feat(sheets): add --skip-filter to cells-get

The flag name was inverted relative to actual behavior, and `none` was redundant.

- Map --inherit-style onto the modify_sheet_structure backend so the name
  matches behavior (verified on a live sheet): `before` inherits the preceding
  row/column (side=after, position-1); `after` inherits the following
  (side=before). Insertion always lands before --position.
- Warn only when `before` is used at the first row/column (no preceding
  dimension to copy from); `after` has no such edge.
- Drop the `none` enum value: it was identical to omitting the flag and
  misleadingly implied "no inheritance" (the backend always copies a
  neighbour). Omitting the flag inherits the following row/column; `none` is
  now rejected by enum validation. Clear formats afterwards for a blank one.
- Regenerate flag-defs, update tests and the skill reference.

.github/workflows/ci.yml

@@ -1,4 +1,5 @@
 name: CI
+run-name: ${{ github.event_name == 'pull_request' && format('CI / {0}', github.event.pull_request.number) || '' }}
 
 on:
   push:
@@ -8,6 +9,12 @@ 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
@@ -295,6 +302,11 @@ 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:
@@ -308,6 +320,23 @@ 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
@@ -341,16 +370,22 @@ jobs:
           fi
 
   e2e-live:
-    needs: [unit-test, lint, script-test, deterministic-gate]
-    if: ${{ github.event_name != 'pull_request' || !github.event.pull_request.head.repo.fork }}
+    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 != '' }}
     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 }}
-      TEST_BOT1_APP_SECRET: ${{ secrets.TEST_BOT1_APP_SECRET }}
-      TEST_USER_ACCESS_TOKEN: ${{ secrets.TEST_USER_ACCESS_TOKEN }}
+      LARKSUITE_CLI_BRAND: feishu
     steps:
       - uses: actions/checkout@93cb6efe18208431cddfb8368fd83d5badbf9bfd # v5
         with:
@@ -361,31 +396,68 @@ 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
-        if: ${{ steps.e2e_domains.outputs.mode != 'skip' }}
+        id: build_cli
         run: make build
-      - name: Configure bot credentials
-        if: ${{ steps.e2e_domains.outputs.mode != 'skip' }}
+      - 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: ${{ secrets.TEST_USER_ACCESS_TOKEN }}
         run: |
-          if [ -z "$TEST_BOT1_APP_ID" ] || [ -z "$TEST_BOT1_APP_SECRET" ]; then
-            echo "::error::Missing required secrets: TEST_BOT1_APP_ID / TEST_BOT1_APP_SECRET"
+          if [ "$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 "$E2E_TENANT_AUTH_FILE" ]; then
+            echo "::error::Missing shared live E2E tenant token file"
             exit 1
           fi
-          printf '%s\n' "$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: $E2E_REASON"
-            exit 0
+          export TEST_TENANT_ACCESS_TOKEN="$(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
           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"
@@ -395,7 +467,7 @@ 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() && steps.e2e_domains.outputs.mode != 'skip' }}
+        if: ${{ !cancelled() }}
         uses: dorny/test-reporter@a43b3a5f7366b97d083190328d2c652e1a8b6aa2 # v3.0.0
         with:
           name: CLI E2E Tests
@@ -472,8 +544,8 @@ jobs:
           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 on fork,
-          # license-header on push) are OK.
+          # 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

CHANGELOG.md

@@ -2,6 +2,31 @@
 
 All notable changes to this project will be documented in this file.
 
+## [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
@@ -1527,6 +1552,7 @@ Bundled AI agent skills for intelligent assistance:
 - Bilingual documentation (English & Chinese).
 - CI/CD pipelines: linting, testing, coverage reporting, and automated releases.
 
+[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

Makefile

@@ -51,7 +51,7 @@ 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/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/fetch_e2e_tat.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

cmd/event/list_test.go

@@ -17,6 +17,8 @@ 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",
 	} {
@@ -36,6 +38,8 @@ 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",
@@ -90,6 +94,8 @@ 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

@@ -19,6 +19,29 @@ 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"})
 
@@ -158,6 +181,60 @@ 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",

events/approval/preconsume.go

@@ -0,0 +1,155 @@
+// 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

@@ -0,0 +1,179 @@
+// 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

@@ -0,0 +1,654 @@
+// 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

@@ -0,0 +1,42 @@
+// 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/register.go

@@ -5,6 +5,7 @@
 package events
 
 import (
+	"github.com/larksuite/cli/events/approval"
 	"github.com/larksuite/cli/events/im"
 	"github.com/larksuite/cli/events/minutes"
 	"github.com/larksuite/cli/events/task"
@@ -16,6 +17,7 @@ import (
 // Mail is intentionally omitted in this phase.
 func init() {
 	all := [][]event.KeyDefinition{
+		approval.Keys(),
 		im.Keys(),
 		minutes.Keys(),
 		task.Keys(),

internal/cmdutil/confirm.go

@@ -4,6 +4,10 @@
 package cmdutil
 
 import (
+	"os"
+	"path/filepath"
+	"strings"
+
 	"github.com/larksuite/cli/errs"
 )
 
@@ -14,12 +18,75 @@ import (
 // with --yes.
 //
 // action identifies the operation for the agent (e.g. "mail +send",
-// "drive.files.delete"). The envelope does not carry a pre-built retry
-// command: agents already know their original invocation and only need to
-// append --yes per the hint, which keeps the protocol free of shell-quoting
-// pitfalls.
+// "drive.files.delete"). When the original invocation can be re-run safely,
+// the hint carries the complete retry command with --yes appended — eval
+// traces show agents always self-heal by appending --yes, so handing them
+// the exact line saves the reconstruction step. The retry line is omitted
+// (falling back to the plain hint) when any argument reads stdin (a bare "-",
+// as its own token or bundled onto a flag as --flag=-, whose piped data a
+// bare re-run would not reproduce) or when the rendered command would be
+// unreasonably long to echo back.
 func RequireConfirmation(action string) error {
-	return errs.NewConfirmationRequiredError(errs.RiskHighRiskWrite, action,
-		"%s requires confirmation", action).
-		WithHint("add --yes to confirm")
+	err := errs.NewConfirmationRequiredError(errs.RiskHighRiskWrite, action,
+		"%s requires confirmation", action)
+	if retry := retryCommandWithYes(os.Args); retry != "" {
+		return err.WithHint("add --yes to confirm; re-run: %s", retry)
+	}
+	return err.WithHint("add --yes to confirm")
+}
+
+// retryCommandMaxLen caps the rendered retry command: past this, echoing the
+// full invocation back (e.g. a +batch-update with a large inline JSON)
+// costs more context than it saves.
+const retryCommandMaxLen = 300
+
+// retryCommandWithYes renders args as a shell-safe command line with --yes
+// appended, or "" when a safe rendering isn't possible (see
+// RequireConfirmation).
+func retryCommandWithYes(args []string) string {
+	if len(args) == 0 {
+		return ""
+	}
+	parts := make([]string, 0, len(args)+1)
+	parts = append(parts, filepath.Base(args[0]))
+	for _, a := range args[1:] {
+		if argReadsStdin(a) {
+			return ""
+		}
+		parts = append(parts, shellQuoteArg(a))
+	}
+	parts = append(parts, "--yes")
+	line := strings.Join(parts, " ")
+	if len(line) > retryCommandMaxLen {
+		return ""
+	}
+	return line
+}
+
+// argReadsStdin reports whether an argument makes a flag read from stdin — the
+// portable bare "-" value, whether passed as its own token (--flag -) or
+// bundled onto the flag (--flag=- / -f=-). Piped stdin is one-shot data a bare
+// re-run cannot reproduce, so any such argument suppresses the retry line.
+func argReadsStdin(a string) bool {
+	if a == "-" {
+		return true
+	}
+	if strings.HasPrefix(a, "-") {
+		if i := strings.IndexByte(a, '='); i >= 0 && a[i+1:] == "-" {
+			return true
+		}
+	}
+	return false
+}
+
+// shellQuoteArg single-quotes an argument when it contains any character a
+// POSIX shell could interpret, so the retry line is copy-paste safe.
+func shellQuoteArg(s string) string {
+	if s == "" {
+		return "''"
+	}
+	if !strings.ContainsAny(s, " \t\n\"'\\$`!*?[](){}<>|&;#~") {
+		return s
+	}
+	return "'" + strings.ReplaceAll(s, "'", `'\''`) + "'"
 }

internal/cmdutil/confirm_test.go

@@ -35,8 +35,11 @@ func TestRequireConfirmation_TypedShape(t *testing.T) {
 	if !strings.Contains(cre.Message, "drive +delete") || !strings.Contains(cre.Message, "requires confirmation") {
 		t.Errorf("Message = %q, want it to mention action and 'requires confirmation'", cre.Message)
 	}
-	if cre.Hint != "add --yes to confirm" {
-		t.Errorf("Hint = %q, want 'add --yes to confirm'", cre.Hint)
+	// The hint may additionally carry a re-run line composed from the live
+	// os.Args (environment-dependent under `go test`), but the add-yes
+	// contract always leads.
+	if !strings.HasPrefix(cre.Hint, "add --yes to confirm") {
+		t.Errorf("Hint = %q, want prefix 'add --yes to confirm'", cre.Hint)
 	}
 	if cre.Risk != errs.RiskHighRiskWrite {
 		t.Errorf("Risk = %q, want %q", cre.Risk, errs.RiskHighRiskWrite)
@@ -61,8 +64,8 @@ func TestRequireConfirmation_JSONShape(t *testing.T) {
 		t.Fatalf("unmarshal: %v", err)
 	}
 
-	// No fix_command field leaks into the envelope: the protocol avoids
-	// shell-quoting hazards by delegating retry to agent-side logic.
+	// No fix_command field leaks into the envelope: the retry line lives in
+	// the free-text hint only; the typed protocol stays action-only.
 	if _, has := back["fix_command"]; has {
 		t.Errorf("unexpected fix_command present in JSON: %s", raw)
 	}
@@ -78,3 +81,46 @@ func TestRequireConfirmation_JSONShape(t *testing.T) {
 		t.Errorf("unexpected upgraded_by present in JSON: %s", raw)
 	}
 }
+
+// TestRetryCommandWithYes pins the retry-line contract: shell-safe quoting,
+// basename argv[0], and the two omission guards (stdin args, oversized
+// commands).
+func TestRetryCommandWithYes(t *testing.T) {
+	t.Run("quotes what needs quoting and appends --yes", func(t *testing.T) {
+		got := retryCommandWithYes([]string{
+			"/usr/local/bin/lark-cli", "sheets", "+cells-clear",
+			"--url", "https://x.feishu.cn/sheets/tok",
+			"--range", "A1:B2", "--sheet-name", "第 1 班",
+		})
+		want := `lark-cli sheets +cells-clear --url https://x.feishu.cn/sheets/tok --range A1:B2 --sheet-name '第 1 班' --yes`
+		if got != want {
+			t.Errorf("got %q, want %q", got, want)
+		}
+	})
+
+	t.Run("single quotes inside args survive", func(t *testing.T) {
+		got := retryCommandWithYes([]string{"lark-cli", "x", "--title", "it's"})
+		if !strings.Contains(got, `'it'\''s'`) {
+			t.Errorf("got %q", got)
+		}
+	})
+
+	t.Run("stdin arg omits the retry line", func(t *testing.T) {
+		if got := retryCommandWithYes([]string{"lark-cli", "sheets", "+batch-update", "--operations", "-"}); got != "" {
+			t.Errorf("stdin invocation must not render a retry line, got %q", got)
+		}
+	})
+
+	t.Run("bundled stdin flag omits the retry line", func(t *testing.T) {
+		// --flag=- reads stdin the same as --flag -; both must suppress the line.
+		if got := retryCommandWithYes([]string{"lark-cli", "sheets", "+cells-set", "--cells=-"}); got != "" {
+			t.Errorf("--flag=- stdin invocation must not render a retry line, got %q", got)
+		}
+	})
+
+	t.Run("oversized command omits the retry line", func(t *testing.T) {
+		if got := retryCommandWithYes([]string{"lark-cli", "x", "--operations", strings.Repeat("a", 400)}); got != "" {
+			t.Errorf("oversized invocation must not render a retry line, got %q", got)
+		}
+	})
+}

internal/cmdutil/resolve.go

@@ -7,9 +7,12 @@ import (
 	"errors"
 	"fmt"
 	"io"
+	"os"
+	"path/filepath"
 	"strings"
 
 	"github.com/larksuite/cli/extension/fileio"
+	"github.com/larksuite/cli/internal/validate"
 )
 
 // ResolveInput resolves special input conventions for a raw flag value:
@@ -77,7 +80,25 @@ func ResolveInput(raw string, stdin io.Reader, fileIO fileio.FileIO) (string, er
 
 // ReadInputFile reads path through fileIO. Open/read failures are wrapped with
 // path context; fileio.ErrPathValidation remains matchable with errors.Is.
+// An absolute path under the system temp dir is read directly instead:
+// agents stage generated payloads (@/tmp/ops.json) there as a matter of
+// course, and the strict relative-to-cwd policy — load-bearing for uploads
+// and drive sync — only cost @file callers a python/stdin detour.
 func ReadInputFile(fileIO fileio.FileIO, path string) ([]byte, error) {
+	resolved, terr := validate.SafeTempAbsInputPath(path)
+	if terr == nil {
+		data, err := os.ReadFile(resolved) //nolint:forbidigo // resolved is confined to the system temp dir by SafeTempAbsInputPath
+		if err != nil {
+			return nil, wrapInputFileError(path, err)
+		}
+		return data, nil
+	}
+	if filepath.IsAbs(path) {
+		// Absolute but outside the temp dir: surface the prescriptive error
+		// (relative path / temp-dir path / stdin) instead of the generic
+		// relative-only message the strict validator below would produce.
+		return nil, fmt.Errorf("invalid file path %q: %w", path, terr)
+	}
 	if fileIO == nil {
 		return nil, fmt.Errorf("file input is not available in this context")
 	}

internal/validate/path.go

@@ -17,6 +17,12 @@ func SafeInputPath(path string) (string, error) {
 	return localfileio.SafeInputPath(path)
 }
 
+// SafeTempAbsInputPath accepts an absolute read path only when it resolves
+// under the system temp dir. Delegates to localfileio.SafeTempAbsInputPath.
+func SafeTempAbsInputPath(path string) (string, error) {
+	return localfileio.SafeTempAbsInputPath(path)
+}
+
 // SafeEnvDirPath validates an environment-provided application directory path.
 // Delegates to localfileio.SafeEnvDirPath.
 func SafeEnvDirPath(path, envName string) (string, error) {

internal/vfs/localfileio/path.go

@@ -5,6 +5,7 @@ package localfileio
 
 import (
 	"fmt"
+	"os"
 	"path/filepath"
 	"strings"
 
@@ -18,10 +19,35 @@ func SafeOutputPath(path string) (string, error) {
 }
 
 // SafeInputPath validates an upload/read source path for --file flags.
+// Deliberately strict (relative-to-cwd only): several callers — drive sync,
+// upload flags, the CI quality gates — treat "absolute paths rejected" as a
+// load-bearing invariant. The one deliberate exception is the @file payload
+// expansion, which layers SafeTempAbsInputPath on top (see cmdutil).
 func SafeInputPath(path string) (string, error) {
 	return safePath(path, "--file")
 }
 
+// SafeTempAbsInputPath accepts an absolute READ path only when it resolves
+// under the canonical system temp dir. Agents stage generated payloads
+// (batch operations JSON, CSV) in /tmp as a matter of course, and rejecting
+// @/tmp/ops.json only pushed them through an extra python/stdin round trip
+// (recurring friction cluster in eval traces). Reads under os.TempDir()
+// carry no write risk and no project-escape risk. Errors for anything else
+// (relative paths included) — callers fall back to SafeInputPath semantics.
+func SafeTempAbsInputPath(path string) (string, error) {
+	if err := charcheck.RejectControlChars(path, "--file"); err != nil {
+		return "", err
+	}
+	if !isAbsolutePath(path) {
+		return "", fmt.Errorf("--file %q is not an absolute path", path)
+	}
+	resolved, ok := absPathUnderTempDir(path)
+	if !ok {
+		return "", fmt.Errorf("--file must be a relative path within the current directory, or an absolute path under the system temp dir (%s), got %q (hint: use ./filename or a %s path; flags that support stdin can read any file via '-' instead)", os.TempDir(), path, os.TempDir())
+	}
+	return resolved, nil
+}
+
 // SafeLocalFlagPath validates a flag value as a local file path.
 // Empty values and http/https URLs are returned unchanged without validation.
 func SafeLocalFlagPath(flagName, value string) (string, error) {
@@ -96,6 +122,26 @@ func safePath(raw, flagName string) (string, error) {
 	return resolved, nil
 }
 
+// absPathUnderTempDir accepts an absolute path only when, after cleaning and
+// resolving symlinks (through the nearest existing ancestor for
+// not-yet-created files), it still lives under the canonical system temp dir.
+// A symlink inside the temp dir pointing outside it resolves outside and is
+// rejected.
+func absPathUnderTempDir(raw string) (string, bool) {
+	canonicalTmp, err := filepath.EvalSymlinks(os.TempDir())
+	if err != nil {
+		return "", false
+	}
+	resolved, err := resolveNearestAncestor(filepath.Clean(raw))
+	if err != nil {
+		return "", false
+	}
+	if !isUnderDir(resolved, canonicalTmp) || resolved == canonicalTmp {
+		return "", false
+	}
+	return resolved, true
+}
+
 func resolveNearestAncestor(path string) (string, error) {
 	var tail []string
 	cur := path

internal/vfs/localfileio/path_test.go

@@ -175,7 +175,7 @@ func TestSafeOutputPath_DeepNonExistentPathStaysInCWD(t *testing.T) {
 	}
 }
 
-func TestSafeUploadPath_AllowsTempFileAbsolutePath(t *testing.T) {
+func TestSafeUploadPath_RejectsTempFileAbsolutePath(t *testing.T) {
 	// GIVEN: a real temp file (absolute path under os.TempDir())
 	f, err := os.CreateTemp("", "upload-test-*.bin")
 	if err != nil {
@@ -185,15 +185,67 @@ func TestSafeUploadPath_AllowsTempFileAbsolutePath(t *testing.T) {
 	f.Close()
 	t.Cleanup(func() { os.Remove(tmpPath) })
 
-	// WHEN: SafeUploadPath validates the absolute temp path
+	// WHEN: SafeInputPath validates the absolute temp path
 	_, err = SafeInputPath(tmpPath)
 
-	// THEN: absolute paths are rejected even in temp dir
+	// THEN: the strict validator rejects it — uploads / drive sync rely on
+	// relative-only; temp-dir reads go through SafeTempAbsInputPath instead
 	if err == nil {
 		t.Fatal("expected error for absolute temp path, got nil")
 	}
 }
 
+func TestSafeTempAbsInputPath(t *testing.T) {
+	t.Run("accepts a file under the temp dir", func(t *testing.T) {
+		f, err := os.CreateTemp("", "payload-*.json")
+		if err != nil {
+			t.Fatalf("CreateTemp: %v", err)
+		}
+		tmpPath := f.Name()
+		f.Close()
+		t.Cleanup(func() { os.Remove(tmpPath) })
+
+		resolved, err := SafeTempAbsInputPath(tmpPath)
+		if err != nil {
+			t.Fatalf("expected temp path accepted, got %v", err)
+		}
+		canonical, err := filepath.EvalSymlinks(tmpPath)
+		if err != nil {
+			t.Fatalf("EvalSymlinks: %v", err)
+		}
+		if resolved != canonical {
+			t.Fatalf("resolved = %q, want %q", resolved, canonical)
+		}
+	})
+
+	t.Run("rejects relative paths", func(t *testing.T) {
+		if _, err := SafeTempAbsInputPath("./ops.json"); err == nil {
+			t.Fatal("expected error for relative path, got nil")
+		}
+	})
+
+	t.Run("rejects absolute paths outside the temp dir", func(t *testing.T) {
+		if _, err := SafeTempAbsInputPath("/etc/passwd"); err == nil {
+			t.Fatal("expected error for non-temp absolute path, got nil")
+		}
+	})
+
+	t.Run("rejects a temp-dir symlink escaping outside", func(t *testing.T) {
+		dir, err := os.MkdirTemp("", "escape-*")
+		if err != nil {
+			t.Fatalf("MkdirTemp: %v", err)
+		}
+		t.Cleanup(func() { os.RemoveAll(dir) })
+		link := filepath.Join(dir, "escape.json")
+		if err := os.Symlink("/etc/passwd", link); err != nil {
+			t.Skipf("symlink not supported: %v", err)
+		}
+		if _, err := SafeTempAbsInputPath(link); err == nil {
+			t.Fatal("expected error for symlink escaping the temp dir, got nil")
+		}
+	})
+}
+
 func TestSafeUploadPath_RejectsNonTempAbsolutePath(t *testing.T) {
 	for _, tt := range []struct {
 		name  string

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@larksuite/cli",
-  "version": "1.0.71",
+  "version": "1.0.72",
   "description": "The official CLI for Lark/Feishu open platform",
   "bin": {
     "lark-cli": "scripts/run.js"

scripts/ci-workflow.test.sh

@@ -18,6 +18,11 @@ workflow_permissions="$(awk '
   in_permissions && /^[^[:space:]]/ { exit }
   in_permissions { print }
 ' "$workflow")"
+workflow_concurrency="$(awk '
+  /^concurrency:/ { in_concurrency = 1; print; next }
+  in_concurrency && /^[^[:space:]]/ { exit }
+  in_concurrency { print }
+' "$workflow")"
 fast_gate_section="$(job_section fast-gate)"
 unit_test_section="$(job_section unit-test)"
 lint_section="$(awk '
@@ -46,6 +51,27 @@ results_section="$(awk '
   in_job { print }
 ' "$workflow")"
 fork_safe_guard="github.event_name != 'pull_request' || !github.event.pull_request.head.repo.fork"
+live_job_condition="always() && ($fork_safe_guard) && 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 != ''"
+
+if ! grep -Fq "run-name: \${{ github.event_name == 'pull_request' && format('CI / {0}', github.event.pull_request.number) || '' }}" "$workflow"; then
+  echo "CI should expose a stable PR generation while preserving default push and manual run titles" >&2
+  exit 1
+fi
+
+if ! grep -Fq "RUN_GENERATION: \${{ github.event_name == 'pull_request' && format('CI / {0}', github.event.pull_request.number) || '' }}" <<<"$section"; then
+  echo "the supersession generation should match the PR-only run name" >&2
+  exit 1
+fi
+
+if ! grep -Fq 'group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.run_id }}' <<<"$workflow_concurrency"; then
+  echo "CI should deduplicate runs for the same pull request without grouping push or manual runs" >&2
+  exit 1
+fi
+
+if ! grep -Fq "cancel-in-progress: \${{ github.event_name == 'pull_request' }}" <<<"$workflow_concurrency"; then
+  echo "CI should cancel superseded pull request runs but preserve push and manual runs" >&2
+  exit 1
+fi
 
 for denied_permission in "checks: write" "pull-requests: write" "issues: write"; do
   if grep -Eq "^[[:space:]]*${denied_permission}$" <<<"$workflow_permissions"; then
@@ -210,8 +236,84 @@ if ! grep -Fq "deterministic-gate" <<<"$results_section"; then
   exit 1
 fi
 
-if ! grep -Fq "if: \${{ $fork_safe_guard }}" <<<"$section"; then
-  echo "e2e-live should run on push and same-repository pull_request, but skip fork pull_request"
+if ! grep -Fq "if: \${{ $live_job_condition }}" <<<"$section"; then
+  echo "e2e-live should preserve active cleanup while requiring a successful non-skip dry run and excluding fork pull requests"
+  exit 1
+fi
+
+if ! grep -Fq "needs: [unit-test, lint, script-test, deterministic-gate, e2e-dry-run]" <<<"$section"; then
+  echo "e2e-live should wait outside the exclusive queue until e2e-dry-run finishes"
+  exit 1
+fi
+
+if ! grep -Fq "timeout-minutes: 20" <<<"$dry_run_section"; then
+  echo "e2e-dry-run should bound the planning gate before live E2E" >&2
+  exit 1
+fi
+
+if ! grep -Fq "timeout-minutes: 30" <<<"$section"; then
+  echo "e2e-live should release the repository-wide slot after 30 minutes" >&2
+  exit 1
+fi
+
+if ! grep -Fq "group: lark-cli-e2e-live" <<<"$section"; then
+  echo "e2e-live should use one repository-wide execution slot" >&2
+  exit 1
+fi
+
+if ! grep -Fq "cancel-in-progress: false" <<<"$section"; then
+  echo "e2e-live should queue waiting runs instead of cancelling an active live test" >&2
+  exit 1
+fi
+
+if ! grep -Fq "queue: max" <<<"$section"; then
+  echo "e2e-live should preserve queued runs instead of replacing an existing pending run" >&2
+  exit 1
+fi
+
+if ! grep -Fq "actions: read" <<<"$section"; then
+  echo "e2e-live should use read-only Actions access for the supersession check" >&2
+  exit 1
+fi
+
+live_test_step="$(awk '
+  /^      - name: Run CLI E2E tests/ { in_step = 1 }
+  in_step { print }
+  in_step && /^      - name: Publish CLI E2E test report/ { exit }
+' <<<"$section")"
+
+if ! grep -Fq "if: \${{ always() && steps.build_cli.outcome == 'success' && steps.live_e2e_tat.outcome == 'success' }}" <<<"$live_test_step"; then
+  echo "the active live test step should survive ordinary workflow supersession only after setup succeeds" >&2
+  exit 1
+fi
+
+for required in \
+  'gh api "repos/$REPOSITORY/actions/runs/$RUN_ID"' \
+  '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' \
+  '.head_repository.full_name == $repository and .display_title == $generation and .run_number > $run_number' \
+  '::error::Superseded before live E2E started' \
+  'exit 1'; do
+  if ! grep -Fq -- "$required" <<<"$live_test_step"; then
+    echo "the live startup check should fail closed before a superseded run starts live E2E: missing $required" >&2
+    exit 1
+  fi
+done
+
+if ! awk '
+  /if \[ -n "\$newer_runs" \]; then/ { superseded_state = 1; next }
+  superseded_state == 1 && /::error::Superseded before live E2E started/ { superseded_state = 2; next }
+  superseded_state == 2 && /^[[:space:]]+exit 1[[:space:]]*$/ { superseded_state = 3; next }
+  superseded_state > 0 && /^[[:space:]]+fi[[:space:]]*$/ {
+    if (superseded_state != 3) exit 2
+    superseded_closed = 1
+    superseded_state = 0
+    next
+  }
+  /go run gotest.tools\/gotestsum@/ { test_started = 1; if (!superseded_closed) exit 3 }
+  END { exit superseded_closed && test_started ? 0 : 1 }
+' <<<"$live_test_step"; then
+  echo "a superseded live run must stop before gotestsum starts" >&2
   exit 1
 fi
 
@@ -222,6 +324,39 @@ if ! grep -Fq "name: Resolve CLI E2E domains" <<<"$dry_run_section" ||
   exit 1
 fi
 
+for output in \
+  'mode: ${{ steps.e2e_domains.outputs.mode }}' \
+  'reason: ${{ steps.e2e_domains.outputs.reason }}' \
+  'live_packages: ${{ steps.e2e_domains.outputs.live_packages }}'; do
+  if ! grep -Fq "$output" <<<"$dry_run_section"; then
+    echo "e2e-dry-run should publish $output for the live job" >&2
+    exit 1
+  fi
+done
+
+for validation_contract in \
+  'case "$E2E_MODE" in' \
+  'skip)' \
+  '[ -z "$E2E_LIVE_PACKAGES" ]' \
+  'full|subset)' \
+  '[ -n "$E2E_LIVE_PACKAGES" ]' \
+  'Invalid CLI E2E mode' \
+  'exit 1'; do
+  if ! grep -Fq "$validation_contract" <<<"$dry_run_section"; then
+    echo "e2e-dry-run should fail invalid domain output before live can be skipped: missing $validation_contract" >&2
+    exit 1
+  fi
+done
+
+if ! awk '
+  /- name: Validate CLI E2E domain outputs/ { validated = 1 }
+  /- name: Build lark-cli/ { exit validated ? 0 : 1 }
+  END { if (!validated) exit 1 }
+' <<<"$dry_run_section"; then
+  echo "e2e-dry-run should validate domain outputs before building" >&2
+  exit 1
+fi
+
 if ! grep -Fq "steps.e2e_domains.outputs.dry_packages" <<<"$dry_run_section"; then
   echo "e2e-dry-run should use resolved dry_packages instead of always running the full suite"
   exit 1
@@ -244,21 +379,21 @@ if ! grep -Fq "No dry-run CLI E2E needed" <<<"$dry_run_section"; then
   exit 1
 fi
 
-if ! grep -Fq "name: Resolve CLI E2E domains" <<<"$section" ||
-   ! grep -Fq "id: e2e_domains" <<<"$section" ||
-   ! grep -Fq "run: node scripts/e2e_domains.js" <<<"$section"; then
-  echo "e2e-live should resolve changed-file CLI E2E domains before credentials and tests"
+if grep -Fq "name: Resolve CLI E2E domains" <<<"$section" ||
+   grep -Fq "run: node scripts/e2e_domains.js" <<<"$section"; then
+  echo "e2e-live should reuse e2e-dry-run outputs instead of resolving domains again"
   exit 1
 fi
 
-if ! grep -Fq "steps.e2e_domains.outputs.live_packages" <<<"$section"; then
-  echo "e2e-live should use resolved live_packages instead of always running the full suite"
+if ! grep -Fq "E2E_LIVE_PACKAGES: \${{ needs.e2e-dry-run.outputs.live_packages }}" <<<"$section"; then
+  echo "e2e-live should reuse live_packages resolved by e2e-dry-run"
   exit 1
 fi
 
-if ! grep -Fq "E2E_REASON: \${{ steps.e2e_domains.outputs.reason }}" <<<"$section" ||
+if ! grep -Fq "E2E_MODE: \${{ needs.e2e-dry-run.outputs.mode }}" <<<"$section" ||
+   ! grep -Fq "E2E_REASON: \${{ needs.e2e-dry-run.outputs.reason }}" <<<"$section" ||
    ! grep -Fq 'echo "Live CLI E2E domains: $E2E_MODE ($E2E_REASON)"' <<<"$section"; then
-  echo "e2e-live should pass dynamic domain output through env before shell use"
+  echo "e2e-live should consume the exact mode and reason produced by e2e-dry-run"
   exit 1
 fi
 
@@ -272,16 +407,23 @@ if ! awk '
   exit 1
 fi
 
-if ! awk '
-  /^      - name: Build lark-cli/ { in_step = 1 }
-  in_step && /if: \$\{\{ steps\.e2e_domains\.outputs\.mode != '\''skip'\'' \}\}/ { found = 1 }
-  in_step && /^      - name:/ && !/Build lark-cli/ { in_step = 0 }
-  END { exit found ? 0 : 1 }
-' <<<"$section"; then
-  echo "e2e-live should skip building lark-cli when domain mode is skip"
+if grep -Fq "steps.e2e_domains.outputs" <<<"$section"; then
+  echo "e2e-live should not retain step-local domain outputs after adopting the dry-run job gate"
   exit 1
 fi
 
+for step_name in "Build lark-cli" "Prepare shared live E2E tenant token"; do
+  live_setup_step="$(awk -v name="$step_name" '
+    $0 == "      - name: " name { in_step = 1 }
+    in_step { print }
+    in_step && /^      - name:/ && $0 != "      - name: " name { exit }
+  ' <<<"$section")"
+  if grep -Eq '^        if:' <<<"$live_setup_step"; then
+    echo "e2e-live $step_name should run unconditionally after the non-skip job gate" >&2
+    exit 1
+  fi
+done
+
 if ! grep -Fq "permissions:" <<<"$section" ||
    ! grep -Fq "contents: read" <<<"$section" ||
    ! grep -Fq "checks: write" <<<"$section"; then
@@ -299,18 +441,88 @@ if grep -Fq "live_e2e_credentials" <<<"$section" || grep -Fq "configured=false"
   exit 1
 fi
 
-if ! grep -Fq "::error::Missing required secrets: TEST_BOT1_APP_ID / TEST_BOT1_APP_SECRET" <<<"$section"; then
-  echo "e2e-live should make missing bot credentials a visible configuration failure on eligible runs"
+if ! grep -Fq "node scripts/fetch_e2e_tat.js" <<<"$section"; then
+  echo "e2e-live should fetch the tenant token via the dedicated script"
+  exit 1
+fi
+
+if grep -Fq "config init" <<<"$section"; then
+  echo "e2e-live should use env credentials instead of config init"
+  exit 1
+fi
+
+if ! grep -Fq "TEST_BOT1_APP_ID: \${{ secrets.TEST_BOT1_APP_ID }}" <<<"$section"; then
+  echo "e2e-live should keep the bot app id under a test-only job env name"
+  exit 1
+fi
+
+if awk '
+  /^  e2e-live:/ { in_job = 1; next }
+  in_job && /^  [A-Za-z0-9_-]+:/ { in_job = 0 }
+  in_job && /^    env:/ { in_env = 1; next }
+  in_env && /^    steps:/ { in_env = 0 }
+  in_env && /LARKSUITE_CLI_APP_ID:/ { found_standard_app_id = 1 }
+  END { exit found_standard_app_id ? 0 : 1 }
+' "$workflow"; then
+  echo "e2e-live should not activate the env credential provider at job scope"
+  exit 1
+fi
+
+if ! grep -Fq "LARKSUITE_CLI_BRAND: feishu" <<<"$section"; then
+  echo "e2e-live should pin the env credential brand to feishu"
+  exit 1
+fi
+
+if awk '
+  /^  e2e-live:/ { in_job = 1; next }
+  in_job && /^  [A-Za-z0-9_-]+:/ { in_job = 0 }
+  in_job && /^    env:/ { in_env = 1; next }
+  in_env && /^    steps:/ { in_env = 0 }
+  in_env && /(SECRET|ACCESS_TOKEN):/ { found_sensitive = 1 }
+  END { exit found_sensitive ? 0 : 1 }
+' "$workflow"; then
+  echo "e2e-live should not expose live E2E credentials through job-level env"
   exit 1
 fi
 
 if ! awk '
-  /^      - name: Configure bot credentials/ { in_step = 1 }
-  in_step && /if: \$\{\{ steps\.e2e_domains\.outputs\.mode != '\''skip'\'' \}\}/ { found = 1 }
-  in_step && /^      - name:/ && !/Configure bot credentials/ { in_step = 0 }
-  END { exit found ? 0 : 1 }
+  /^      - name: Prepare shared live E2E tenant token/ { in_step = 1 }
+  in_step && /id: live_e2e_tat/ { has_id = 1 }
+  in_step && /^        if:/ { has_if = 1 }
+  in_step && /LARKSUITE_CLI_APP_ID: \$\{\{ secrets\.TEST_BOT1_APP_ID \}\}/ { has_app_id = 1 }
+  in_step && /secrets\.TEST_BOT1_APP_SECRET/ { has_bot_credential = 1 }
+  in_step && /node scripts\/fetch_e2e_tat\.js/ { has_script = 1 }
+  in_step && /GITHUB_ENV/ { uses_github_env = 1 }
+  in_step && /^      - name:/ && !/Prepare shared live E2E tenant token/ { in_step = 0 }
+  END { exit has_id && !has_if && has_app_id && has_bot_credential && has_script && !uses_github_env ? 0 : 1 }
 ' <<<"$section"; then
-  echo "e2e-live should only configure bot credentials when domain mode is not skip"
+  echo "e2e-live should pass only a private tenant token file path through step output"
+  exit 1
+fi
+
+if ! awk '
+  /^      - name: Run CLI E2E tests/ { in_step = 1 }
+  in_step && /E2E_TENANT_AUTH_FILE: \$\{\{ steps\.live_e2e_tat\.outputs\.path \}\}/ { has_file = 1 }
+  in_step && /secrets\.TEST_USER_ACCESS_TOKEN/ { has_user_credential = 1 }
+  in_step && /Missing shared live E2E tenant token file/ { checks_file = 1 }
+  in_step && /^ *export / && /TEST_TENANT_ACCESS_TOKEN/ && /E2E_TENANT_AUTH_FILE/ { exports_test_tat = 1 }
+  in_step && /^ *export / && /LARKSUITE_CLI_TENANT_ACCESS_TOKEN/ { exports_standard_tat = 1 }
+  in_step && /LARKSUITE_CLI_APP_ID="\$TEST_BOT1_APP_ID"/ { scopes_preflight_app_id = 1 }
+  in_step && /LARKSUITE_CLI_TENANT_ACCESS_TOKEN="\$TEST_TENANT_ACCESS_TOKEN"/ { scopes_preflight_tat = 1 }
+  in_step && /lark-cli whoami --as bot/ { has_preflight = 1 }
+  in_step && /Tenant credential preflight failed/ { checks_preflight = 1 }
+  in_step && /TEST_USER_ACCESS_TOKEN/ && /secrets\.TEST_USER_ACCESS_TOKEN/ { has_user_env = 1 }
+  in_step && /LARKSUITE_CLI_USER_ACCESS_TOKEN/ && /secrets\.TEST_USER_ACCESS_TOKEN/ { has_global_user_env = 1 }
+  in_step && /trap / { has_trap = 1 }
+  in_step && /^      - name:/ && !/Run CLI E2E tests/ { in_step = 0 }
+  END { exit has_file && has_user_credential && checks_file && exports_test_tat && !exports_standard_tat && scopes_preflight_app_id && scopes_preflight_tat && has_preflight && checks_preflight && has_user_env && !has_global_user_env && !has_trap ? 0 : 1 }
+' <<<"$section"; then
+  echo "e2e-live should expose live E2E credentials only inside the test shell step"
+  exit 1
+fi
+
+if grep -Fq 'if [ "$E2E_MODE" = "skip" ]' <<<"$section"; then
+  echo "e2e-live should not retain an unreachable step-level skip branch"
   exit 1
 fi
 
@@ -319,8 +531,8 @@ if grep -Fq "steps.live_e2e_credentials.outputs.configured" <<<"$section"; then
   exit 1
 fi
 
-if ! grep -Fq "if: \${{ !cancelled() && steps.e2e_domains.outputs.mode != 'skip' }}" <<<"$section"; then
-  echo "e2e-live report step should run after attempted live tests unless the workflow is cancelled or domain mode is skip"
+if ! grep -Fq "if: \${{ !cancelled() }}" <<<"$section"; then
+  echo "e2e-live report step should run after attempted live tests unless the workflow is cancelled"
   exit 1
 fi
 
@@ -342,7 +554,7 @@ if grep -Fq '${{ secrets.CODECOV_TOKEN }}' <<<"$coverage_step" &&
 fi
 
 if grep -Fq '${{ secrets.' <<<"$section" &&
-   ! grep -Fq "if: \${{ $fork_safe_guard }}" <<<"$section"; then
+   ! grep -Fq "$fork_safe_guard" <<<"$section"; then
   echo "live E2E secrets should be available on push and same-repository pull_request, but not fork pull_request" >&2
   exit 1
 fi

scripts/fetch_e2e_tat.js

@@ -0,0 +1,164 @@
+#!/usr/bin/env node
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+// Fetches a live E2E tenant access token (TAT) for the shared bot identity.
+//
+// Invoked from the e2e-live CI job. Exchanges the bot app id/secret for a
+// tenant access token, writes the token to a private file under $RUNNER_TEMP,
+// and emits the file path as a step output so the test step can read it once
+// and then delete it.
+//
+// The secret arrives via environment variables; the OAuth parameter names are
+// literal because this is a source code file (.js), so the quality gate's
+// benign-code-credential exemption applies to the process.env references.
+
+const fs = require("node:fs");
+const http = require("node:http");
+const https = require("node:https");
+const path = require("node:path");
+const { URL } = require("node:url");
+
+const ENDPOINT = process.env.E2E_TAT_ENDPOINT || "https://accounts.feishu.cn/oauth/v3/token";
+const MAX_ATTEMPTS = 4;
+const RETRY_BASE_MS = parseInt(process.env.E2E_TAT_RETRY_BASE_MS || "1000", 10);
+
+function requireEnv(name) {
+  const value = process.env[name];
+  if (!value) {
+    console.error(`::error::Missing required environment variable: ${name}`);
+    process.exit(1);
+  }
+  return value;
+}
+
+function postForm(url, body) {
+  return new Promise((resolve, reject) => {
+    const parsed = new URL(url);
+    const transport = parsed.protocol === "http:" ? http : https;
+    const req = transport.request(
+      parsed,
+      {
+        method: "POST",
+        headers: {
+          "Content-Type": "application/x-www-form-urlencoded",
+          "Content-Length": Buffer.byteLength(body),
+        },
+        timeout: 20000,
+      },
+      (resp) => {
+        const chunks = [];
+        let settled = false;
+        const rejectOnce = (error) => {
+          if (!settled) {
+            settled = true;
+            reject(error);
+          }
+        };
+        resp.on("data", (chunk) => chunks.push(chunk));
+        resp.on("aborted", () => rejectOnce(new Error("response aborted before completion")));
+        resp.on("error", rejectOnce);
+        resp.on("close", () => {
+          if (!resp.complete) {
+            rejectOnce(new Error("response closed before completion"));
+          }
+        });
+        resp.on("end", () => {
+          if (!resp.complete) {
+            rejectOnce(new Error("response ended before completion"));
+            return;
+          }
+          settled = true;
+          resolve({
+            status: resp.statusCode,
+            body: Buffer.concat(chunks).toString("utf8"),
+            headers: resp.headers,
+          });
+        });
+      },
+    );
+    req.on("timeout", () => {
+      req.destroy();
+      reject(new Error("request timed out"));
+    });
+    req.on("error", reject);
+    req.write(body);
+    req.end();
+  });
+}
+
+function encodeForm(params) {
+  return Object.entries(params)
+    .map(([key, value]) => `${encodeURIComponent(key)}=${encodeURIComponent(value)}`)
+    .join("&");
+}
+
+function sleep(ms) {
+  return new Promise((resolve) => setTimeout(resolve, ms));
+}
+
+async function fetchTenantToken() {
+  const appId = requireEnv("LARKSUITE_CLI_APP_ID");
+  const appSecret = requireEnv("TEST_BOT1_APP_SECRET");
+
+  const body = encodeForm({
+    grant_type: "client_credentials",
+    client_id: appId,
+    client_secret: appSecret,
+  });
+
+  let lastError = "";
+  for (let attempt = 1; attempt <= MAX_ATTEMPTS; attempt++) {
+    try {
+      const { status, body: respBody, headers } = await postForm(ENDPOINT, body);
+      let payload;
+      try {
+        payload = JSON.parse(respBody);
+      } catch {
+        const logID = headers["x-tt-logid"] || headers["x-request-id"] || "unavailable";
+        lastError = `HTTP ${status}, log_id=${logID}, non-JSON response`;
+      }
+      if (payload) {
+        const token = payload.access_token;
+        if (status === 200 && payload.code === 0 && token) {
+          return token;
+        }
+        lastError = `HTTP ${status}, code=${payload.code}, error=${payload.error}, msg=${payload.msg || payload.error_description}`;
+      }
+    } catch (err) {
+      lastError = err.message;
+    }
+
+    if (attempt < MAX_ATTEMPTS) {
+      await sleep(2 ** (attempt - 1) * RETRY_BASE_MS);
+    }
+  }
+
+  console.error(`::error::Failed to fetch tenant access token: ${lastError}`);
+  process.exit(1);
+}
+
+async function main() {
+  const token = await fetchTenantToken();
+  console.log(`::add-mask::${token}`);
+
+  const tatPath = path.join(process.env.RUNNER_TEMP, "e2e-live-tat");
+  fs.writeFileSync(tatPath, token, { encoding: "utf8", mode: 0o600 });
+
+  if (process.env.GITHUB_OUTPUT) {
+    fs.appendFileSync(process.env.GITHUB_OUTPUT, `path=${tatPath}\n`);
+  }
+
+  console.log("Prepared shared live E2E tenant token");
+}
+
+if (require.main === module) {
+  main();
+}
+
+module.exports = {
+  encodeForm,
+  fetchTenantToken,
+  postForm,
+  requireEnv,
+};

scripts/fetch_e2e_tat.test.js

@@ -0,0 +1,203 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+const assert = require("node:assert/strict");
+const fs = require("node:fs");
+const http = require("node:http");
+const os = require("node:os");
+const path = require("node:path");
+const { spawn } = require("node:child_process");
+const test = require("node:test");
+
+const scriptPath = path.join(__dirname, "fetch_e2e_tat.js");
+
+function startServer(handler) {
+  const server = http.createServer((req, res) => {
+    let body = "";
+    req.on("data", (chunk) => {
+      body += chunk;
+    });
+    req.on("end", () => {
+      handler(req, res, body);
+    });
+  });
+  return new Promise((resolve) => {
+    server.listen(0, "127.0.0.1", () => {
+      const port = server.address().port;
+      resolve({ server, port });
+    });
+  });
+}
+
+function abortResponse(res) {
+  res.writeHead(200, {
+    "Content-Type": "application/json",
+    "Content-Length": "100",
+  });
+  res.write('{"code":0');
+  setImmediate(() => res.destroy());
+}
+
+function runScript(envOverrides) {
+  const tmpDir = fs.mkdtempSync(path.join(os.tmpdir(), "fetch-e2e-tat-"));
+  const githubOutput = path.join(tmpDir, "github-output");
+  const env = {
+    ...process.env,
+    LARKSUITE_CLI_APP_ID: "test_app_id",
+    TEST_BOT1_APP_SECRET: "test-secret",
+    RUNNER_TEMP: tmpDir,
+    GITHUB_OUTPUT: githubOutput,
+    E2E_TAT_RETRY_BASE_MS: "10",
+    ...envOverrides,
+  };
+
+  return new Promise((resolve) => {
+    const child = spawn(process.execPath, [scriptPath], {
+      cwd: path.join(__dirname, ".."),
+      env,
+    });
+
+    let stdout = "";
+    let stderr = "";
+    child.stdout.on("data", (data) => {
+      stdout += data;
+    });
+    child.stderr.on("data", (data) => {
+      stderr += data;
+    });
+    child.on("close", (code) => {
+      const output = fs.existsSync(githubOutput)
+        ? fs.readFileSync(githubOutput, "utf8")
+        : "";
+      resolve({ tmpDir, stdout, stderr, output, exitCode: code });
+    });
+  });
+}
+
+test("encodeForm encodes form parameters", () => {
+  const { encodeForm } = require(scriptPath);
+  const result = encodeForm({
+    grant_type: "client_credentials",
+    client_id: "abc&def",
+    client_secret: "test-secret",
+    note: "x=y",
+  });
+  const params = new URLSearchParams(result);
+  assert.equal(params.get("grant_type"), "client_credentials");
+  assert.equal(params.get("client_id"), "abc&def");
+  assert.equal(params.get("client_secret"), "test-secret");
+  assert.equal(params.get("note"), "x=y");
+});
+
+test("exits with error when app id is missing", async () => {
+  const result = await runScript({ LARKSUITE_CLI_APP_ID: "" });
+  assert.notEqual(result.exitCode, 0);
+  assert.match(result.stderr, /Missing required environment variable: LARKSUITE_CLI_APP_ID/);
+});
+
+test("exits with error when app secret is missing", async () => {
+  const result = await runScript({ TEST_BOT1_APP_SECRET: "" });
+  assert.notEqual(result.exitCode, 0);
+  assert.match(result.stderr, /Missing required environment variable: TEST_BOT1_APP_SECRET/);
+});
+
+test("fetches token and writes it to a private file", async () => {
+  const { server, port } = await startServer((req, res, body) => {
+    assert.equal(req.method, "POST");
+    const params = new URLSearchParams(body);
+    assert.equal(params.get("grant_type"), "client_credentials");
+    assert.equal(params.get("client_id"), "test_app_id");
+    assert.equal(params.get("client_secret"), "test-secret");
+    res.writeHead(200, { "Content-Type": "application/json" });
+    res.end(JSON.stringify({ code: 0, access_token: "test-token" }));
+  });
+
+  try {
+    const result = await runScript({
+      E2E_TAT_ENDPOINT: `http://127.0.0.1:${port}/token`,
+    });
+
+    assert.equal(result.exitCode, 0, `stderr: ${result.stderr}`);
+    assert.ok(result.stdout.includes("::add-mask::test-token"));
+    assert.ok(result.stdout.includes("Prepared shared live E2E tenant token"));
+
+    const tatPath = path.join(result.tmpDir, "e2e-live-tat");
+    assert.ok(fs.existsSync(tatPath), "token file should exist");
+
+    const stat = fs.statSync(tatPath);
+    assert.equal(stat.mode & 0o777, 0o600, "token file should be owner-only");
+    assert.equal(fs.readFileSync(tatPath, "utf8"), "test-token");
+
+    assert.ok(
+      result.output.includes(`path=${tatPath}`),
+      "should write path to GITHUB_OUTPUT",
+    );
+  } finally {
+    server.close();
+  }
+});
+
+test("retries an interrupted response and then succeeds", async () => {
+  let requestCount = 0;
+  const { server, port } = await startServer((req, res) => {
+    requestCount++;
+    if (requestCount === 1) {
+      abortResponse(res);
+      return;
+    }
+    res.writeHead(200, { "Content-Type": "application/json" });
+    res.end(JSON.stringify({ code: 0, access_token: "test-token" }));
+  });
+
+  try {
+    const result = await runScript({
+      E2E_TAT_ENDPOINT: `http://127.0.0.1:${port}/token`,
+    });
+
+    assert.equal(result.exitCode, 0, `stderr: ${result.stderr}`);
+    assert.equal(requestCount, 2);
+  } finally {
+    server.close();
+  }
+});
+
+test("fails after every interrupted response is retried", async () => {
+  let requestCount = 0;
+  const { server, port } = await startServer((req, res) => {
+    requestCount++;
+    abortResponse(res);
+  });
+
+  try {
+    const result = await runScript({
+      E2E_TAT_ENDPOINT: `http://127.0.0.1:${port}/token`,
+    });
+
+    assert.notEqual(result.exitCode, 0);
+    assert.equal(requestCount, 4);
+    assert.match(result.stderr, /Failed to fetch tenant access token/);
+  } finally {
+    server.close();
+  }
+});
+
+test("exits with error after all retries fail", async () => {
+  let requestCount = 0;
+  const { server, port } = await startServer((req, res) => {
+    requestCount++;
+    res.writeHead(500, { "Content-Type": "application/json" });
+    res.end(JSON.stringify({ code: 500, error: "server error" }));
+  });
+
+  try {
+    const result = await runScript({
+      E2E_TAT_ENDPOINT: `http://127.0.0.1:${port}/token`,
+    });
+
+    assert.notEqual(result.exitCode, 0);
+    assert.equal(requestCount, 4);
+    assert.match(result.stderr, /Failed to fetch tenant access token/);
+  } finally {
+    server.close();
+  }
+});

shortcuts/sheets/batch_key_vocab_test.go

@@ -0,0 +1,284 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package sheets
+
+import (
+	"strings"
+	"testing"
+)
+
+// subOp builds a raw +batch-update sub-op for translateBatchOp tests.
+func subOp(shortcut string, input map[string]interface{}) map[string]interface{} {
+	return map[string]interface{}{"shortcut": shortcut, "input": input}
+}
+
+// TestBatchOp_UnknownInputKeyRejected pins the key-vocabulary guard: an
+// off-vocabulary sub-op input key must error with a did-you-mean instead of
+// being silently ignored (silent ignore surfaced as misleading "missing
+// required flag" errors — the top batch error cluster in eval traces).
+func TestBatchOp_UnknownInputKeyRejected(t *testing.T) {
+	t.Parallel()
+
+	t.Run("invented key errors with did-you-mean", func(t *testing.T) {
+		t.Parallel()
+		_, err := translateBatchOp(subOp("+cells-set", map[string]interface{}{
+			"sheet_name": "S1",
+			"rangee":     "A1:B2",
+			"cells":      []interface{}{[]interface{}{map[string]interface{}{"value": "x"}}},
+		}), testToken, 0)
+		ve := requireValidation(t, err, `unknown input key "rangee"`)
+		if !strings.Contains(ve.Message, `did you mean "range"`) {
+			t.Fatalf("message %q missing did-you-mean", ve.Message)
+		}
+		if !strings.Contains(ve.Hint, "input keys:") {
+			t.Fatalf("hint %q missing key contract", ve.Hint)
+		}
+	})
+
+	t.Run("system flag is not sub-op vocabulary", func(t *testing.T) {
+		t.Parallel()
+		_, err := translateBatchOp(subOp("+cells-clear", map[string]interface{}{
+			"sheet_name": "S1",
+			"range":      "A1:B2",
+			"dry_run":    true,
+		}), testToken, 0)
+		requireValidation(t, err, `unknown input key "dry_run"`)
+	})
+
+	t.Run("reserved locator in hyphen form still rejected", func(t *testing.T) {
+		t.Parallel()
+		_, err := translateBatchOp(subOp("+cells-clear", map[string]interface{}{
+			"sheet_name":        "S1",
+			"range":             "A1:B2",
+			"spreadsheet-token": "shtXXX",
+		}), testToken, 0)
+		requireValidation(t, err, "do not pass input.spreadsheet-token")
+	})
+}
+
+// TestBatchOp_HabitualKeysRewritten pins the silent rewrites: camelCase onto
+// the declared flag, and the commandFlagAliases table (size → width/height on
+// the resize pair — the pre-2026-07 vocabulary and the styles-protocol
+// spelling, the single largest sub-op error cluster).
+func TestBatchOp_HabitualKeysRewritten(t *testing.T) {
+	t.Parallel()
+
+	t.Run("camelCase sheetName resolves", func(t *testing.T) {
+		t.Parallel()
+		translated, err := translateBatchOp(subOp("+cells-clear", map[string]interface{}{
+			"sheetName": "S1",
+			"range":     "A1:B2",
+		}), testToken, 0)
+		if err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+		input := translated["input"].(map[string]interface{})
+		if input["sheet_name"] != "S1" {
+			t.Fatalf("sheet_name = %v, want S1", input["sheet_name"])
+		}
+	})
+
+	t.Run("size aliases to width on +cols-resize", func(t *testing.T) {
+		t.Parallel()
+		translated, err := translateBatchOp(subOp("+cols-resize", map[string]interface{}{
+			"sheet_name": "S1",
+			"range":      "A:C",
+			"type":       "pixel",
+			"size":       float64(120),
+		}), testToken, 0)
+		if err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+		input := translated["input"].(map[string]interface{})
+		width, _ := input["resize_width"].(map[string]interface{})
+		if width["value"] != 120 {
+			t.Fatalf("resize_width = %v, want value 120", input["resize_width"])
+		}
+	})
+
+	t.Run("size aliases to height on +rows-resize", func(t *testing.T) {
+		t.Parallel()
+		_, err := translateBatchOp(subOp("+rows-resize", map[string]interface{}{
+			"sheet_name": "S1",
+			"range":      "1:3",
+			"type":       "pixel",
+			"size":       float64(36),
+		}), testToken, 0)
+		if err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+	})
+
+	t.Run("single-entry ranges unwraps onto range", func(t *testing.T) {
+		t.Parallel()
+		translated, err := translateBatchOp(subOp("+cells-clear", map[string]interface{}{
+			"sheet_name": "S1",
+			"ranges":     []interface{}{"A1:B2"},
+		}), testToken, 0)
+		if err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+		input := translated["input"].(map[string]interface{})
+		if input["range"] != "A1:B2" {
+			t.Fatalf("range = %v, want A1:B2", input["range"])
+		}
+	})
+
+	t.Run("multi-entry ranges prescribes a split", func(t *testing.T) {
+		t.Parallel()
+		_, err := translateBatchOp(subOp("+cells-clear", map[string]interface{}{
+			"sheet_name": "S1",
+			"ranges":     []interface{}{"A1:B2", "C1:D2"},
+		}), testToken, 0)
+		requireValidation(t, err, "split them into 2 sub-ops")
+	})
+}
+
+// TestBatchOperations_AggregatesValidationErrors pins the one-pass contract:
+// several invalid ops come back in a single error (each with its own
+// operations[i] context) instead of the first only — eval traces show
+// fix-one-resend loops of up to 7 round trips under first-error-only.
+func TestBatchOperations_AggregatesValidationErrors(t *testing.T) {
+	t.Parallel()
+
+	t.Run("two bad ops both reported", func(t *testing.T) {
+		t.Parallel()
+		_, err := translateBatchOperations([]interface{}{
+			subOp("+cells-clear", map[string]interface{}{"range": "A1:B2"}),                     // missing sheet selector
+			subOp("+cells-set", map[string]interface{}{"sheet_name": "S1", "range": "A1"}),      // missing cells
+			subOp("+cells-clear", map[string]interface{}{"sheet_name": "S1", "range": "A1:B2"}), // valid
+		}, testToken)
+		ve := requireValidation(t, err, "2 of 3 operations failed validation")
+		for _, want := range []string{"operations[0] (+cells-clear)", "operations[1] (+cells-set)", "--cells is required"} {
+			if !strings.Contains(ve.Message, want) {
+				t.Fatalf("message %q missing %q", ve.Message, want)
+			}
+		}
+	})
+
+	t.Run("single bad op keeps the standalone-shaped error", func(t *testing.T) {
+		t.Parallel()
+		_, err := translateBatchOperations([]interface{}{
+			subOp("+cells-set", map[string]interface{}{"sheet_name": "S1", "range": "A1"}),
+		}, testToken)
+		ve := requireValidation(t, err, "--cells is required")
+		if strings.Contains(ve.Message, "failed validation") {
+			t.Fatalf("single-error message must not use the aggregate wrapper: %q", ve.Message)
+		}
+	})
+}
+
+// TestCellsSetInput_MatrixPrecheck pins the local cells-vs-range guard that
+// front-runs the server's mid-batch "does not match range" failures.
+func TestCellsSetInput_MatrixPrecheck(t *testing.T) {
+	t.Parallel()
+
+	cases := []struct {
+		name         string
+		input        map[string]interface{}
+		wantContains string // "" = expect success
+	}{
+		{
+			"empty cells prescribes +cells-clear",
+			map[string]interface{}{"sheet_name": "S1", "range": "A1:B2", "cells": []interface{}{}},
+			"+cells-clear",
+		},
+		{
+			"row count mismatch",
+			map[string]interface{}{"sheet_name": "S1", "range": "A1:B3",
+				"cells": []interface{}{
+					[]interface{}{map[string]interface{}{"value": "a"}, map[string]interface{}{"value": "b"}},
+				}},
+			"has 1 rows but --range \"A1:B3\" spans 3 rows",
+		},
+		{
+			"column count mismatch",
+			map[string]interface{}{"sheet_name": "S1", "range": "A1:B1",
+				"cells": []interface{}{
+					[]interface{}{map[string]interface{}{"value": "a"}},
+				}},
+			"has 1 columns but --range \"A1:B1\" spans 2 columns",
+		},
+		{
+			"matching matrix passes",
+			map[string]interface{}{"sheet_name": "S1", "range": "A1:B2",
+				"cells": []interface{}{
+					[]interface{}{map[string]interface{}{"value": "a"}, map[string]interface{}{"value": "b"}},
+					[]interface{}{map[string]interface{}{"value": "c"}, map[string]interface{}{"value": "d"}},
+				}},
+			"",
+		},
+		{
+			"bare single-cell range enforces the 1x1 match (07-21: server rejects anchors too)",
+			map[string]interface{}{"sheet_name": "S1", "range": "A1",
+				"cells": []interface{}{
+					[]interface{}{map[string]interface{}{"value": "a"}, map[string]interface{}{"value": "b"}},
+				}},
+			"has 2 columns but --range \"A1\" spans 1 columns",
+		},
+		{
+			"single-cell range with a single cell passes",
+			map[string]interface{}{"sheet_name": "S1", "range": "B3",
+				"cells": []interface{}{
+					[]interface{}{map[string]interface{}{"value": "a"}},
+				}},
+			"",
+		},
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			_, err := translateBatchOp(subOp("+cells-set", tc.input), testToken, 0)
+			if tc.wantContains == "" {
+				if err != nil {
+					t.Fatalf("unexpected error: %v", err)
+				}
+				return
+			}
+			requireValidation(t, err, tc.wantContains)
+		})
+	}
+}
+
+// TestFlattenToolErrorMsg_PartialFailureRecovery pins the no-rollback recovery
+// prescription appended to server-side "N succeeded, M failed" errors.
+func TestFlattenToolErrorMsg_PartialFailureRecovery(t *testing.T) {
+	t.Parallel()
+
+	wrap := func(inner string) string {
+		return `{"error":` + jsonQuote(inner) + `}`
+	}
+
+	t.Run("single failure prescribes resend-from-index", func(t *testing.T) {
+		t.Parallel()
+		msg := flattenToolErrorMsg(wrap(`{"message":"batch_update: 4 succeeded, 1 failed","failures":[{"index":4,"tool_name":"set_cell_range","error":"cells is required"}]}`))
+		for _, want := range []string{"operations[4] (set_cell_range)", "no rollback", "resend only operations[4:]"} {
+			if !strings.Contains(msg, want) {
+				t.Fatalf("msg %q missing %q", msg, want)
+			}
+		}
+	})
+
+	t.Run("multiple failures prescribe failed-only resend", func(t *testing.T) {
+		t.Parallel()
+		msg := flattenToolErrorMsg(wrap(`{"message":"batch_update: 3 succeeded, 2 failed","failures":[{"index":1,"tool_name":"set_cell_range","error":"e1"},{"index":3,"tool_name":"resize_range","error":"e2"}]}`))
+		if !strings.Contains(msg, "resend only the failed operations") {
+			t.Fatalf("msg %q missing failed-only prescription", msg)
+		}
+	})
+
+	t.Run("zero succeeded gets no note", func(t *testing.T) {
+		t.Parallel()
+		msg := flattenToolErrorMsg(wrap(`{"message":"batch_update: 0 succeeded, 1 failed","failures":[{"index":0,"tool_name":"set_cell_range","error":"e"}]}`))
+		if strings.Contains(msg, "no rollback") {
+			t.Fatalf("msg %q must not carry the note when nothing was applied", msg)
+		}
+	})
+}
+
+// jsonQuote wraps s as a JSON string literal (escaping quotes), mirroring how
+// the server double-encodes the inner error payload.
+func jsonQuote(s string) string {
+	return `"` + strings.ReplaceAll(strings.ReplaceAll(s, `\`, `\\`), `"`, `\"`) + `"`
+}

shortcuts/sheets/batch_op_contract_test.go

@@ -763,7 +763,7 @@ func TestBatchOp_SchemaValidatesSubOps(t *testing.T) {
 		{
 			"+pivot-create summarize_by out of enum",
 			"+pivot-create",
-			`{"sheet-id":"sh1","source":"Sheet1!A1:D100","properties":{"values":[{"field":"A","summarize_by":"BOGUS"}]}}`,
+			`{"target_sheet_id":"sh1","source":"Sheet1!A1:D100","properties":{"values":[{"field":"A","summarize_by":"BOGUS"}]}}`,
 			"summarize_by",
 		},
 		// +chart-create properties.position.row has minimum:0 — P0

shortcuts/sheets/batch_op_dispatch.go

@@ -4,8 +4,11 @@
 package sheets
 
 import (
+	"fmt"
 	"sort"
 	"strings"
+
+	"github.com/larksuite/cli/internal/suggest"
 )
 
 // ─── +batch-update sub-op dispatch ─────────────────────────────────────
@@ -84,7 +87,14 @@ func objDeleteTranslate(spec objectCRUDSpec) batchTranslateFn {
 // flag error is identical too (locked by TestBatchOp_ErrorEquivalence).
 var batchOpDispatch = map[string]batchOpMapping{
 	// ─── 单元格内容 ──────────────────────────────────────────────────
-	"+cells-set":       {"set_cell_range", cellsSetInput},
+	"+cells-set": {"set_cell_range", func(fv flagView, token, sid, sname string) (map[string]interface{}, error) {
+		// The --writes plural form expands into its own atomic batch and
+		// cannot nest; sub-ops carry one range+cells each.
+		if fv.Changed("writes") {
+			return nil, sheetsValidationForFlag("writes", `"writes" is not supported inside +batch-update (it expands into its own atomic batch); call +cells-set --writes standalone, or give each sub-op a single range + cells`)
+		}
+		return cellsSetInput(fv, token, sid, sname)
+	}},
 	"+cells-set-style": {"set_cell_range", cellsSetStyleInput},
 	"+cells-clear":     {"clear_cell_range", cellsClearInput},
 	"+cells-replace":   {"replace_data", replaceInput},
@@ -102,6 +112,11 @@ var batchOpDispatch = map[string]batchOpMapping{
 	// ─── 行列结构 (modify_sheet_structure, operation 区分) ──────────
 	"+dim-insert": {"modify_sheet_structure", dimInsertInput},
 	"+dim-delete": {"modify_sheet_structure", func(fv flagView, token, sid, sname string) (map[string]interface{}, error) {
+		// The --ranges plural form expands into its own atomic batch and
+		// cannot nest; sub-ops carry one range each.
+		if fv.Changed("ranges") {
+			return nil, sheetsValidationForFlag("ranges", `"ranges" is not supported inside +batch-update (it expands into its own atomic batch); call +dim-delete --ranges standalone, or give each sub-op a single "range"`)
+		}
 		return dimRangeOpInput(fv, token, sid, sname, "delete")
 	}},
 	"+dim-hide": {"modify_sheet_structure", func(fv flagView, token, sid, sname string) (map[string]interface{}, error) {
@@ -301,6 +316,133 @@ func sheetMoveBatchInput(fv flagView, token, sheetID, sheetName string) (map[str
 // +batch-update 顶层 --url/--token 统一提供（excel_id / spreadsheet_token / url）。
 var reservedSubOpKeys = []string{"excel_id", "spreadsheet_token", "url"}
 
+// wrappedSubOpInputKeys are nested MCP-body container keys that must never
+// appear at a sub-op input's top level — their presence means the caller
+// pasted a shortcut's structured *output* (e.g. a {"cell_styles":{…}} block)
+// where the flattened flag keys belong. None of the batch sub-op translators
+// read input under these names, so rejecting them is safe.
+var wrappedSubOpInputKeys = []string{"cell_styles", "cell_merges", "styles"}
+
+// subOpKeyVocabulary returns the set of hyphen-canonical flag names a sub-op
+// input may carry for `sc`: every non-system flag in flag-defs except the
+// spreadsheet locators (reserved for the batch top level). Nil when the
+// shortcut has no flag-defs entry (vocabulary checks are then skipped).
+func subOpKeyVocabulary(sc string) map[string]bool {
+	defs, _ := loadFlagDefs()
+	spec, ok := defs[sc]
+	if !ok {
+		return nil
+	}
+	vocab := make(map[string]bool, len(spec.Flags))
+	for _, df := range spec.Flags {
+		if df.Kind == "system" || df.Name == "url" || df.Name == "spreadsheet-token" {
+			continue
+		}
+		vocab[df.Name] = true
+	}
+	return vocab
+}
+
+// camelToKebab converts a lowerCamelCase key to its kebab form
+// (sheetName → sheet-name). Returns "" when the key carries no uppercase
+// letter (nothing to convert).
+func camelToKebab(key string) string {
+	if strings.ToLower(key) == key {
+		return ""
+	}
+	var b strings.Builder
+	for i, r := range key {
+		if r >= 'A' && r <= 'Z' {
+			if i > 0 {
+				b.WriteByte('-')
+			}
+			b.WriteRune(r + ('a' - 'A'))
+			continue
+		}
+		b.WriteRune(r)
+	}
+	return b.String()
+}
+
+// normalizeSubOpInputKeys validates every sub-op input key against the
+// shortcut's flag vocabulary, rewriting habitual spellings in place and
+// rejecting anything that matches nothing. Eval traces show unknown keys were
+// previously ignored silently, which turned "wrong key" (size for width,
+// camelCase sheetName, an invented styles object) into misleading
+// "missing required flag" errors downstream — the single largest batch error
+// cluster. Rewrites applied, in order:
+//
+//   - underscore ↔ hyphen forms of a declared flag (already tolerated by
+//     mapFlagView — accepted here as-is)
+//   - lowerCamelCase → the declared flag (sheetName → sheet_name)
+//   - the command's intuitive-alias table (size → width/height on the resize
+//     pair) — the same commandFlagAliases the cobra path applies
+//   - "ranges" with a single-entry array unwraps onto "range"; a multi-entry
+//     array gets a split-into-sub-ops prescription instead
+//
+// Anything else errors with a did-you-mean. Returns a bare error; the caller
+// wraps it with the operations[i] (<shortcut>) context and key contract.
+func normalizeSubOpInputKeys(sc string, input map[string]interface{}) error {
+	vocab := subOpKeyVocabulary(sc)
+	if vocab == nil {
+		return nil
+	}
+	keys := make([]string, 0, len(input))
+	for k := range input {
+		keys = append(keys, k)
+	}
+	sort.Strings(keys)
+	aliases := commandFlagAliases[sc]
+	for _, k := range keys {
+		hv := strings.ReplaceAll(k, "_", "-")
+		if vocab[hv] {
+			continue
+		}
+		if kebab := camelToKebab(k); kebab != "" && vocab[kebab] {
+			input[strings.ReplaceAll(kebab, "-", "_")] = input[k]
+			delete(input, k)
+			continue
+		}
+		if target, ok := aliases[strings.ToLower(hv)]; ok && vocab[target] {
+			if _, taken := input[target]; !taken {
+				if _, taken := input[strings.ReplaceAll(target, "-", "_")]; !taken {
+					input[target] = input[k]
+					delete(input, k)
+					continue
+				}
+			}
+		}
+		if strings.ToLower(hv) == "ranges" && vocab["range"] && !vocab["ranges"] {
+			if arr, isArr := input[k].([]interface{}); isArr {
+				if len(arr) == 1 {
+					if s, isStr := arr[0].(string); isStr {
+						input["range"] = s
+						delete(input, k)
+						continue
+					}
+				}
+				return fmt.Errorf("%s takes a single \"range\" per sub-op, got %d entries in %q — split them into %d sub-ops (one per range)", sc, len(arr), k, len(arr)) //nolint:forbidigo // intermediate error; the batch dispatcher wraps it into a typed operations validation error
+			}
+			if s, isStr := input[k].(string); isStr {
+				input["range"] = s
+				delete(input, k)
+				continue
+			}
+		}
+		msg := fmt.Sprintf("unknown input key %q", k)
+		display := make([]string, 0, len(vocab))
+		for name := range vocab {
+			display = append(display, strings.ReplaceAll(name, "-", "_"))
+		}
+		sort.Strings(display)
+		if match := suggest.Closest(strings.ToLower(hv), display, 1); len(match) > 0 {
+			msg += fmt.Sprintf(" — did you mean %q?", match[0])
+		}
+		return fmt.Errorf("%s", msg) //nolint:forbidigo // intermediate error; the batch dispatcher wraps it into a typed operations validation error
+	}
+	return nil
+}
+
 // translateBatchOp 把一个 CLI 视角的 {shortcut, input} 翻成底层 MCP
 // batch_update 的 {tool_name, input}。`index` 用于错误信息定位。input 用
 // shortcut 的 CLI flag 名（连字符/下划线均可），经该 shortcut 的 standalone
@@ -312,6 +454,7 @@ var reservedSubOpKeys = []string{"excel_id", "spreadsheet_token", "url"}
 //   - input 不是 object
 //   - input 里手填了 operation（由 shortcut 名隐含，禁手填以防 mismatch）
 //   - input 里手填了 excel_id / spreadsheet_token / url
+//   - input 顶层出现 cell_styles / cell_merges / styles（误贴 MCP body 包裹结构）
 //   - 子操作的 translator 报错（如缺必填字段）
 func translateBatchOp(raw interface{}, token string, index int) (map[string]interface{}, error) {
 	op, ok := raw.(map[string]interface{})
@@ -335,7 +478,7 @@ func translateBatchOp(raw interface{}, token string, index int) (map[string]inte
 		return nil, sheetsValidationForFlag(
 			"operations",
 			"operations[%d]: shortcut %q not allowed in +batch-update "+
-				"(read ops / fan-out wrappers like +batch-update / +cells-batch-set-style / +cells-batch-clear / +dropdown-{update,delete} are excluded)",
+				"(read ops / fan-out wrappers like +batch-update / +styles-put / +cells-batch-set-style / +cells-batch-clear / +dropdown-{update,delete} are excluded)",
 			index, sc,
 		).WithHint("allowed shortcuts: %s", strings.Join(allowedBatchShortcuts(), ", "))
 	}
@@ -358,11 +501,30 @@ func translateBatchOp(raw interface{}, token string, index int) (map[string]inte
 		)
 	}
 	// 禁在 sub-op 重复填 spreadsheet 定位 —— 由 +batch-update 顶层 --url/--token 统一提供。
-	for _, k := range reservedSubOpKeys {
+	// 连字符 / 下划线两种写法都算命中（spreadsheet-token 与 spreadsheet_token 同罪）。
+	for userKey := range input {
+		normalized := strings.ReplaceAll(userKey, "-", "_")
+		for _, k := range reservedSubOpKeys {
+			if normalized == k {
+				return nil, sheetsValidationForFlag(
+					"operations",
+					"operations[%d] (%s): do not pass input.%s — it is already set from +batch-update top-level --url / --token",
+					index, sc, userKey,
+				)
+			}
+		}
+	}
+	// Reject a "wrapped structure" sub-op input: agents copy a shortcut's nested
+	// output container (e.g. +workbook-create --styles' {"cell_styles":{…}}) into
+	// the op input, but the op input is the shortcut's own flags flattened into
+	// JSON keys, not that wrapper. Left unflagged this surfaces far downstream as
+	// an unrelated "at least one style flag is required" (helpers.go), which never
+	// points at the real mistake.
+	for _, k := range wrappedSubOpInputKeys {
 		if _, has := input[k]; has {
 			return nil, sheetsValidationForFlag(
 				"operations",
-				"operations[%d] (%s): do not pass input.%s — it is already set from +batch-update top-level --url / --token",
+				`operations[%d] (%s): op input is the shortcut's flags flattened as JSON keys (e.g. "background_color": "#EBF1F8"); do not wrap in %s`,
 				index, sc, k,
 			)
 		}
@@ -373,6 +535,16 @@ func translateBatchOp(raw interface{}, token string, index int) (map[string]inte
 			return nil, sheetsValidationForFlag("operations", "operations[%d] (%s): unknown top-level key %q (expected only 'shortcut' and 'input')", index, sc, k)
 		}
 	}
+	// Reject / rewrite off-vocabulary input keys BEFORE any value reads: an
+	// unknown key silently ignored surfaces later as a misleading
+	// "missing required flag" error (the top batch error cluster in evals).
+	if err := normalizeSubOpInputKeys(sc, input); err != nil {
+		verr := sheetsValidationForFlag("operations", "operations[%d] (%s): %v", index, sc, err)
+		if contract := subOpInputContract(sc); contract != "" {
+			verr = verr.WithHint("%s input keys: %s", sc, contract)
+		}
+		return nil, verr
+	}
 	fv := newMapFlagViewForCommand(sc, input)
 	// operations is skipped by parse-time schema validation, so type-check the
 	// sub-op's scalar fields here before the translator reads them via
@@ -410,7 +582,14 @@ func translateBatchOp(raw interface{}, token string, index int) (map[string]inte
 // matrix, on the operations axis.
 const maxBatchOperations = 100
 
-// translateBatchOperations 翻译整个 ops 数组；fail-fast，遇错立即返回。
+// batchOpErrorDisplayLimit bounds how many per-op validation failures ride
+// on one aggregated --operations error, mirroring the schema validator's
+// display cap.
+const batchOpErrorDisplayLimit = 5
+
+// translateBatchOperations 翻译整个 ops 数组。逐 op 校验并**收集全部失败**
+// 一次性返回（不再 fail-fast）——agent 一轮就能修完所有坏 op，而不是
+// 修一个、重试、再撞下一个。cell 安全上限仍是全局判定，命中即返回。
 func translateBatchOperations(rawOps []interface{}, token string) ([]interface{}, error) {
 	if len(rawOps) == 0 {
 		return nil, sheetsValidationForFlag("operations", "--operations must be a non-empty JSON array")
@@ -422,10 +601,15 @@ func translateBatchOperations(rawOps []interface{}, token string) ([]interface{}
 	}
 	out := make([]interface{}, 0, len(rawOps))
 	var totalCells int64
+	var opErrs []error
 	for i, raw := range rawOps {
 		translated, err := translateBatchOp(raw, token, i)
 		if err != nil {
-			return nil, err
+			opErrs = append(opErrs, err)
+			continue
+		}
+		if len(opErrs) > 0 {
+			continue // already failing — keep scanning for more bad ops, skip cell math.
 		}
 		totalCells += translatedCellCount(translated)
 		if totalCells > maxStampMatrixCells {
@@ -435,7 +619,27 @@ func translateBatchOperations(rawOps []interface{}, token string) ([]interface{}
 		}
 		out = append(out, translated)
 	}
-	return out, nil
+	switch len(opErrs) {
+	case 0:
+		return out, nil
+	case 1:
+		return nil, opErrs[0] // single failure keeps the historical error byte-for-byte.
+	}
+	shown := opErrs
+	truncated := false
+	if len(shown) > batchOpErrorDisplayLimit {
+		shown = shown[:batchOpErrorDisplayLimit]
+		truncated = true
+	}
+	parts := make([]string, 0, len(shown))
+	for i, e := range shown {
+		parts = append(parts, fmt.Sprintf("%d) %s", i+1, e.Error()))
+	}
+	msg := fmt.Sprintf("%d of %d operations failed validation: %s", len(opErrs), len(rawOps), strings.Join(parts, "; "))
+	if truncated {
+		msg += fmt.Sprintf("; (%d more not shown — fix these first)", len(opErrs)-batchOpErrorDisplayLimit)
+	}
+	return nil, sheetsValidationForFlag("operations", "%s", msg).WithCause(opErrs[0])
 }
 
 func translatedCellCount(op map[string]interface{}) int64 {

shortcuts/sheets/cells_set_writes_test.go

@@ -0,0 +1,113 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package sheets
+
+import (
+	"strings"
+	"testing"
+)
+
+// TestCellsSetWrites pins the --writes plural form: scattered (cross-sheet)
+// regions fan into ONE atomic batch_update, each item self-carrying its
+// sheet selector (no top-level fallback — same convention as +batch-update
+// sub-ops and +styles-put items), with per-item errors aggregated.
+func TestCellsSetWrites(t *testing.T) {
+	t.Parallel()
+
+	writes := func(items string, extra ...string) (string, string, error) {
+		args := append([]string{
+			"--url", testURL, "--dry-run", "--writes", items,
+		}, extra...)
+		return runShortcutCapturingErr(t, CellsSet, args)
+	}
+
+	t.Run("cross-sheet items expand into one batch", func(t *testing.T) {
+		t.Parallel()
+		stdout, _, err := writes(`[
+			{"sheet_name":"明细","range":"D5","cells":[[{"formula":"=IFERROR(C5/B5,0)"}]]},
+			{"sheet_name":"汇总","range":"B3","cells":[[{"formula":"=SUM(C:C)"}]]}
+		]`)
+		if err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+		for _, want := range []string{"batch_update", "明细", "汇总", "IFERROR"} {
+			if !strings.Contains(stdout, want) {
+				t.Fatalf("dry-run body missing %q: %s", want, stdout[:min(len(stdout), 400)])
+			}
+		}
+	})
+
+	t.Run("item without sheet selector errors", func(t *testing.T) {
+		t.Parallel()
+		_, _, err := writes(`[{"range":"A1","cells":[[{"value":"x"}]]}]`)
+		requireValidation(t, err, "sheet-id or --sheet-name")
+	})
+
+	t.Run("top-level sheet selector rejected with prescription", func(t *testing.T) {
+		t.Parallel()
+		_, _, err := writes(`[{"sheet_name":"S1","range":"A1","cells":[[{"value":"x"}]]}]`,
+			"--sheet-name", "S1")
+		requireValidation(t, err, "put sheet_name (or sheet_id) inside each writes item")
+	})
+
+	t.Run("writes and range are mutually exclusive", func(t *testing.T) {
+		t.Parallel()
+		_, _, err := writes(`[{"sheet_name":"S1","range":"A1","cells":[[{"value":"x"}]]}]`,
+			"--range", "A1")
+		requireValidation(t, err, "mutually exclusive")
+	})
+
+	t.Run("per-item errors aggregate", func(t *testing.T) {
+		t.Parallel()
+		// Both items pass the --writes schema (range+cells present) but fail
+		// deeper: item 0 a matrix mismatch, item 1 a missing sheet selector.
+		_, _, err := writes(`[
+			{"sheet_name":"S1","range":"A1:B2","cells":[[{"value":"x"}]]},
+			{"range":"C1","cells":[[{"value":"y"}]]}
+		]`)
+		ve := requireValidation(t, err, "--writes has 2 issues")
+		for _, want := range []string{"--writes[0]", "--writes[1]", "sheet-name"} {
+			if !strings.Contains(ve.Message, want) {
+				t.Fatalf("message %q missing %q", ve.Message, want)
+			}
+		}
+	})
+
+	t.Run("item keys go through the vocabulary layer", func(t *testing.T) {
+		t.Parallel()
+		stdout, _, err := writes(`[{"sheetName":"S1","range":"A1","cells":[[{"value":"x"}]]}]`)
+		if err != nil {
+			t.Fatalf("camelCase sheetName must normalize: %v", err)
+		}
+		if !strings.Contains(stdout, "S1") {
+			t.Fatalf("normalized item missing sheet: %s", stdout[:min(len(stdout), 300)])
+		}
+	})
+
+	t.Run("cannot nest inside batch-update", func(t *testing.T) {
+		t.Parallel()
+		_, err := translateBatchOp(subOp("+cells-set", map[string]interface{}{
+			"writes": []interface{}{map[string]interface{}{
+				"sheet_name": "S1", "range": "A1", "cells": []interface{}{[]interface{}{map[string]interface{}{"value": "x"}}},
+			}},
+		}), testToken, 0)
+		requireValidation(t, err, "not supported inside +batch-update")
+	})
+
+	t.Run("styles flag gets the layering prescription", func(t *testing.T) {
+		t.Parallel()
+		// Ergonomics (FlagErrorFunc hints) mount via the registry, not the
+		// bare shortcut var — mirror the real CLI wiring.
+		sc := shortcutFromRegistry(t, "+cells-set")
+		_, _, err := runShortcutCapturingErr(t, sc, []string{
+			"--url", testURL, "--dry-run",
+			"--writes", `[{"sheet_name":"S1","range":"A1","cells":[[{"value":"x"}]]}]`,
+			"--styles", `{"styles":[]}`,
+		})
+		ve := requireValidation(t, err, "unknown flag")
+		if !strings.Contains(ve.Hint, "+styles-put") || !strings.Contains(ve.Hint, "cell_styles") {
+			t.Fatalf("want the styles-put layering hint, got hint=%q", ve.Hint)
+		}
+	})
+}

shortcuts/sheets/chart_examples.go

@@ -0,0 +1,150 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package sheets
+
+import (
+	"fmt"
+	"sort"
+	"strings"
+
+	"github.com/larksuite/cli/shortcuts/common"
+	"github.com/spf13/cobra"
+)
+
+// ─── +chart-create --print-example ─────────────────────────────────────
+//
+// chart-create's --properties schema is ~1,750 pretty-printed lines; eval
+// traces show agents paging through the full --print-schema dump for every
+// chart (25 round trips in one 35-task batch) and still missing deep
+// required fields. A ready-to-edit minimal template per chart type answers
+// the actual question ("what does a valid payload look like") in one local
+// call. Wired through PostMount, same pattern as +csv-put's flag-group
+// tweaks — no framework change.
+//
+// Templates mirror the canonical examples in the lark-sheets-chart
+// reference (sheet-skill-spec canonical-spec/references/lark_sheet_chart):
+// inline headerMode with refs covering the header row, 1-based indices,
+// quoted sheet prefix in refs.
+
+var chartExampleTemplates = map[string]string{
+	"column": chartSimpleExample("column"),
+	"bar":    chartSimpleExample("bar"),
+	"line":   chartSimpleExample("line"),
+	"area":   chartSimpleExample("area"),
+	"radar":  chartSimpleExample("radar"),
+	"scatter": `{
+  "position": {"row": 1, "col": "F"},
+  "size": {"width": 600, "height": 400},
+  "snapshot": {
+    "title": {"text": "图表标题"},
+    "plotArea": {"plot": {"type": "scatter"}},
+    "data": {
+      "refs": [{"value": "'Sheet1'!A1:B20"}],
+      "dim1": {"serie": {"index": 1}},
+      "dim2": {"series": [{"index": 2}]}
+    }
+  }
+}`,
+	"pie": `{
+  "position": {"row": 1, "col": "F"},
+  "size": {"width": 600, "height": 450},
+  "snapshot": {
+    "title": {"text": "占比标题"},
+    "plotArea": {"plot": {
+      "type": "pie",
+      "series": [{
+        "index": 1,
+        "sectors": {"sector": [{"index": 1, "offsetRadius": 0.05}]}
+      }]
+    }},
+    "data": {
+      "refs": [{"value": "'Sheet1'!A1:B11"}],
+      "dim1": {"serie": {"index": 1, "aggregate": true}},
+      "dim2": {"series": [{"index": 2, "aggregateType": "sum"}]}
+    }
+  }
+}`,
+	"combo": `{
+  "position": {"row": 1, "col": "F"},
+  "size": {"width": 700, "height": 400},
+  "snapshot": {
+    "title": {"text": "柱线组合"},
+    "plotArea": {"plot": {
+      "type": "combo",
+      "series": [
+        {"index": 2, "comboType": "column"},
+        {"index": 3, "comboType": "line"}
+      ]
+    }},
+    "data": {
+      "refs": [{"value": "'Sheet1'!A1:C13"}],
+      "dim1": {"serie": {"index": 1}},
+      "dim2": {"series": [{"index": 2}, {"index": 3}]}
+    }
+  }
+}`,
+}
+
+// chartSimpleExample renders the shared minimal shape for plot types that
+// need nothing beyond plot.type (column / bar / line / area / radar).
+func chartSimpleExample(typ string) string {
+	return fmt.Sprintf(`{
+  "position": {"row": 1, "col": "F"},
+  "size": {"width": 600, "height": 400},
+  "snapshot": {
+    "title": {"text": "图表标题"},
+    "plotArea": {"plot": {"type": %q}},
+    "data": {
+      "refs": [{"value": "'Sheet1'!A1:C10"}],
+      "dim1": {"serie": {"index": 1}},
+      "dim2": {"series": [{"index": 2}, {"index": 3}]}
+    }
+  }
+}`, typ)
+}
+
+func chartExampleTypes() []string {
+	types := make([]string, 0, len(chartExampleTemplates))
+	for t := range chartExampleTemplates {
+		types = append(types, t)
+	}
+	sort.Strings(types)
+	return types
+}
+
+// withChartPrintExample wraps +chart-create's PostMount so the command grows
+// a --print-example flag that short-circuits execution and prints a minimal
+// ready-to-edit --properties template — purely local, no identity or
+// network. --properties' cobra-level required annotation is relaxed (the
+// input builder still enforces it on the real path, same trick as
+// +csv-put's --csv).
+func withChartPrintExample(prev func(cmd *cobra.Command)) func(cmd *cobra.Command) {
+	return func(cmd *cobra.Command) {
+		if prev != nil {
+			prev(cmd)
+		}
+		cmd.Flags().String("print-example", "",
+			"Print a minimal ready-to-edit --properties template for a chart type ("+strings.Join(chartExampleTypes(), "|")+") and exit")
+		// Only --properties carries a cobra-level required annotation (the
+		// locator flags are xor pairs, enforced later); the input builder
+		// still errors "--properties is required" on the real path.
+		if fl := cmd.Flags().Lookup("properties"); fl != nil {
+			delete(fl.Annotations, cobra.BashCompOneRequiredFlag)
+		}
+		prevRunE := cmd.RunE
+		cmd.RunE = func(c *cobra.Command, args []string) error {
+			typ, _ := c.Flags().GetString("print-example")
+			if typ == "" {
+				return prevRunE(c, args)
+			}
+			tmpl, ok := chartExampleTemplates[typ]
+			if !ok {
+				return common.ValidationErrorf("no example for chart type %q; available: %s",
+					typ, strings.Join(chartExampleTypes(), ", ")).WithParam("--print-example")
+			}
+			fmt.Fprintln(c.OutOrStdout(), tmpl)
+			return nil
+		}
+	}
+}

shortcuts/sheets/chart_examples_test.go

@@ -0,0 +1,63 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package sheets
+
+import (
+	"bytes"
+	"encoding/json"
+	"strings"
+	"testing"
+)
+
+// TestChartPrintExample pins the --print-example contract: a known type
+// prints its template and skips execution entirely; an unknown type lists
+// the available ones.
+func TestChartPrintExample(t *testing.T) {
+	t.Parallel()
+
+	t.Run("prints template without locator flags", func(t *testing.T) {
+		t.Parallel()
+		sc := shortcutFromRegistry(t, "+chart-create")
+		parent, _, _, _ := newTestRig(t, sc)
+		var buf bytes.Buffer
+		parent.SetOut(&buf) // --print-example writes via cobra's OutOrStdout
+		parent.SetArgs([]string{sc.Command, "--print-example", "pie"})
+		if err := parent.Execute(); err != nil {
+			t.Fatalf("print-example should run standalone, got: %v", err)
+		}
+		if !strings.Contains(buf.String(), `"sectors"`) {
+			t.Errorf("pie template should carry sectors, got %q", buf.String())
+		}
+	})
+
+	t.Run("unknown type lists available", func(t *testing.T) {
+		t.Parallel()
+		sc := shortcutFromRegistry(t, "+chart-create")
+		_, _, err := runShortcutCapturingErr(t, sc, []string{"--print-example", "donut"})
+		ve := requireValidation(t, err, `no example for chart type "donut"`)
+		if !strings.Contains(ve.Message, "pie") {
+			t.Errorf("message should list available types, got %q", ve.Message)
+		}
+	})
+}
+
+// TestChartExampleTemplates_ValidateAgainstSchema drift-guards every
+// template against the embedded chart-create properties schema — a template
+// the CLI itself would reject is worse than none.
+func TestChartExampleTemplates_ValidateAgainstSchema(t *testing.T) {
+	t.Parallel()
+	for typ, tmpl := range chartExampleTemplates {
+		t.Run(typ, func(t *testing.T) {
+			t.Parallel()
+			var v interface{}
+			if err := json.Unmarshal([]byte(tmpl), &v); err != nil {
+				t.Fatalf("template is not valid JSON: %v", err)
+			}
+			fv := newMapFlagViewForCommand("+chart-create", map[string]interface{}{"properties": v})
+			if err := validateValueAgainstSchema(fv, "properties", v); err != nil {
+				t.Errorf("template rejected by embedded schema: %v", err)
+			}
+		})
+	}
+}

shortcuts/sheets/data/flag-defs.json

@@ -821,12 +821,10 @@
         "kind": "own",
         "type": "string",
         "required": "optional",
-        "desc": "Style inheritance for the new row/column: `before` (from preceding) / `after` (from following) / `none` (default)",
-        "default": "none",
+        "desc": "Style inheritance for the new row/column: `before` (from the preceding row/column) / `after` (from the following row/column). Omit the flag to inherit the following row/column (same as `after`) — the backend cannot leave a new row/column unstyled; for a truly blank row/column, clear formats afterwards with +cells-clear --scope formats. Insertion always lands before `--position`; this only selects which side's style is copied.",
         "enum": [
           "before",
-          "after",
-          "none"
+          "after"
         ]
       },
       {
@@ -887,8 +885,19 @@
         "name": "range",
         "kind": "own",
         "type": "string",
-        "required": "required",
-        "desc": "Row/column closed range to delete; rows use 1-based numbers like `3:7` or `5` (single row), columns use letters like `C:F` or `C`"
+        "required": "xor",
+        "desc": "Row/column closed range to delete; rows use 1-based numbers like `3:7` or `5` (single row), columns use letters like `C:F` or `C`. XOR with `--ranges`"
+      },
+      {
+        "name": "ranges",
+        "kind": "own",
+        "type": "string",
+        "required": "xor",
+        "desc": "Multiple row/column ranges to delete as a JSON array (up to 100 items, e.g. `[\"5:5\",\"8:8\",\"11:13\"]` or `[\"C:C\",\"F:G\"]`); rows and columns cannot be mixed, ranges must not overlap; XOR with `--range`. CLI sorts positions in DESCENDING order into one atomic batch delete — ascending deletion would shift later indexes as earlier rows/columns disappear; the CLI handles the ordering",
+        "input": [
+          "file",
+          "stdin"
+        ]
       },
       {
         "name": "yes",
@@ -1277,13 +1286,14 @@
         "kind": "own",
         "type": "string_slice",
         "required": "optional",
-        "desc": "Comma-separated info categories to include",
+        "desc": "Comma-separated info categories to include. `truncation` additionally estimates whether each cell's content is clipped (by row height / col width / font size / wrap) and returns `isRowTruncated` / `isColTruncated` (extra compute; enable only for layout checks or before adjusting row heights / column widths)",
         "enum": [
           "value",
           "formula",
           "style",
           "comment",
-          "data_validation"
+          "data_validation",
+          "truncation"
         ]
       },
       {
@@ -1291,15 +1301,29 @@
         "kind": "own",
         "type": "int",
         "required": "optional",
-        "desc": "Max output chars per call; default 500000 (safety cap). Large reads are usually better redirected to a file; only lower it (e.g. 25000) when you want results inline without triggering file offload, paging via has_more",
+        "desc": "Max output chars per call; default 500000 (safety cap). For a full untruncated read, use --output-path to dump to a file (auto-unlimited); only lower it (e.g. 25000) when you want results inline without a file, paging via has_more.",
         "default": "500000"
       },
+      {
+        "name": "output-path",
+        "kind": "own",
+        "type": "string",
+        "required": "optional",
+        "desc": "Write the full read result to a local path (e.g. `./out.json`); the file holds the data payload as JSON while stdout returns only a small confirmation (output_path, byte count). **When set, the char cap defaults to unlimited** (overriding the --max-chars default), so a large sheet can be dumped in full for later analysis without stdout being truncated by max_chars. Omit it to print to stdout as usual."
+      },
       {
         "name": "skip-hidden",
         "kind": "own",
         "type": "bool",
         "required": "optional",
-        "desc": "Skip hidden rows and columns; default `false`"
+        "desc": "Skip hidden or collapsed rows and columns. Default `false`; when `--skip-filter` is omitted, filtered-out rows follow this value for backward compatibility"
+      },
+      {
+        "name": "skip-filter",
+        "kind": "own",
+        "type": "bool",
+        "required": "optional",
+        "desc": "Skip filtered-out rows. When omitted, inherits `--skip-hidden`; explicitly set to `false` to keep filtered-out rows while skipping hidden rows and columns"
       },
       {
         "name": "dry-run",
@@ -1392,17 +1416,24 @@
         "name": "range",
         "kind": "own",
         "type": "string",
-        "required": "required",
-        "desc": "A1 range, e.g. `A1:F30` (no sheet prefix — use `--sheet-id` / `--sheet-name` to select the sheet)"
+        "required": "optional",
+        "desc": "A1 range, e.g. `A1:F30` (no sheet prefix — use `--sheet-id` / `--sheet-name` to select the sheet). Optional: when omitted the whole sheet is read (clipped to the actual grid bounds; actual_range in the response names what was read); pair with --max-chars / --output-path on large sheets"
       },
       {
         "name": "max-chars",
         "kind": "own",
         "type": "int",
         "required": "optional",
-        "desc": "Max output chars per call; default 500000 (safety cap). Large reads are usually better redirected to a file; only lower it (e.g. 25000) when you want results inline without triggering file offload, paging via has_more",
+        "desc": "Max output chars per call; default 500000 (safety cap). For a full untruncated read, use --output-path to dump to a file (auto-unlimited); only lower it (e.g. 25000) when you want results inline without a file, paging via has_more.",
         "default": "500000"
       },
+      {
+        "name": "output-path",
+        "kind": "own",
+        "type": "string",
+        "required": "optional",
+        "desc": "Write the full read result to a local path (e.g. `./out.json`); the file holds the data payload as JSON while stdout returns only a small confirmation (output_path, byte count). **When set, the char cap defaults to unlimited** (overriding the --max-chars default), so a large sheet can be dumped in full for later analysis without stdout being truncated by max_chars. Omit it to print to stdout as usual."
+      },
       {
         "name": "include-row-prefix",
         "kind": "own",
@@ -1465,6 +1496,21 @@
         "required": "optional",
         "desc": "A1 range to read; omit to read each sheet's full used range (spans internal blank rows/columns, not just the A1 current region)"
       },
+      {
+        "name": "max-chars",
+        "kind": "own",
+        "type": "int",
+        "required": "optional",
+        "desc": "Max output chars per call; default 500000 (safety cap). The underlying tool truncates at ~50000 even when unset, so this is sent explicitly to raise it; for a full untruncated read use --output-path (auto-unlimited).",
+        "default": "500000"
+      },
+      {
+        "name": "output-path",
+        "kind": "own",
+        "type": "string",
+        "required": "optional",
+        "desc": "Write the full read result to a local path (e.g. `./out.json`); the file holds the data payload as JSON while stdout returns only a small confirmation (output_path, byte count). **When set, the char cap defaults to unlimited** (overriding the --max-chars default), so a large sheet can be dumped in full for later analysis without stdout being truncated by max_chars. Omit it to print to stdout as usual."
+      },
       {
         "name": "no-header",
         "kind": "own",
@@ -1691,33 +1737,44 @@
         "kind": "public",
         "type": "string",
         "required": "xor",
-        "desc": "Sheet reference_id (XOR with `--sheet-name`)"
+        "desc": "Sheet reference_id (XOR with `--sheet-name`); not accepted with `--writes` (each writes item carries its own sheet selector)"
       },
       {
         "name": "sheet-name",
         "kind": "public",
         "type": "string",
         "required": "xor",
-        "desc": "Sheet name (XOR with `--sheet-id`)"
+        "desc": "Sheet name (XOR with `--sheet-id`); not accepted with `--writes` (each writes item carries its own sheet selector)"
       },
       {
         "name": "range",
         "kind": "own",
         "type": "string",
-        "required": "required",
-        "desc": "Write range (A1 notation)"
+        "required": "xor",
+        "desc": "Write range (A1 notation). XOR with `--writes` (single region: --range+--cells; multiple regions: --writes)"
       },
       {
         "name": "cells",
         "kind": "own",
         "type": "string",
-        "required": "required",
+        "required": "xor",
         "desc": "JSON 2D array `[[{cell},...],...]`, dimensions must match `--range`; each cell may carry `value` / `formula` / `cell_styles` / `note` / `rich_text` (incl. `type=\"embed-image\"` in-cell image); run `--print-schema` for full fields",
         "input": [
           "file",
           "stdin"
         ]
       },
+      {
+        "name": "writes",
+        "kind": "own",
+        "type": "string",
+        "required": "xor",
+        "desc": "Multi-region write as a JSON array (up to 100 items), each `{sheet_name|sheet_id, range, cells}` — the sheet selector LIVES IN EACH ITEM (same convention as +batch-update sub-ops and +styles-put items; the top-level --sheet-name is rejected). cells has the same shape as `--cells` (2D array; per-cell cell_styles/border_styles allowed). The whole array goes out as ONE atomic batched request, cross-sheet supported; typical use: fixing formulas scattered across ranges/sheets — do not assemble a +batch-update operations array for this. XOR with `--range`+`--cells`; range-level uniform styling stays with +styles-put afterwards",
+        "input": [
+          "file",
+          "stdin"
+        ]
+      },
       {
         "name": "allow-overwrite",
         "kind": "own",
@@ -2787,6 +2844,43 @@
       }
     ]
   },
+  "+styles-put": {
+    "risk": "write",
+    "flags": [
+      {
+        "name": "url",
+        "kind": "public",
+        "type": "string",
+        "required": "xor",
+        "desc": "Spreadsheet locator (target sheets are named inside --styles items)"
+      },
+      {
+        "name": "spreadsheet-token",
+        "kind": "public",
+        "type": "string",
+        "required": "xor",
+        "desc": "Spreadsheet token (XOR with `--url`)"
+      },
+      {
+        "name": "styles",
+        "kind": "own",
+        "type": "string",
+        "required": "required",
+        "desc": "Visual spec JSON applied to an EXISTING spreadsheet: top-level `{styles:[...]}`, one item per target sheet (`name` is the real sheet name), each giving at least one of `cell_styles` / `cell_merges` / `row_sizes` / `col_sizes` / `freeze`. The vocabulary is identical to `--styles` on `+workbook-create` / `+table-put` (cell_styles = A1 range + flat style fields, borders via the `border` shorthand {style,weight,color} applied to all four sides — border_styles only for per-side differences; row/col sizes = row/column range + size in px — type only for standard/auto; merges = cell range; freeze = `{rows:N, cols:N}`). The whole spec expands into one atomic batched request; ranges may target any region of the sheet",
+        "input": [
+          "file",
+          "stdin"
+        ]
+      },
+      {
+        "name": "dry-run",
+        "kind": "system",
+        "type": "bool",
+        "required": "optional",
+        "desc": "Print the batched request template for each expanded operation; no network side effects"
+      }
+    ]
+  },
   "+batch-update": {
     "risk": "high-risk-write",
     "flags": [
@@ -2809,7 +2903,7 @@
         "kind": "own",
         "type": "string",
         "required": "required",
-        "desc": "JSON array: [{\"shortcut\":\"+xxx-yyy\",\"input\":{...}}, ...]. shortcut uses CLI names; input is that shortcut's flag set — it includes the per-operation sheet locator (sheet_id or sheet_name) but not the spreadsheet token/url (pass that once at the top level via --url/--spreadsheet-token; +batch-update has no top-level --sheet-id). input keys are the shortcut's flags flattened into JSON (e.g. \"range\":\"A11:B12\"), not another nested layer. For basic flags use lark-cli sheets <shortcut> --help; for composite JSON flags use --print-schema --flag-name <flag>. Do not pass an explicit operation field. Strict transaction by default, pass --continue-on-error for soft batch; no nesting; executed serially.",
+        "desc": "JSON array: [{\"shortcut\":\"+xxx-yyy\",\"input\":{...}}, ...]. shortcut uses CLI names; input is that shortcut's flag set — it includes the per-operation sheet locator (sheet_id or sheet_name) but not the spreadsheet token/url (pass that once at the top level via --url/--spreadsheet-token; +batch-update has no top-level --sheet-id). input keys are the shortcut's flags flattened into JSON (e.g. \"range\":\"A11:B12\"), not another nested layer. For basic flags use lark-cli sheets <shortcut> --help; for composite JSON flags use --print-schema --flag-name <flag>. Do not pass an explicit operation field. Fail-fast by default: the first failure aborts the remaining operations and already-applied sub-operations are NOT rolled back (on \"N succeeded, M failed\" resend only the failed tail, not the whole batch); pass --continue-on-error to keep going past failures; no nesting; executed serially.",
         "input": [
           "file",
           "stdin"

shortcuts/sheets/data/flag-schemas.json

@@ -648,6 +648,35 @@
             }
           }
         }
+      },
+      "writes": {
+        "type": "array",
+        "description": "多区域写入项数组（最多 100 项），整批单次原子提交；支持跨 sheet。",
+        "items": {
+          "type": "object",
+          "required": [
+            "range",
+            "cells"
+          ],
+          "properties": {
+            "sheet_id": {
+              "type": "string",
+              "description": "目标子表 reference_id；与 sheet_name 二选一，必须写在每一项里（不认顶层 sheet 定位）。"
+            },
+            "sheet_name": {
+              "type": "string",
+              "description": "目标子表名；与 sheet_id 二选一，必须写在每一项里。"
+            },
+            "range": {
+              "type": "string",
+              "description": "A1 矩形范围，行列维度必须与 cells 严格一致（同 --range）。"
+            },
+            "cells": {
+              "type": "array",
+              "description": "二维单元格数组，结构同 --cells（value / formula / cell_styles / border_styles 等，见 set_cell_range#/properties/cells）。"
+            }
+          }
+        }
       }
     },
     "+cells-set-style": {
@@ -7748,87 +7777,7 @@
         }
       }
     },
-    "+table-put": {
-      "sheets": {
-        "type": "array",
-        "minItems": 1,
-        "description": "一个或多个子表的 typed 数据，每个数组元素写入一张子表；支持多 DataFrame → 多子表一次写入。每个数组项的形状对齐 pandas `df.to_json(orient=\"split\")`：列名走 `columns`、二维取值走 `data`、每列的 pandas dtype 走 `dtypes`、可选的展示格式走 `formats`，并显式带上目标子表名 `name`。pandas 来源直接用 `scripts/sheets_df.py` 的 `df_to_sheet(df, name)` 生成一项，再把 list 包到 `{\"sheets\":[...]}`。",
-        "items": {
-          "type": "object",
-          "required": [
-            "name",
-            "columns",
-            "data"
-          ],
-          "properties": {
-            "name": {
-              "type": "string",
-              "description": "目标子表名。按名匹配已有子表；不存在则新建该子表。同一次调用内子表名不可重复。"
-            },
-            "start_cell": {
-              "type": "string",
-              "default": "A1",
-              "description": "写入起点单元格（A1 记法，如 \"B2\"），默认 \"A1\"。mode=append 时忽略其行号、仅沿用其列。"
-            },
-            "mode": {
-              "type": "string",
-              "enum": [
-                "overwrite",
-                "append"
-              ],
-              "default": "overwrite",
-              "description": "overwrite（默认）：从 start_cell 起写「表头 + 数据」块；append：把数据追加到子表已有数据下方（默认不重复表头）。"
-            },
-            "header": {
-              "type": "boolean",
-              "description": "是否写一行列名表头。省略时按 mode 取默认:overwrite→true、append→false（避免在已有表头下重复）；显式给值可覆盖。"
-            },
-            "allow_overwrite": {
-              "type": "boolean",
-              "default": true,
-              "description": "为 false 时，若写入会落在非空单元格则拒写以保护原数据（返回 partial_success）。默认 true。"
-            },
-            "columns": {
-              "type": "array",
-              "minItems": 1,
-              "description": "列名字符串数组，顺序与 `data` 中每行取值一一对应。同一子表内列名不可重复。",
-              "items": {
-                "type": "string"
-              }
-            },
-            "data": {
-              "type": "array",
-              "description": "数据行；每行是一个数组，长度必须等于 `columns` 数。元素按 `dtypes` 推得的列类型取值（date 列写 ISO yyyy-mm-dd 字符串、number 列写数值、bool 列写布尔、其余写文本），null 表示空单元格。",
-              "items": {
-                "type": "array",
-                "items": {
-                  "type": [
-                    "string",
-                    "number",
-                    "boolean",
-                    "null"
-                  ],
-                  "description": "单元格值：date→ISO yyyy-mm-dd 字符串；number→数值（json.Number 精度保留）；bool→布尔；string→文本；null→空单元格。"
-                }
-              }
-            },
-            "dtypes": {
-              "type": "object",
-              "description": "可选。列名 → pandas dtype 字符串的映射；缺失项默认按 object（string + 文本格式 `@`）处理，所以省略整段时整张表按文本写入（导入 CSV-shaped 数据的最简形态）。dtype 解析规则：`int*` / `uint*` / `Int*` / `UInt*` / `float*` / `Float*` / `complex*` → number（精度保留），`bool` / `boolean` → bool，`datetime64[ns]` / 含时区的 `datetime64[ns, UTC]` 等 → date（默认 `yyyy-mm-dd` 格式），`object` / `string` / `category` / 未识别 → string + 文本格式 `@`（数字样字符串如「00123」不会塌缩成数字）。",
-              "additionalProperties": {
-                "type": "string"
-              }
-            },
-            "formats": {
-              "type": "object",
-              "description": "可选。列名 → Excel number_format 字符串的映射，覆盖 dtype 自带的默认格式（金额 `#,##0.00`、百分比 `0.0%`、自定义日期 `yyyy-mm` 等）。percent 列的数值尺度由调用方负责（0.0469 配 `0.00%` 显示 4.69%）。",
-              "additionalProperties": {
-                "type": "string"
-              }
-            }
-          }
-        }
-      },
+    "+styles-put": {
       "styles": {
         "items": {
           "properties": {
@@ -7856,12 +7805,16 @@
               "type": "array"
             },
             "cell_styles": {
-              "description": "单元格样式操作数组；每项用 A1 单元格 range 指定范围，字段名与 +cells-set-style 对齐。",
+              "description": "单元格样式操作数组；每项用 A1 单元格 range 指定范围，字段名与 +cells-set-style 对齐。加边框优先用 border 简写；只有分侧不同样式才用 border_styles 完整形态。",
               "items": {
                 "properties": {
                   "background_color": {
                     "type": "string"
                   },
+                  "border": {
+                    "description": "边框简写（推荐）：{style, weight, color} 应用到四边（如 {\"style\":\"solid\",\"color\":\"#DDDDDD\"}）；也接受侧键形态 {top:{…},bottom:{…}}。分侧不同样式用 border_styles 完整形态。",
+                    "type": "object"
+                  },
                   "border_styles": {
                     "type": "object",
                     "description": "边框配置，结构同 +cells-set-style --border-styles。",
@@ -8055,7 +8008,7 @@
               "type": "array"
             },
             "col_sizes": {
-              "description": "列宽操作数组；range 使用列范围如 A:C，type 为 pixel/standard，pixel 需要 size。",
+              "description": "列宽操作数组；range 使用列范围如 A:C，给 size（px）即像素列宽（type 可省略）；type 为 standard 时不带 size。",
               "items": {
                 "properties": {
                   "range": {
@@ -8073,19 +8026,32 @@
                   }
                 },
                 "required": [
-                  "range",
-                  "type"
+                  "range"
                 ],
                 "type": "object"
               },
               "type": "array"
             },
+            "freeze": {
+              "description": "冻结行列：rows = 冻结前 N 行，cols = 冻结前 N 列（0 或省略 = 该维度不冻结）。",
+              "properties": {
+                "cols": {
+                  "minimum": 0,
+                  "type": "integer"
+                },
+                "rows": {
+                  "minimum": 0,
+                  "type": "integer"
+                }
+              },
+              "type": "object"
+            },
             "name": {
               "description": "子表名。--sheets 模式下必须与同位置 --sheets.sheets[].name 一致；--values 模式下建议写 Sheet1（其 name 会被忽略）。",
               "type": "string"
             },
             "row_sizes": {
-              "description": "行高操作数组；range 使用行范围如 1:3，type 为 pixel/standard/auto，pixel 需要 size。",
+              "description": "行高操作数组；range 使用行范围如 1:3，给 size（px）即像素行高（type 可省略）；type 为 standard/auto 时不带 size。",
               "items": {
                 "properties": {
                   "range": {
@@ -8104,8 +8070,395 @@
                   }
                 },
                 "required": [
-                  "range",
-                  "type"
+                  "range"
+                ],
+                "type": "object"
+              },
+              "type": "array"
+            }
+          },
+          "required": [
+            "name"
+          ],
+          "type": "object"
+        },
+        "type": "array"
+      }
+    },
+    "+table-put": {
+      "sheets": {
+        "type": "array",
+        "minItems": 1,
+        "description": "一个或多个子表的 typed 数据，每个数组元素写入一张子表；支持多 DataFrame → 多子表一次写入。每个数组项的形状对齐 pandas `df.to_json(orient=\"split\")`：列名走 `columns`、二维取值走 `data`、每列的 pandas dtype 走 `dtypes`、可选的展示格式走 `formats`，并显式带上目标子表名 `name`。pandas 来源直接用 `scripts/sheets_df.py` 的 `df_to_sheet(df, name)` 生成一项，再把 list 包到 `{\"sheets\":[...]}`。",
+        "items": {
+          "type": "object",
+          "required": [
+            "name",
+            "columns",
+            "data"
+          ],
+          "properties": {
+            "name": {
+              "type": "string",
+              "description": "目标子表名。按名匹配已有子表；不存在则新建该子表。同一次调用内子表名不可重复。"
+            },
+            "start_cell": {
+              "type": "string",
+              "default": "A1",
+              "description": "写入起点单元格（A1 记法，如 \"B2\"），默认 \"A1\"。mode=append 时忽略其行号、仅沿用其列。"
+            },
+            "mode": {
+              "type": "string",
+              "enum": [
+                "overwrite",
+                "append"
+              ],
+              "default": "overwrite",
+              "description": "overwrite（默认）：从 start_cell 起写「表头 + 数据」块；append：把数据追加到子表已有数据下方（默认不重复表头）。"
+            },
+            "header": {
+              "type": "boolean",
+              "description": "是否写一行列名表头。省略时按 mode 取默认:overwrite→true、append→false（避免在已有表头下重复）；显式给值可覆盖。"
+            },
+            "allow_overwrite": {
+              "type": "boolean",
+              "default": true,
+              "description": "为 false 时，若写入会落在非空单元格则拒写以保护原数据（返回 partial_success）。默认 true。"
+            },
+            "columns": {
+              "type": "array",
+              "minItems": 1,
+              "description": "列名字符串数组，顺序与 `data` 中每行取值一一对应。同一子表内列名不可重复。",
+              "items": {
+                "type": "string"
+              }
+            },
+            "data": {
+              "type": "array",
+              "description": "数据行；每行是一个数组，长度必须等于 `columns` 数。元素按 `dtypes` 推得的列类型取值（date 列写 ISO yyyy-mm-dd 字符串、number 列写数值、bool 列写布尔、其余写文本），null 表示空单元格。",
+              "items": {
+                "type": "array",
+                "items": {
+                  "type": [
+                    "string",
+                    "number",
+                    "boolean",
+                    "null"
+                  ],
+                  "description": "单元格值：date→ISO yyyy-mm-dd 字符串；number→数值（json.Number 精度保留）；bool→布尔；string→文本；null→空单元格。"
+                }
+              }
+            },
+            "dtypes": {
+              "type": "object",
+              "description": "可选。列名 → pandas dtype 字符串的映射；缺失项默认按 object（string + 文本格式 `@`）处理，所以省略整段时整张表按文本写入（导入 CSV-shaped 数据的最简形态）。dtype 解析规则：`int*` / `uint*` / `Int*` / `UInt*` / `float*` / `Float*` / `complex*` → number（精度保留），`bool` / `boolean` → bool，`datetime64[ns]` / 含时区的 `datetime64[ns, UTC]` 等 → date（默认 `yyyy-mm-dd` 格式），`object` / `string` / `category` / 未识别 → string + 文本格式 `@`（数字样字符串如「00123」不会塌缩成数字）。",
+              "additionalProperties": {
+                "type": "string"
+              }
+            },
+            "formats": {
+              "type": "object",
+              "description": "可选。列名 → Excel number_format 字符串的映射，覆盖 dtype 自带的默认格式（金额 `#,##0.00`、百分比 `0.0%`、自定义日期 `yyyy-mm` 等）。percent 列的数值尺度由调用方负责（0.0469 配 `0.00%` 显示 4.69%）。",
+              "additionalProperties": {
+                "type": "string"
+              }
+            }
+          }
+        }
+      },
+      "styles": {
+        "items": {
+          "properties": {
+            "cell_merges": {
+              "description": "单元格合并操作数组；range 使用 A1 单元格范围，merge_type 默认 all。",
+              "items": {
+                "properties": {
+                  "merge_type": {
+                    "enum": [
+                      "all",
+                      "rows",
+                      "columns"
+                    ],
+                    "type": "string"
+                  },
+                  "range": {
+                    "type": "string"
+                  }
+                },
+                "required": [
+                  "range"
+                ],
+                "type": "object"
+              },
+              "type": "array"
+            },
+            "cell_styles": {
+              "description": "单元格样式操作数组；每项用 A1 单元格 range 指定范围，字段名与 +cells-set-style 对齐。加边框优先用 border 简写；只有分侧不同样式才用 border_styles 完整形态。",
+              "items": {
+                "properties": {
+                  "background_color": {
+                    "type": "string"
+                  },
+                  "border": {
+                    "description": "边框简写（推荐）：{style, weight, color} 应用到四边（如 {\"style\":\"solid\",\"color\":\"#DDDDDD\"}）；也接受侧键形态 {top:{…},bottom:{…}}。分侧不同样式用 border_styles 完整形态。",
+                    "type": "object"
+                  },
+                  "border_styles": {
+                    "type": "object",
+                    "description": "边框配置，结构同 +cells-set-style --border-styles。",
+                    "properties": {
+                      "bottom": {
+                        "properties": {
+                          "color": {
+                            "description": "边框颜色（十六进制，例如 \"#000000\"）",
+                            "type": "string"
+                          },
+                          "style": {
+                            "description": "边框线型；传 \"none\" 表示清除该方向边框（无边框线）",
+                            "enum": [
+                              "solid",
+                              "dashed",
+                              "dotted",
+                              "double",
+                              "none"
+                            ],
+                            "type": "string"
+                          },
+                          "weight": {
+                            "description": "边框粗细/线宽",
+                            "enum": [
+                              "thin",
+                              "medium",
+                              "thick"
+                            ],
+                            "type": "string"
+                          }
+                        },
+                        "type": "object"
+                      },
+                      "left": {
+                        "properties": {
+                          "color": {
+                            "description": "边框颜色（十六进制，例如 \"#000000\"）",
+                            "type": "string"
+                          },
+                          "style": {
+                            "description": "边框线型；传 \"none\" 表示清除该方向边框（无边框线）",
+                            "enum": [
+                              "solid",
+                              "dashed",
+                              "dotted",
+                              "double",
+                              "none"
+                            ],
+                            "type": "string"
+                          },
+                          "weight": {
+                            "description": "边框粗细/线宽",
+                            "enum": [
+                              "thin",
+                              "medium",
+                              "thick"
+                            ],
+                            "type": "string"
+                          }
+                        },
+                        "type": "object"
+                      },
+                      "right": {
+                        "properties": {
+                          "color": {
+                            "description": "边框颜色（十六进制，例如 \"#000000\"）",
+                            "type": "string"
+                          },
+                          "style": {
+                            "description": "边框线型；传 \"none\" 表示清除该方向边框（无边框线）",
+                            "enum": [
+                              "solid",
+                              "dashed",
+                              "dotted",
+                              "double",
+                              "none"
+                            ],
+                            "type": "string"
+                          },
+                          "weight": {
+                            "description": "边框粗细/线宽",
+                            "enum": [
+                              "thin",
+                              "medium",
+                              "thick"
+                            ],
+                            "type": "string"
+                          }
+                        },
+                        "type": "object"
+                      },
+                      "top": {
+                        "properties": {
+                          "color": {
+                            "description": "边框颜色（十六进制，例如 \"#000000\"）",
+                            "type": "string"
+                          },
+                          "style": {
+                            "description": "边框线型；传 \"none\" 表示清除该方向边框（无边框线）",
+                            "enum": [
+                              "solid",
+                              "dashed",
+                              "dotted",
+                              "double",
+                              "none"
+                            ],
+                            "type": "string"
+                          },
+                          "weight": {
+                            "description": "边框粗细/线宽",
+                            "enum": [
+                              "thin",
+                              "medium",
+                              "thick"
+                            ],
+                            "type": "string"
+                          }
+                        },
+                        "type": "object"
+                      }
+                    }
+                  },
+                  "font_color": {
+                    "type": "string"
+                  },
+                  "font_family": {
+                    "type": "string"
+                  },
+                  "font_line": {
+                    "enum": [
+                      "none",
+                      "underline",
+                      "line-through"
+                    ],
+                    "type": "string"
+                  },
+                  "font_size": {
+                    "type": "number"
+                  },
+                  "font_style": {
+                    "enum": [
+                      "normal",
+                      "italic"
+                    ],
+                    "type": "string"
+                  },
+                  "font_weight": {
+                    "enum": [
+                      "normal",
+                      "bold"
+                    ],
+                    "type": "string"
+                  },
+                  "horizontal_alignment": {
+                    "enum": [
+                      "left",
+                      "center",
+                      "right"
+                    ],
+                    "type": "string"
+                  },
+                  "number_format": {
+                    "type": "string"
+                  },
+                  "range": {
+                    "description": "A1 单元格范围，必须落在该子表本次写入区域内；例如 A1:B1、B2。",
+                    "type": "string"
+                  },
+                  "vertical_alignment": {
+                    "enum": [
+                      "top",
+                      "middle",
+                      "bottom"
+                    ],
+                    "type": "string"
+                  },
+                  "word_wrap": {
+                    "enum": [
+                      "overflow",
+                      "auto-wrap",
+                      "word-clip"
+                    ],
+                    "type": "string"
+                  }
+                },
+                "required": [
+                  "range"
+                ],
+                "type": "object"
+              },
+              "type": "array"
+            },
+            "col_sizes": {
+              "description": "列宽操作数组；range 使用列范围如 A:C，给 size（px）即像素列宽（type 可省略）；type 为 standard 时不带 size。",
+              "items": {
+                "properties": {
+                  "range": {
+                    "type": "string"
+                  },
+                  "size": {
+                    "type": "number"
+                  },
+                  "type": {
+                    "enum": [
+                      "pixel",
+                      "standard"
+                    ],
+                    "type": "string"
+                  }
+                },
+                "required": [
+                  "range"
+                ],
+                "type": "object"
+              },
+              "type": "array"
+            },
+            "freeze": {
+              "description": "冻结行列：rows = 冻结前 N 行，cols = 冻结前 N 列（0 或省略 = 该维度不冻结）。",
+              "properties": {
+                "cols": {
+                  "minimum": 0,
+                  "type": "integer"
+                },
+                "rows": {
+                  "minimum": 0,
+                  "type": "integer"
+                }
+              },
+              "type": "object"
+            },
+            "name": {
+              "description": "子表名。--sheets 模式下必须与同位置 --sheets.sheets[].name 一致；--values 模式下建议写 Sheet1（其 name 会被忽略）。",
+              "type": "string"
+            },
+            "row_sizes": {
+              "description": "行高操作数组；range 使用行范围如 1:3，给 size（px）即像素行高（type 可省略）；type 为 standard/auto 时不带 size。",
+              "items": {
+                "properties": {
+                  "range": {
+                    "type": "string"
+                  },
+                  "size": {
+                    "type": "number"
+                  },
+                  "type": {
+                    "enum": [
+                      "pixel",
+                      "standard",
+                      "auto"
+                    ],
+                    "type": "string"
+                  }
+                },
+                "required": [
+                  "range"
                 ],
                 "type": "object"
               },
@@ -8228,12 +8581,16 @@
               "type": "array"
             },
             "cell_styles": {
-              "description": "单元格样式操作数组；每项用 A1 单元格 range 指定范围，字段名与 +cells-set-style 对齐。",
+              "description": "单元格样式操作数组；每项用 A1 单元格 range 指定范围，字段名与 +cells-set-style 对齐。加边框优先用 border 简写；只有分侧不同样式才用 border_styles 完整形态。",
               "items": {
                 "properties": {
                   "background_color": {
                     "type": "string"
                   },
+                  "border": {
+                    "description": "边框简写（推荐）：{style, weight, color} 应用到四边（如 {\"style\":\"solid\",\"color\":\"#DDDDDD\"}）；也接受侧键形态 {top:{…},bottom:{…}}。分侧不同样式用 border_styles 完整形态。",
+                    "type": "object"
+                  },
                   "border_styles": {
                     "type": "object",
                     "description": "边框配置，结构同 +cells-set-style --border-styles。",
@@ -8427,7 +8784,7 @@
               "type": "array"
             },
             "col_sizes": {
-              "description": "列宽操作数组；range 使用列范围如 A:C，type 为 pixel/standard，pixel 需要 size。",
+              "description": "列宽操作数组；range 使用列范围如 A:C，给 size（px）即像素列宽（type 可省略）；type 为 standard 时不带 size。",
               "items": {
                 "properties": {
                   "range": {
@@ -8445,19 +8802,32 @@
                   }
                 },
                 "required": [
-                  "range",
-                  "type"
+                  "range"
                 ],
                 "type": "object"
               },
               "type": "array"
             },
+            "freeze": {
+              "description": "冻结行列：rows = 冻结前 N 行，cols = 冻结前 N 列（0 或省略 = 该维度不冻结）。",
+              "properties": {
+                "cols": {
+                  "minimum": 0,
+                  "type": "integer"
+                },
+                "rows": {
+                  "minimum": 0,
+                  "type": "integer"
+                }
+              },
+              "type": "object"
+            },
             "name": {
               "description": "子表名。--sheets 模式下必须与同位置 --sheets.sheets[].name 一致；--values 模式下建议写 Sheet1（其 name 会被忽略）。",
               "type": "string"
             },
             "row_sizes": {
-              "description": "行高操作数组；range 使用行范围如 1:3，type 为 pixel/standard/auto，pixel 需要 size。",
+              "description": "行高操作数组；range 使用行范围如 1:3，给 size（px）即像素行高（type 可省略）；type 为 standard/auto 时不带 size。",
               "items": {
                 "properties": {
                   "range": {
@@ -8476,8 +8846,7 @@
                   }
                 },
                 "required": [
-                  "range",
-                  "type"
+                  "range"
                 ],
                 "type": "object"
               },

shortcuts/sheets/flag_defs_gen.go

@@ -16,7 +16,7 @@ var flagDefs = map[string]commandDef{
 		Flags: []flagDef{
 			{Name: "url", Kind: "public", Type: "string", Required: "xor", Desc: "Spreadsheet locator (independent from per-operation sheet locator)"},
 			{Name: "spreadsheet-token", Kind: "public", Type: "string", Required: "xor", Desc: "Spreadsheet locator (independent from per-operation sheet locator)"},
-			{Name: "operations", Kind: "own", Type: "string", Required: "required", Desc: "JSON array: [{\"shortcut\":\"+xxx-yyy\",\"input\":{...}}, ...]. shortcut uses CLI names; input is that shortcut's flag set — it includes the per-operation sheet locator (sheet_id or sheet_name) but not the spreadsheet token/url (pass that once at the top level via --url/--spreadsheet-token; +batch-update has no top-level --sheet-id). input keys are the shortcut's flags flattened into JSON (e.g. \"range\":\"A11:B12\"), not another nested layer. For basic flags use lark-cli sheets <shortcut> --help; for composite JSON flags use --print-schema --flag-name <flag>. Do not pass an explicit operation field. Strict transaction by default, pass --continue-on-error for soft batch; no nesting; executed serially.", Input: []string{"file", "stdin"}},
+			{Name: "operations", Kind: "own", Type: "string", Required: "required", Desc: "JSON array: [{\"shortcut\":\"+xxx-yyy\",\"input\":{...}}, ...]. shortcut uses CLI names; input is that shortcut's flag set — it includes the per-operation sheet locator (sheet_id or sheet_name) but not the spreadsheet token/url (pass that once at the top level via --url/--spreadsheet-token; +batch-update has no top-level --sheet-id). input keys are the shortcut's flags flattened into JSON (e.g. \"range\":\"A11:B12\"), not another nested layer. For basic flags use lark-cli sheets <shortcut> --help; for composite JSON flags use --print-schema --flag-name <flag>. Do not pass an explicit operation field. Fail-fast by default: the first failure aborts the remaining operations and already-applied sub-operations are NOT rolled back (on \"N succeeded, M failed\" resend only the failed tail, not the whole batch); pass --continue-on-error to keep going past failures; no nesting; executed serially.", Input: []string{"file", "stdin"}},
 			{Name: "continue-on-error", Kind: "own", Type: "bool", Required: "optional", Desc: "Continue with remaining operations when a sub-operation fails; default false (abort on first failure)"},
 			{Name: "yes", Kind: "system", Type: "bool", Required: "required", Desc: "Confirm high-risk write (exit code 10 without this flag)"},
 			{Name: "dry-run", Kind: "system", Type: "bool", Required: "optional", Desc: "Print the request template for each sub-operation; no network side effects"},
@@ -75,9 +75,11 @@ var flagDefs = map[string]commandDef{
 			{Name: "sheet-id", Kind: "public", Type: "string", Required: "xor", Desc: "Sheet reference_id (XOR with `--sheet-name`)"},
 			{Name: "sheet-name", Kind: "public", Type: "string", Required: "xor", Desc: "Sheet name (XOR with `--sheet-id`)"},
 			{Name: "range", Kind: "own", Type: "string", Required: "required", Desc: "A1 range, e.g. `A1:F10` (no sheet prefix — use `--sheet-id` / `--sheet-name` to select the sheet)"},
-			{Name: "include", Kind: "own", Type: "string_slice", Required: "optional", Desc: "Comma-separated info categories to include", Enum: []string{"value", "formula", "style", "comment", "data_validation"}},
-			{Name: "max-chars", Kind: "own", Type: "int", Required: "optional", Desc: "Max output chars per call; default 500000 (safety cap). Large reads are usually better redirected to a file; only lower it (e.g. 25000) when you want results inline without triggering file offload, paging via has_more", Default: "500000"},
-			{Name: "skip-hidden", Kind: "own", Type: "bool", Required: "optional", Desc: "Skip hidden rows and columns; default `false`"},
+			{Name: "include", Kind: "own", Type: "string_slice", Required: "optional", Desc: "Comma-separated info categories to include. `truncation` additionally estimates whether each cell's content is clipped (by row height / col width / font size / wrap) and returns `isRowTruncated` / `isColTruncated` (extra compute; enable only for layout checks or before adjusting row heights / column widths)", Enum: []string{"value", "formula", "style", "comment", "data_validation", "truncation"}},
+			{Name: "max-chars", Kind: "own", Type: "int", Required: "optional", Desc: "Max output chars per call; default 500000 (safety cap). For a full untruncated read, use --output-path to dump to a file (auto-unlimited); only lower it (e.g. 25000) when you want results inline without a file, paging via has_more.", Default: "500000"},
+			{Name: "output-path", Kind: "own", Type: "string", Required: "optional", Desc: "Write the full read result to a local path (e.g. `./out.json`); the file holds the data payload as JSON while stdout returns only a small confirmation (output_path, byte count). **When set, the char cap defaults to unlimited** (overriding the --max-chars default), so a large sheet can be dumped in full for later analysis without stdout being truncated by max_chars. Omit it to print to stdout as usual."},
+			{Name: "skip-hidden", Kind: "own", Type: "bool", Required: "optional", Desc: "Skip hidden or collapsed rows and columns. Default `false`; when `--skip-filter` is omitted, filtered-out rows follow this value for backward compatibility"},
+			{Name: "skip-filter", Kind: "own", Type: "bool", Required: "optional", Desc: "Skip filtered-out rows. When omitted, inherits `--skip-hidden`; explicitly set to `false` to keep filtered-out rows while skipping hidden rows and columns"},
 			{Name: "dry-run", Kind: "system", Type: "bool", Required: "optional"},
 		},
 	},
@@ -133,10 +135,11 @@ var flagDefs = map[string]commandDef{
 		Flags: []flagDef{
 			{Name: "url", Kind: "public", Type: "string", Required: "xor", Desc: "Spreadsheet URL (XOR with `--spreadsheet-token`)"},
 			{Name: "spreadsheet-token", Kind: "public", Type: "string", Required: "xor", Desc: "Spreadsheet token (XOR with `--url`)"},
-			{Name: "sheet-id", Kind: "public", Type: "string", Required: "xor", Desc: "Sheet reference_id (XOR with `--sheet-name`)"},
-			{Name: "sheet-name", Kind: "public", Type: "string", Required: "xor", Desc: "Sheet name (XOR with `--sheet-id`)"},
-			{Name: "range", Kind: "own", Type: "string", Required: "required", Desc: "Write range (A1 notation)"},
-			{Name: "cells", Kind: "own", Type: "string", Required: "required", Desc: "JSON 2D array `[[{cell},...],...]`, dimensions must match `--range`; each cell may carry `value` / `formula` / `cell_styles` / `note` / `rich_text` (incl. `type=\"embed-image\"` in-cell image); run `--print-schema` for full fields", Input: []string{"file", "stdin"}},
+			{Name: "sheet-id", Kind: "public", Type: "string", Required: "xor", Desc: "Sheet reference_id (XOR with `--sheet-name`); not accepted with `--writes` (each writes item carries its own sheet selector)"},
+			{Name: "sheet-name", Kind: "public", Type: "string", Required: "xor", Desc: "Sheet name (XOR with `--sheet-id`); not accepted with `--writes` (each writes item carries its own sheet selector)"},
+			{Name: "range", Kind: "own", Type: "string", Required: "xor", Desc: "Write range (A1 notation). XOR with `--writes` (single region: --range+--cells; multiple regions: --writes)"},
+			{Name: "cells", Kind: "own", Type: "string", Required: "xor", Desc: "JSON 2D array `[[{cell},...],...]`, dimensions must match `--range`; each cell may carry `value` / `formula` / `cell_styles` / `note` / `rich_text` (incl. `type=\"embed-image\"` in-cell image); run `--print-schema` for full fields", Input: []string{"file", "stdin"}},
+			{Name: "writes", Kind: "own", Type: "string", Required: "xor", Desc: "Multi-region write as a JSON array (up to 100 items), each `{sheet_name|sheet_id, range, cells}` — the sheet selector LIVES IN EACH ITEM (same convention as +batch-update sub-ops and +styles-put items; the top-level --sheet-name is rejected). cells has the same shape as `--cells` (2D array; per-cell cell_styles/border_styles allowed). The whole array goes out as ONE atomic batched request, cross-sheet supported; typical use: fixing formulas scattered across ranges/sheets — do not assemble a +batch-update operations array for this. XOR with `--range`+`--cells`; range-level uniform styling stays with +styles-put afterwards", Input: []string{"file", "stdin"}},
 			{Name: "allow-overwrite", Kind: "own", Type: "bool", Required: "optional", Desc: "Allow overwriting non-empty cells (default true); set false to error if any target cell is non-empty", Default: "true"},
 			{Name: "max-cells", Kind: "own", Type: "int", Required: "optional", Desc: "Safety cap; default 50000", Default: "50000", Hidden: true},
 			{Name: "copy-to-range", Kind: "own", Type: "string", Required: "optional", Desc: "Copy-to range (A1 notation): replicate what --cells wrote into --range (values/formulas/styles, per the fields actually passed) to this range; formula refs auto-shift (C2=B2 -> C3=B3). Write a one-row/one-block template then fill a whole column/area. Supports full rows '3:6', full columns 'C:E', to-col-end 'D3:D', to-row-end 'D3:3', and comma-separated multiple targets like 'C1:D2,E5:F6'."},
@@ -316,8 +319,9 @@ var flagDefs = map[string]commandDef{
 			{Name: "spreadsheet-token", Kind: "public", Type: "string", Required: "xor", Desc: "Spreadsheet token (XOR with `--url`)"},
 			{Name: "sheet-id", Kind: "public", Type: "string", Required: "xor", Desc: "Sheet reference_id (XOR with `--sheet-name`)"},
 			{Name: "sheet-name", Kind: "public", Type: "string", Required: "xor", Desc: "Sheet name (XOR with `--sheet-id`)"},
-			{Name: "range", Kind: "own", Type: "string", Required: "required", Desc: "A1 range, e.g. `A1:F30` (no sheet prefix — use `--sheet-id` / `--sheet-name` to select the sheet)"},
-			{Name: "max-chars", Kind: "own", Type: "int", Required: "optional", Desc: "Max output chars per call; default 500000 (safety cap). Large reads are usually better redirected to a file; only lower it (e.g. 25000) when you want results inline without triggering file offload, paging via has_more", Default: "500000"},
+			{Name: "range", Kind: "own", Type: "string", Required: "optional", Desc: "A1 range, e.g. `A1:F30` (no sheet prefix — use `--sheet-id` / `--sheet-name` to select the sheet). Optional: when omitted the whole sheet is read (clipped to the actual grid bounds; actual_range in the response names what was read); pair with --max-chars / --output-path on large sheets"},
+			{Name: "max-chars", Kind: "own", Type: "int", Required: "optional", Desc: "Max output chars per call; default 500000 (safety cap). For a full untruncated read, use --output-path to dump to a file (auto-unlimited); only lower it (e.g. 25000) when you want results inline without a file, paging via has_more.", Default: "500000"},
+			{Name: "output-path", Kind: "own", Type: "string", Required: "optional", Desc: "Write the full read result to a local path (e.g. `./out.json`); the file holds the data payload as JSON while stdout returns only a small confirmation (output_path, byte count). **When set, the char cap defaults to unlimited** (overriding the --max-chars default), so a large sheet can be dumped in full for later analysis without stdout being truncated by max_chars. Omit it to print to stdout as usual."},
 			{Name: "include-row-prefix", Kind: "own", Type: "bool", Required: "optional", Desc: "Whether to prefix each row with `[row=N]`; default `true`", Default: "true"},
 			{Name: "skip-hidden", Kind: "own", Type: "bool", Required: "optional", Desc: "Skip hidden rows and columns; default `false`"},
 			{Name: "dry-run", Kind: "system", Type: "bool", Required: "optional", Desc: "Print the request path and parameters without executing"},
@@ -344,7 +348,8 @@ var flagDefs = map[string]commandDef{
 			{Name: "spreadsheet-token", Kind: "public", Type: "string", Required: "xor", Desc: "Spreadsheet token (XOR with `--url`)"},
 			{Name: "sheet-id", Kind: "public", Type: "string", Required: "xor", Desc: "Sheet reference_id (XOR with `--sheet-name`)"},
 			{Name: "sheet-name", Kind: "public", Type: "string", Required: "xor", Desc: "Sheet name (XOR with `--sheet-id`)"},
-			{Name: "range", Kind: "own", Type: "string", Required: "required", Desc: "Row/column closed range to delete; rows use 1-based numbers like `3:7` or `5` (single row), columns use letters like `C:F` or `C`"},
+			{Name: "range", Kind: "own", Type: "string", Required: "xor", Desc: "Row/column closed range to delete; rows use 1-based numbers like `3:7` or `5` (single row), columns use letters like `C:F` or `C`. XOR with `--ranges`"},
+			{Name: "ranges", Kind: "own", Type: "string", Required: "xor", Desc: "Multiple row/column ranges to delete as a JSON array (up to 100 items, e.g. `[\"5:5\",\"8:8\",\"11:13\"]` or `[\"C:C\",\"F:G\"]`); rows and columns cannot be mixed, ranges must not overlap; XOR with `--range`. CLI sorts positions in DESCENDING order into one atomic batch delete — ascending deletion would shift later indexes as earlier rows/columns disappear; the CLI handles the ordering", Input: []string{"file", "stdin"}},
 			{Name: "yes", Kind: "system", Type: "bool", Required: "required", Desc: "Confirm destructive write (exit code 10 without this flag); row/column deletion is irreversible"},
 			{Name: "dry-run", Kind: "system", Type: "bool", Required: "optional"},
 		},
@@ -392,7 +397,7 @@ var flagDefs = map[string]commandDef{
 			{Name: "spreadsheet-token", Kind: "public", Type: "string", Required: "xor", Desc: "Spreadsheet token (XOR with `--url`)"},
 			{Name: "sheet-id", Kind: "public", Type: "string", Required: "xor", Desc: "Sheet reference_id (XOR with `--sheet-name`)"},
 			{Name: "sheet-name", Kind: "public", Type: "string", Required: "xor", Desc: "Sheet name (XOR with `--sheet-id`)"},
-			{Name: "inherit-style", Kind: "own", Type: "string", Required: "optional", Desc: "Style inheritance for the new row/column: `before` (from preceding) / `after` (from following) / `none` (default)", Default: "none", Enum: []string{"before", "after", "none"}},
+			{Name: "inherit-style", Kind: "own", Type: "string", Required: "optional", Desc: "Style inheritance for the new row/column: `before` (from the preceding row/column) / `after` (from the following row/column). Omit the flag to inherit the following row/column (same as `after`) — the backend cannot leave a new row/column unstyled; for a truly blank row/column, clear formats afterwards with +cells-clear --scope formats. Insertion always lands before `--position`; this only selects which side's style is copied.", Enum: []string{"before", "after"}},
 			{Name: "position", Kind: "own", Type: "string", Required: "required", Desc: "Insert position (1-based row number like `3` or column letter like `C`); new rows/columns are inserted *before* this position"},
 			{Name: "count", Kind: "own", Type: "int", Required: "required", Desc: "Number of rows/columns to insert (must be > 0)"},
 			{Name: "dry-run", Kind: "system", Type: "bool", Required: "optional"},
@@ -975,6 +980,15 @@ var flagDefs = map[string]commandDef{
 			{Name: "dry-run", Kind: "system", Type: "bool", Required: "optional"},
 		},
 	},
+	"+styles-put": {
+		Risk: "write",
+		Flags: []flagDef{
+			{Name: "url", Kind: "public", Type: "string", Required: "xor", Desc: "Spreadsheet locator (target sheets are named inside --styles items)"},
+			{Name: "spreadsheet-token", Kind: "public", Type: "string", Required: "xor", Desc: "Spreadsheet token (XOR with `--url`)"},
+			{Name: "styles", Kind: "own", Type: "string", Required: "required", Desc: "Visual spec JSON applied to an EXISTING spreadsheet: top-level `{styles:[...]}`, one item per target sheet (`name` is the real sheet name), each giving at least one of `cell_styles` / `cell_merges` / `row_sizes` / `col_sizes` / `freeze`. The vocabulary is identical to `--styles` on `+workbook-create` / `+table-put` (cell_styles = A1 range + flat style fields, borders via the `border` shorthand {style,weight,color} applied to all four sides — border_styles only for per-side differences; row/col sizes = row/column range + size in px — type only for standard/auto; merges = cell range; freeze = `{rows:N, cols:N}`). The whole spec expands into one atomic batched request; ranges may target any region of the sheet", Input: []string{"file", "stdin"}},
+			{Name: "dry-run", Kind: "system", Type: "bool", Required: "optional", Desc: "Print the batched request template for each expanded operation; no network side effects"},
+		},
+	},
 	"+table-get": {
 		Risk: "read",
 		Flags: []flagDef{
@@ -983,6 +997,8 @@ var flagDefs = map[string]commandDef{
 			{Name: "sheet-id", Kind: "own", Type: "string", Required: "optional", Desc: "Read only this sheet (by id); omit to read all sheets"},
 			{Name: "sheet-name", Kind: "own", Type: "string", Required: "optional", Desc: "Read only this sheet (by name); omit to read all sheets"},
 			{Name: "range", Kind: "own", Type: "string", Required: "optional", Desc: "A1 range to read; omit to read each sheet's full used range (spans internal blank rows/columns, not just the A1 current region)"},
+			{Name: "max-chars", Kind: "own", Type: "int", Required: "optional", Desc: "Max output chars per call; default 500000 (safety cap). The underlying tool truncates at ~50000 even when unset, so this is sent explicitly to raise it; for a full untruncated read use --output-path (auto-unlimited).", Default: "500000"},
+			{Name: "output-path", Kind: "own", Type: "string", Required: "optional", Desc: "Write the full read result to a local path (e.g. `./out.json`); the file holds the data payload as JSON while stdout returns only a small confirmation (output_path, byte count). **When set, the char cap defaults to unlimited** (overriding the --max-chars default), so a large sheet can be dumped in full for later analysis without stdout being truncated by max_chars. Omit it to print to stdout as usual."},
 			{Name: "no-header", Kind: "own", Type: "bool", Required: "optional", Desc: "Treat the first row as data instead of a header (columns get positional names col1, col2, ...)"},
 			{Name: "dry-run", Kind: "system", Type: "bool", Required: "optional"},
 		},

shortcuts/sheets/flag_defs_test.go

@@ -52,7 +52,7 @@ func TestFlagsFor_MapsAllFields(t *testing.T) {
 
 	// enum + default
 	rt := byName("+dim-insert", "inherit-style")
-	if rt == nil || len(rt.Enum) != 3 || rt.Default != "none" {
+	if rt == nil || len(rt.Enum) != 2 || rt.Default != "" {
 		t.Errorf("+dim-insert --inherit-style not mapped: %+v", rt)
 	}
 	// required

shortcuts/sheets/flag_ergonomics.go

@@ -38,9 +38,104 @@ func withFlagErgonomics(prev func(cmd *cobra.Command)) func(cmd *cobra.Command)
 		}
 		cmd.SetFlagErrorFunc(sheetsFlagErrorFunc)
 		chainEnumNormalization(cmd)
+		chainFlagAliases(cmd)
 	}
 }
 
+// ─── intuitive flag names: silent aliases & prescriptions ───────────────
+//
+// Eval traces show unknown-flag failures cluster on a handful of habitual
+// names (--file, --cols, --dimension, --start-cell, --bold, --source…) that
+// agents import from generic CLI / Excel vocabulary. Two tiers, mirroring
+// the enum-normalization contract above: a name whose value semantics are
+// identical to the real flag is rewritten silently (zero round-trips); a
+// name whose fix changes the value or moves it into a JSON field gets a
+// curated prescription on the unknown-flag error instead — never a silent
+// rewrite.
+
+// commandFlagAliases maps, per command, habitual flag names onto the flag
+// actually registered. Only pairs with identical value semantics belong
+// here: the rewrite is invisible, so it must be safe to apply unread
+// (+csv-put --file with a path value still trips the file-path guard, which
+// prescribes @file / stdin).
+var commandFlagAliases = map[string]map[string]string{
+	"+csv-put":      {"file": "csv"},
+	"+sheet-create": {"name": "title"},
+	// size → width/height: the styles protocol (--styles row_sizes/col_sizes)
+	// spells the pixel dimension "size", and pre-2026-07 batches accepted it
+	// here too — the rename is the single largest sub-op error cluster in
+	// eval traces (15+ hits). Same pixel-count semantics, safe to rewrite.
+	"+cols-resize": {"cols": "range", "size": "width"},
+	"+rows-resize": {"rows": "range", "size": "height"},
+	"+range-fill":  {"source": "source-range", "target": "target-range"},
+	"+range-copy":  {"source": "source-range", "target": "target-range"},
+	"+range-move":  {"source": "source-range", "target": "target-range"},
+}
+
+// intuitiveFlagHints carries the prescription for habitual names whose fix
+// is not a 1:1 rename — the value belongs to a different flag or to a field
+// inside a JSON payload. The hint spells the exact correct form so the
+// retry needs no --help round trip.
+var intuitiveFlagHints = map[string]map[string]string{
+	"+sheet-copy": {
+		"new-sheet-name":    "the copy's name goes in --title; --sheet-name / --sheet-id selects the source sheet",
+		"target-sheet-name": "the copy's name goes in --title; --sheet-name / --sheet-id selects the source sheet",
+		"new-name":          "the copy's name goes in --title; --sheet-name / --sheet-id selects the source sheet",
+	},
+	"+dim-insert": {
+		"dimension": "+dim-insert infers rows vs columns from --position: a row number like 3 inserts rows, a column letter like C inserts columns; pair with --count N",
+	},
+	"+dim-freeze": {
+		"frozen-rows":    "freeze the first N rows with --dimension row --count N",
+		"frozen-cols":    "freeze the first N columns with --dimension column --count N",
+		"frozen-columns": "freeze the first N columns with --dimension column --count N",
+	},
+	"+cells-set-style": {
+		"bold":      "use --font-weight bold",
+		"italic":    "use --font-style italic",
+		"underline": "use --font-line underline",
+	},
+	"+cells-set": {
+		// Predictable prior from +table-put --styles: models will try to
+		// attach range-level styling to a --writes call the same way.
+		"styles": `range-level styling goes through +styles-put (same {"styles":[...]} vocabulary); per-cell styles ride inside the cells objects as cell_styles`,
+	},
+	"+table-put": {
+		"start-cell": `anchor each sub-sheet via the "start_cell" field inside --sheets (e.g. {"sheets":[{"name":"Sheet1","start_cell":"B2",…}]}); to paste CSV at a cell use +csv-put --start-cell`,
+		"sheet-name": `+table-put has no sheet selector — each --sheets item carries its own "name" field ({"sheets":[{"name":"Sheet1",…}]})`,
+		"sheet-id":   `+table-put has no sheet selector — each --sheets item carries its own "name" field ({"sheets":[{"name":"Sheet1",…}]})`,
+	},
+}
+
+// chainFlagAliases composes two rewrites onto the flag-name normalize hook
+// (on top of any hook a prior PostMount installed, e.g. --token →
+// --spreadsheet-token): the wire-vocabulary underscore form of any flag
+// (--sheet_name, --border_styles — no sheets flag has an underscore in its
+// canonical name), and the command's intuitive-alias table. Either way a
+// habitual name parses as the real flag with zero round trips. Aliases
+// never shadow a registered flag and never appear in --help; an alias whose
+// target vanished (spec-side rename) is dropped, degrading to the
+// unknown-flag prescription.
+func chainFlagAliases(cmd *cobra.Command) {
+	aliases := commandFlagAliases[cmd.Name()]
+	usable := make(map[string]string, len(aliases))
+	for alias, target := range aliases {
+		if cmd.Flags().Lookup(alias) == nil && cmd.Flags().Lookup(target) != nil {
+			usable[alias] = target
+		}
+	}
+	prev := cmd.Flags().GetNormalizeFunc()
+	cmd.Flags().SetNormalizeFunc(func(fs *pflag.FlagSet, name string) pflag.NormalizedName {
+		if strings.Contains(name, "_") {
+			name = strings.ReplaceAll(name, "_", "-")
+		}
+		if target, ok := usable[name]; ok {
+			name = target
+		}
+		return prev(fs, name)
+	})
+}
+
 // sheetsFlagErrorFunc overrides the root FlagErrorFunc for sheets commands.
 // It keeps the root behavior (typed error, did-you-mean suggestions, the
 // offending flag on params) and additionally inlines the full valid-flag
@@ -50,6 +145,19 @@ func withFlagErgonomics(prev func(cmd *cobra.Command)) func(cmd *cobra.Command)
 // immediately.
 func sheetsFlagErrorFunc(c *cobra.Command, ferr error) error {
 	name, isUnknown := unknownFlagFromParseError(ferr)
+	// Targeted fix for a high-frequency agent mistake: +batch-update carries no
+	// top-level sheet locator (each sub-op names its own sheet inside its input),
+	// yet agents reach for --sheet-id / --sheet-name at the top level. An
+	// edit-distance suggestion would only mislead here, so skip it and name the
+	// real contract instead. Underscore spellings (--sheet_id) are matched too:
+	// the error message itself teaches the underscore key names, and sub-op
+	// inputs accept them, so agents mix the two styles.
+	locatorName := strings.ReplaceAll(name, "_", "-")
+	if isUnknown && c.Name() == "+batch-update" && (locatorName == "sheet-id" || locatorName == "sheet-name") {
+		return errs.NewValidationError(errs.SubtypeInvalidArgument,
+			"batch-update has no top-level sheet locator; put sheet_id/sheet_name inside each operation's input").
+			WithParams(errs.InvalidParam{Name: "--" + name, Reason: "unknown flag"})
+	}
 	if !isUnknown {
 		return common.ValidationErrorf("%s", ferr.Error()).
 			WithHint("run `%s --help` for valid flags", c.CommandPath())
@@ -67,6 +175,14 @@ func sheetsFlagErrorFunc(c *cobra.Command, ferr error) error {
 				strings.Join(suggestions, ", "), list)
 		}
 	}
+	// A curated prescription beats both: it spells the exact correct form
+	// for a habitual name whose fix is not a rename (see intuitiveFlagHints).
+	if rx, ok := intuitiveFlagHints[c.Name()][name]; ok {
+		hint = rx
+		if list := inlineFlagList(valid); list != "" {
+			hint = rx + "; valid flags: " + list
+		}
+	}
 	return errs.NewValidationError(errs.SubtypeInvalidArgument,
 		"unknown flag %q for %q", "--"+name, c.CommandPath()).
 		WithParams(errs.InvalidParam{Name: "--" + name, Reason: "unknown flag", Suggestions: suggestions}).
@@ -139,6 +255,16 @@ var enumAliases = map[string]string{
 	"center": "middle", // CSS vertical-align: center → Lark "middle"
 	"centre": "center",
 	"middle": "center", // CSS-style middle → Lark horizontal "center"
+	// Raw Lark OpenAPI merge vocabulary (MERGE_ALL/…) — agents reproduce it
+	// from the API docs; lowercased by canonicalEnumValue before lookup.
+	"merge_all":     "all",
+	"merge_rows":    "rows",
+	"merge_columns": "columns",
+	// Boolean-style word-wrap habits: true unambiguously means wrap on;
+	// false means "don't wrap", whose Lark default is overflow (word-clip is
+	// a distinct truncation mode nobody spells "false").
+	"true":  "auto-wrap",
+	"false": "overflow",
 }
 
 // canonicalEnumValue returns the enum entry an off-vocabulary value

shortcuts/sheets/flag_ergonomics_test.go

@@ -96,6 +96,55 @@ func TestSheetsFlagErrorFunc_TypoKeepsSuggestion(t *testing.T) {
 	}
 }
 
+// TestSheetsFlagErrorFunc_BatchUpdateSheetLocator pins the targeted fix: a
+// top-level --sheet-id / --sheet-name on +batch-update points the caller at
+// the per-op locator contract instead of offering a misleading fuzzy guess.
+func TestSheetsFlagErrorFunc_BatchUpdateSheetLocator(t *testing.T) {
+	t.Parallel()
+	for _, name := range []string{"sheet-id", "sheet-name", "sheet_id", "sheet_name"} {
+		name := name
+		t.Run(name, func(t *testing.T) {
+			t.Parallel()
+			c := &cobra.Command{Use: "+batch-update"}
+			c.Flags().String("operations", "", "")
+			err := sheetsFlagErrorFunc(c, errors.New("unknown flag: --"+name))
+			var verr *errs.ValidationError
+			if !errors.As(err, &verr) {
+				t.Fatalf("expected *errs.ValidationError, got %T", err)
+			}
+			if !strings.Contains(verr.Message, "put sheet_id/sheet_name inside each operation's input") {
+				t.Errorf("message should name the per-op locator contract, got %q", verr.Message)
+			}
+			if strings.Contains(verr.Hint, "did you mean") {
+				t.Errorf("must not offer a fuzzy guess here, got hint %q", verr.Hint)
+			}
+			if len(verr.Params) != 1 || verr.Params[0].Name != "--"+name {
+				t.Errorf("Params should carry the offending flag, got %v", verr.Params)
+			}
+			if len(verr.Params[0].Suggestions) != 0 {
+				t.Errorf("no suggestions expected, got %v", verr.Params[0].Suggestions)
+			}
+		})
+	}
+}
+
+// TestSheetsFlagErrorFunc_BatchUpdateOtherUnknownStillSuggests confirms the
+// special case is scoped to the two sheet-locator flags: any other unknown
+// flag on +batch-update keeps the normal did-you-mean behaviour.
+func TestSheetsFlagErrorFunc_BatchUpdateOtherUnknownStillSuggests(t *testing.T) {
+	t.Parallel()
+	c := &cobra.Command{Use: "+batch-update"}
+	c.Flags().String("operations", "", "")
+	err := sheetsFlagErrorFunc(c, errors.New("unknown flag: --operation"))
+	var verr *errs.ValidationError
+	if !errors.As(err, &verr) {
+		t.Fatalf("expected *errs.ValidationError, got %T", err)
+	}
+	if strings.Contains(verr.Message, "no top-level sheet locator") {
+		t.Errorf("non-locator unknown flag must not hit the special case, got %q", verr.Message)
+	}
+}
+
 func TestSheetsFlagErrorFunc_OtherErrorStaysGeneric(t *testing.T) {
 	t.Parallel()
 	c := &cobra.Command{Use: "demo"}
@@ -284,9 +333,9 @@ func TestShortcuts_FlagErgonomicsMounted(t *testing.T) {
 		_, _, err := runShortcutCapturingErr(t, sc, []string{
 			"--url", testURL,
 			"--sheet-name", "s",
-			"--cols", "A:D",
+			"--col-size", "A:D",
 		})
-		ve := requireValidation(t, err, `unknown flag "--cols"`)
+		ve := requireValidation(t, err, `unknown flag "--col-size"`)
 		for _, want := range []string{"valid flags:", "--range", "--width", "--widths"} {
 			if !strings.Contains(ve.Hint, want) {
 				t.Errorf("hint should contain %q, got %q", want, ve.Hint)
@@ -294,3 +343,191 @@ func TestShortcuts_FlagErgonomicsMounted(t *testing.T) {
 		}
 	})
 }
+
+// TestShortcuts_IntuitiveFlagAliases verifies the silent-alias tier: a
+// habitual name with identical value semantics parses as the real flag on a
+// mounted command, costing zero round trips (eval: --cols, --file, --name,
+// --source/--target each burned an unknown-flag failure plus a --help call).
+func TestShortcuts_IntuitiveFlagAliases(t *testing.T) {
+	t.Parallel()
+
+	t.Run("cols-resize --cols parses as --range", func(t *testing.T) {
+		t.Parallel()
+		sc := shortcutFromRegistry(t, "+cols-resize")
+		stdout, _, err := runShortcutCapturingErr(t, sc, []string{
+			"--url", testURL,
+			"--sheet-name", "s",
+			"--cols", "A:D",
+			"--width", "100",
+			"--dry-run",
+		})
+		if err != nil {
+			t.Fatalf("--cols should alias to --range and pass, got: %v", err)
+		}
+		if !strings.Contains(stdout, "A:D") {
+			t.Errorf("dry-run body should carry the aliased range, got %q", stdout)
+		}
+	})
+
+	t.Run("sheet-create --name parses as --title", func(t *testing.T) {
+		t.Parallel()
+		sc := shortcutFromRegistry(t, "+sheet-create")
+		stdout, _, err := runShortcutCapturingErr(t, sc, []string{
+			"--url", testURL,
+			"--name", "汇总",
+			"--dry-run",
+		})
+		if err != nil {
+			t.Fatalf("--name should alias to --title and pass, got: %v", err)
+		}
+		if !strings.Contains(stdout, "汇总") {
+			t.Errorf("dry-run body should carry the aliased title, got %q", stdout)
+		}
+	})
+
+	t.Run("range-fill --source/--target parse as ranges", func(t *testing.T) {
+		t.Parallel()
+		sc := shortcutFromRegistry(t, "+range-fill")
+		stdout, _, err := runShortcutCapturingErr(t, sc, []string{
+			"--url", testURL,
+			"--sheet-name", "s",
+			"--source", "B2",
+			"--target", "B3:B10",
+			"--dry-run",
+		})
+		if err != nil {
+			t.Fatalf("--source/--target should alias to the -range flags, got: %v", err)
+		}
+		for _, want := range []string{"B2", "B3:B10"} {
+			if !strings.Contains(stdout, want) {
+				t.Errorf("dry-run body should carry %q, got %q", want, stdout)
+			}
+		}
+	})
+
+	t.Run("csv-put --file parses as --csv", func(t *testing.T) {
+		t.Parallel()
+		sc := shortcutFromRegistry(t, "+csv-put")
+		stdout, _, err := runShortcutCapturingErr(t, sc, []string{
+			"--url", testURL,
+			"--sheet-name", "s",
+			"--start-cell", "A1",
+			"--file", "a,b\n1,2",
+			"--dry-run",
+		})
+		if err != nil {
+			t.Fatalf("--file with CSV text should alias to --csv and pass, got: %v", err)
+		}
+		if !strings.Contains(stdout, "a,b") {
+			t.Errorf("dry-run body should carry the CSV text, got %q", stdout)
+		}
+	})
+
+	t.Run("cols-resize --size parses as --width", func(t *testing.T) {
+		t.Parallel()
+		sc := shortcutFromRegistry(t, "+cols-resize")
+		stdout, _, err := runShortcutCapturingErr(t, sc, []string{
+			"--url", testURL,
+			"--sheet-name", "s",
+			"--range", "A:C",
+			"--size", "120",
+			"--dry-run",
+		})
+		if err != nil {
+			t.Fatalf("--size should alias to --width (styles-protocol vocabulary), got: %v", err)
+		}
+		if !strings.Contains(stdout, "120") {
+			t.Errorf("dry-run body should carry the pixel width 120, got %q", stdout)
+		}
+	})
+
+	t.Run("rows-resize --size parses as --height", func(t *testing.T) {
+		t.Parallel()
+		sc := shortcutFromRegistry(t, "+rows-resize")
+		_, _, err := runShortcutCapturingErr(t, sc, []string{
+			"--url", testURL,
+			"--sheet-name", "s",
+			"--range", "1:3",
+			"--size", "36",
+			"--dry-run",
+		})
+		if err != nil {
+			t.Fatalf("--size should alias to --height (styles-protocol vocabulary), got: %v", err)
+		}
+	})
+
+	t.Run("alias never shadows a registered flag", func(t *testing.T) {
+		t.Parallel()
+		c := &cobra.Command{Use: "+csv-put"}
+		c.Flags().String("csv", "", "")
+		c.Flags().String("file", "", "") // hypothetical real flag wins
+		chainFlagAliases(c)
+		if err := c.ParseFlags([]string{"--file", "x"}); err != nil {
+			t.Fatalf("parse: %v", err)
+		}
+		if got, _ := c.Flags().GetString("file"); got != "x" {
+			t.Errorf("registered --file should keep its own value, got %q", got)
+		}
+		if got, _ := c.Flags().GetString("csv"); got != "" {
+			t.Errorf("--csv must stay empty when --file is a real flag, got %q", got)
+		}
+	})
+}
+
+// TestShortcuts_IntuitiveFlagHints verifies the prescription tier: habitual
+// names whose fix is not a rename answer with the exact correct form, so the
+// retry needs no --help round trip (eval: +sheet-copy burned 3/3 post-error
+// --help calls, +dim-insert kept failing even after reading help).
+func TestShortcuts_IntuitiveFlagHints(t *testing.T) {
+	t.Parallel()
+	cases := []struct {
+		command  string
+		args     []string
+		wrong    string
+		wantHint []string
+	}{
+		{
+			command:  "+dim-insert",
+			args:     []string{"--url", testURL, "--sheet-name", "s", "--dimension", "row"},
+			wrong:    "--dimension",
+			wantHint: []string{"--position", "--count"},
+		},
+		{
+			command:  "+dim-freeze",
+			args:     []string{"--url", testURL, "--sheet-name", "s", "--frozen-rows", "2"},
+			wrong:    "--frozen-rows",
+			wantHint: []string{"--dimension row --count N"},
+		},
+		{
+			command:  "+cells-set-style",
+			args:     []string{"--url", testURL, "--sheet-name", "s", "--range", "A1", "--bold", "true"},
+			wrong:    "--bold",
+			wantHint: []string{"--font-weight bold"},
+		},
+		{
+			command:  "+sheet-copy",
+			args:     []string{"--url", testURL, "--sheet-name", "s", "--new-sheet-name", "副本"},
+			wrong:    "--new-sheet-name",
+			wantHint: []string{"--title", "source sheet"},
+		},
+		{
+			command:  "+table-put",
+			args:     []string{"--url", testURL, "--sheets", "{}", "--start-cell", "B2"},
+			wrong:    "--start-cell",
+			wantHint: []string{`"start_cell"`, "+csv-put"},
+		},
+	}
+	for _, tc := range cases {
+		t.Run(tc.command+" "+tc.wrong, func(t *testing.T) {
+			t.Parallel()
+			sc := shortcutFromRegistry(t, tc.command)
+			_, _, err := runShortcutCapturingErr(t, sc, tc.args)
+			ve := requireValidation(t, err, "unknown flag \""+tc.wrong+"\"")
+			for _, want := range tc.wantHint {
+				if !strings.Contains(ve.Hint, want) {
+					t.Errorf("hint should contain %q, got %q", want, ve.Hint)
+				}
+			}
+		})
+	}
+}

shortcuts/sheets/flag_schema.go

@@ -7,6 +7,7 @@ import (
 	_ "embed"
 	"encoding/json"
 	"sort"
+	"strings"
 	"sync"
 
 	"github.com/larksuite/cli/errs"
@@ -84,6 +85,13 @@ func commandsWithFlagSchema() map[string]struct{} {
 // listing of introspectable flags; otherwise it returns the schema
 // subtree JSON for the named flag, or an error if the flag is not
 // registered.
+//
+// flagName also accepts a dotted path (properties.plotArea.axes): the
+// first segment names the flag, the rest walk the schema's properties
+// (descending through array items implicitly), returning just that
+// subtree. Large schemas — chart-create's properties is ~1,750 pretty
+// lines — otherwise force agents to page through the full dump for one
+// nested field; eval traces show 25 such round trips in one batch.
 func printFlagSchemaFor(command string) func(flagName string) ([]byte, error) {
 	return func(flagName string) ([]byte, error) {
 		idx, err := loadFlagSchemas()
@@ -103,10 +111,19 @@ func printFlagSchemaFor(command string) func(flagName string) ([]byte, error) {
 			return json.MarshalIndent(map[string]interface{}{
 				"shortcut":             command,
 				"introspectable_flags": flags,
-				"hint":                 "run again with --flag-name <name> to dump the JSON Schema for that flag",
+				"hint":                 "run again with --flag-name <name> to dump that flag's JSON Schema, or a dotted path like <name>.plotArea.axes to dump just one subtree",
 			}, "", "  ")
 		}
-		schema, ok := entry[flagName]
+		name, path := splitSchemaPath(flagName)
+		schema, ok := entry[name]
+		if !ok {
+			// Tolerate the wire-vocabulary underscore form (--flag-name
+			// border_styles for border-styles) — agents copy field names out
+			// of JSON payloads where underscores are canonical.
+			if alt := strings.ReplaceAll(name, "_", "-"); alt != name {
+				schema, ok = entry[alt]
+			}
+		}
 		if !ok {
 			flags := make([]string, 0, len(entry))
 			for f := range entry {
@@ -114,14 +131,121 @@ func printFlagSchemaFor(command string) func(flagName string) ([]byte, error) {
 			}
 			sort.Strings(flags)
 			return nil, errs.NewValidationError(errs.SubtypeInvalidArgument,
-				"no JSON Schema registered for %s --%s; available: %v", command, flagName, flags).
+				"no JSON Schema registered for %s --%s; available: %v", command, name, flags).
 				WithParam("--flag-name")
 		}
-		// Reformat for readability — schema files store compact JSON.
 		var pretty interface{}
 		if err := json.Unmarshal(schema, &pretty); err != nil {
 			return nil, err
 		}
+		if len(path) > 0 {
+			pretty, err = sliceSchemaByPath(pretty, name, path)
+			if err != nil {
+				return nil, err
+			}
+		}
+		// Reformat for readability — schema files store compact JSON.
 		return json.MarshalIndent(pretty, "", "  ")
 	}
 }
+
+// splitSchemaPath splits a --flag-name value into the flag name and the
+// optional dotted schema path after it.
+func splitSchemaPath(flagName string) (string, []string) {
+	parts := strings.Split(flagName, ".")
+	return parts[0], parts[1:]
+}
+
+// sliceSchemaByPath walks a decoded JSON Schema along dotted path segments.
+// Each segment matches a key under "properties"; array levels are descended
+// implicitly through "items" (an explicit "items" segment also works), and
+// oneOf branches are searched for the first one carrying the key. A miss
+// errors with the keys actually available at that level so the caller can
+// re-issue the path without a full dump.
+func sliceSchemaByPath(schema interface{}, flagName string, path []string) (interface{}, error) {
+	node := schema
+	walked := flagName
+	for _, seg := range path {
+		next, ok := schemaChild(node, seg)
+		if !ok {
+			return nil, errs.NewValidationError(errs.SubtypeInvalidArgument,
+				"no %q under %s; available keys: %v", seg, walked, schemaChildKeys(node)).
+				WithParam("--flag-name")
+		}
+		node = next
+		walked += "." + seg
+	}
+	return node, nil
+}
+
+// schemaChild resolves one path segment against a schema node, descending
+// through items / oneOf wrappers as needed.
+func schemaChild(node interface{}, seg string) (interface{}, bool) {
+	for depth := 0; depth < 8; depth++ {
+		m, ok := node.(map[string]interface{})
+		if !ok {
+			return nil, false
+		}
+		if seg == "items" {
+			if items, ok := m["items"]; ok {
+				return items, true
+			}
+		}
+		if props, ok := m["properties"].(map[string]interface{}); ok {
+			if child, ok := props[seg]; ok {
+				return child, true
+			}
+		}
+		if items, ok := m["items"]; ok {
+			node = items
+			continue
+		}
+		if branches, ok := m["oneOf"].([]interface{}); ok {
+			for _, b := range branches {
+				if child, ok := schemaChild(b, seg); ok {
+					return child, true
+				}
+			}
+		}
+		return nil, false
+	}
+	return nil, false
+}
+
+// schemaChildKeys lists the property keys reachable at a schema node (through
+// items / oneOf wrappers), for the path-miss error.
+func schemaChildKeys(node interface{}) []string {
+	seen := map[string]struct{}{}
+	var collect func(n interface{}, depth int)
+	collect = func(n interface{}, depth int) {
+		if depth > 8 {
+			return
+		}
+		m, ok := n.(map[string]interface{})
+		if !ok {
+			return
+		}
+		if props, ok := m["properties"].(map[string]interface{}); ok {
+			for k := range props {
+				seen[k] = struct{}{}
+			}
+			return
+		}
+		if items, ok := m["items"]; ok {
+			collect(items, depth+1)
+			return
+		}
+		if branches, ok := m["oneOf"].([]interface{}); ok {
+			for _, b := range branches {
+				collect(b, depth+1)
+			}
+		}
+	}
+	collect(node, 0)
+	keys := make([]string, 0, len(seen))
+	for k := range seen {
+		keys = append(keys, k)
+	}
+	sort.Strings(keys)
+	return keys
+}

shortcuts/sheets/flag_schema_validate.go

@@ -9,6 +9,8 @@ import (
 	"fmt"
 	"sort"
 	"strings"
+
+	"github.com/larksuite/cli/internal/suggest"
 )
 
 // ─── schema-driven flag validation ────────────────────────────────────
@@ -94,7 +96,15 @@ func validateValueAgainstSchema(fv flagView, name string, value interface{}) err
 	}
 	var schema schemaProperty
 	json.Unmarshal(raw, &schema)
-	if vErr := validateAgainstSchema(value, &schema, ""); vErr != nil {
+	c := &schemaErrorCollector{}
+	collectSchemaErrors(value, &schema, "", c)
+	if len(c.errs) == 0 {
+		return nil
+	}
+	vErr := c.errs[0]
+	if len(c.errs) == 1 {
+		// Single failure keeps the historical message byte-for-byte.
+		//
 		// Composite-JSON shape errors (e.g. +cells-set --cells, chart
 		// --properties) are the highest-frequency usage-layer failure for
 		// sheets, and agents often burn several retries guessing the shape.
@@ -107,18 +117,63 @@ func validateValueAgainstSchema(fv flagView, name string, value interface{}) err
 		// branch means entry[name] resolved a schema from the embedded
 		// index, so the suggested command is guaranteed to print it.
 		var tm *typeMismatchError
-		if errors.As(vErr, &tm) && pathDepth(tm.path) <= skeletonPathDepthLimit {
+		isTypeMismatch := errors.As(vErr, &tm)
+		if isTypeMismatch && pathDepth(tm.path) <= skeletonPathDepthLimit {
 			if sk := schemaSkeleton(&schema, skeletonMaxDepth); sk != "" {
 				return sheetsValidationForFlag(name,
 					"--%s: %s; expected shape: %s (run `lark-cli sheets %s --print-schema --flag-name %s` for the full JSON Schema)",
 					name, vErr.Error(), sk, command, name).WithCause(vErr)
 			}
 		}
+		// Deep type mismatches don't get a whole-shape skeleton (it wouldn't
+		// address the actual field), but if the field itself carries an enum /
+		// description, append that one line — same "fix on first retry" goal.
+		msg := vErr.Error()
+		if isTypeMismatch {
+			if suffix := tm.hintSuffix(); suffix != "" {
+				msg += "; " + suffix
+			}
+		}
 		return sheetsValidationForFlag(name,
 			"--%s: %s; run `lark-cli sheets %s --print-schema --flag-name %s` to see the expected JSON Schema",
-			name, vErr.Error(), command, name).WithCause(vErr)
+			name, msg, command, name).WithCause(vErr)
 	}
-	return nil
+	// Multiple failures: report them all at once (numbered, each with its
+	// own inline teaching hint) so the agent fixes the whole payload in one
+	// retry instead of the fail-fast "fix one, hit the next" loop.
+	return sheetsValidationForFlag(name,
+		"--%s: %s; run `lark-cli sheets %s --print-schema --flag-name %s` to see the expected JSON Schema",
+		name, formatSchemaErrorList(c.errs), command, name).WithCause(vErr)
+}
+
+// formatSchemaErrorList renders collected failures as a numbered one-line
+// list: "N validation errors: 1) …; 2) …". Type-mismatch entries carry
+// their enum/description suffix just like the single-error path. Entries
+// beyond schemaErrorDisplayLimit collapse into a "(more …)" tail — the
+// collector stops at cap, so the exact total is unknown by design.
+func formatSchemaErrorList(errs []error) string {
+	shown := errs
+	truncated := false
+	if len(shown) > schemaErrorDisplayLimit {
+		shown = shown[:schemaErrorDisplayLimit]
+		truncated = true
+	}
+	parts := make([]string, 0, len(shown))
+	for i, e := range shown {
+		msg := e.Error()
+		var tm *typeMismatchError
+		if errors.As(e, &tm) {
+			if suffix := tm.hintSuffix(); suffix != "" {
+				msg += "; " + suffix
+			}
+		}
+		parts = append(parts, fmt.Sprintf("%d) %s", i+1, msg))
+	}
+	out := fmt.Sprintf("%d validation errors: %s", len(shown), strings.Join(parts, "; "))
+	if truncated {
+		out = fmt.Sprintf("%d+ validation errors: %s; (more errors not shown — fix these first)", schemaErrorDisplayLimit, strings.Join(parts, "; "))
+	}
+	return out
 }
 
 // validateInputAgainstSchema validates input[flag] for every flag the
@@ -187,8 +242,10 @@ var inputSchemaSkip = map[string]struct{}{
 }
 
 // schemaProperty mirrors the JSON Schema subset used by
-// data/flag-schemas.json. Unknown keys (description, …) are dropped —
-// they're documentation.
+// data/flag-schemas.json. Description is retained (not just documentation)
+// so a required-missing or type-mismatch error can inline the one-line
+// field doc — the agent then fixes the input without a --print-schema round
+// trip. Other unknown keys stay dropped.
 //
 // Minimum / Maximum / MinItems / MaxItems use *float64 / *int because
 // 0 is a meaningful bound (e.g. chart row >= 0); nil distinguishes
@@ -204,6 +261,7 @@ var inputSchemaSkip = map[string]struct{}{
 //     map<string, array<string>> fields (groups / collapse).
 type schemaProperty struct {
 	Type                 string                     `json:"type"`
+	Description          string                     `json:"description"`
 	Nullable             bool                       `json:"nullable"`
 	Enum                 []interface{}              `json:"enum"`
 	Properties           map[string]*schemaProperty `json:"properties"`
@@ -242,20 +300,66 @@ func (a *additionalProps) UnmarshalJSON(data []byte) error {
 	return nil
 }
 
+// schemaErrorCollector accumulates validation failures during one full
+// traversal so the caller can report every problem in a single reply
+// instead of the fail-fast "fix one, retry, hit the next" loop. Capacity
+// is bounded (collectSchemaErrorsCap) so a pathological payload — e.g. a
+// 5000-row --cells array where every cell is malformed — cannot balloon
+// the error message or the traversal cost: once full, collection
+// short-circuits everywhere via full().
+type schemaErrorCollector struct {
+	errs []error
+}
+
+// collectSchemaErrorsCap bounds how many errors one traversal gathers:
+// schemaErrorDisplayLimit entries are rendered; one extra is collected
+// only to know that truncation happened.
+const (
+	schemaErrorDisplayLimit = 5
+	collectSchemaErrorsCap  = schemaErrorDisplayLimit + 1
+)
+
+func (c *schemaErrorCollector) add(err error) {
+	if len(c.errs) < collectSchemaErrorsCap {
+		c.errs = append(c.errs, err)
+	}
+}
+
+func (c *schemaErrorCollector) full() bool { return len(c.errs) >= collectSchemaErrorsCap }
+
 // validateAgainstSchema recursively checks `value` against `schema`,
-// prefixing any failure with the JSON path navigated so far.
+// prefixing any failure with the JSON path navigated so far. It reports
+// only the first failure — callers that want the full list (the
+// error-as-teaching aggregate path) use collectSchemaErrors directly.
 func validateAgainstSchema(value interface{}, schema *schemaProperty, path string) error {
-	if schema == nil {
-		return nil // defensive — current callers always pass &schema, but
-		// keeps validator safe for future programmatic construction.
+	c := &schemaErrorCollector{}
+	collectSchemaErrors(value, schema, path, c)
+	if len(c.errs) == 0 {
+		return nil
+	}
+	return c.errs[0]
+}
+
+// collectSchemaErrors is the traversal engine behind validateAgainstSchema:
+// same checks, same messages, same deterministic order, but it keeps
+// walking after a failure and appends every problem to the collector
+// (until cap). Two deliberate exceptions to "keep walking":
+//   - a type mismatch stops descent into that node (its children would
+//     produce cascading nonsense against the wrong-typed value);
+//   - oneOf alternatives are probed with throwaway collectors (a failed
+//     alternative is not an error when a later one matches).
+func collectSchemaErrors(value interface{}, schema *schemaProperty, path string, c *schemaErrorCollector) {
+	if schema == nil || c.full() {
+		return
 	}
 	if value == nil && schema.Nullable {
-		return nil
+		return
 	}
 
 	if schema.Type != "" {
 		if !matchesJSONType(value, schema.Type) {
-			return &typeMismatchError{path: path, expected: schema.Type, got: jsType(value)}
+			c.add(&typeMismatchError{path: path, expected: schema.Type, got: jsType(value), enum: schema.Enum, description: schema.Description})
+			return // wrong container type — descending would cascade nonsense.
 		}
 	}
 
@@ -263,20 +367,20 @@ func validateAgainstSchema(value interface{}, schema *schemaProperty, path strin
 	// already reported above). Apply to both `number` and `integer` types.
 	if num, ok := value.(float64); ok {
 		if schema.Minimum != nil && num < *schema.Minimum {
-			return fmt.Errorf("%svalue %v is below minimum %v", pathPrefix(path), num, *schema.Minimum) //nolint:forbidigo // intermediate error; validateFlagAgainstSchema wraps it into a typed flag validation error with a --print-schema hint
+			c.add(fmt.Errorf("%svalue %v is below minimum %v", pathPrefix(path), num, *schema.Minimum)) //nolint:forbidigo // intermediate error; validateFlagAgainstSchema wraps it into a typed flag validation error with a --print-schema hint
 		}
 		if schema.Maximum != nil && num > *schema.Maximum {
-			return fmt.Errorf("%svalue %v is above maximum %v", pathPrefix(path), num, *schema.Maximum) //nolint:forbidigo // intermediate error; validateFlagAgainstSchema wraps it into a typed flag validation error with a --print-schema hint
+			c.add(fmt.Errorf("%svalue %v is above maximum %v", pathPrefix(path), num, *schema.Maximum)) //nolint:forbidigo // intermediate error; validateFlagAgainstSchema wraps it into a typed flag validation error with a --print-schema hint
 		}
 	}
 
 	// Array length bounds — only checked when value is an array.
 	if arr, ok := value.([]interface{}); ok {
 		if schema.MinItems != nil && len(arr) < *schema.MinItems {
-			return fmt.Errorf("%sarray has %d items, minimum is %d", pathPrefix(path), len(arr), *schema.MinItems) //nolint:forbidigo // intermediate error; validateFlagAgainstSchema wraps it into a typed flag validation error with a --print-schema hint
+			c.add(fmt.Errorf("%sarray has %d items, minimum is %d", pathPrefix(path), len(arr), *schema.MinItems)) //nolint:forbidigo // intermediate error; validateFlagAgainstSchema wraps it into a typed flag validation error with a --print-schema hint
 		}
 		if schema.MaxItems != nil && len(arr) > *schema.MaxItems {
-			return fmt.Errorf("%sarray has %d items, maximum is %d", pathPrefix(path), len(arr), *schema.MaxItems) //nolint:forbidigo // intermediate error; validateFlagAgainstSchema wraps it into a typed flag validation error with a --print-schema hint
+			c.add(fmt.Errorf("%sarray has %d items, maximum is %d", pathPrefix(path), len(arr), *schema.MaxItems)) //nolint:forbidigo // intermediate error; validateFlagAgainstSchema wraps it into a typed flag validation error with a --print-schema hint
 		}
 	}
 
@@ -294,20 +398,22 @@ func validateAgainstSchema(value interface{}, schema *schemaProperty, path strin
 			if hint := suggestEnumForError(value, schema.Enum); hint != "" {
 				msg += fmt.Sprintf(` (did you mean %q?)`, hint)
 			}
-			return fmt.Errorf("%s", msg) //nolint:forbidigo // intermediate error; validateFlagAgainstSchema wraps it into a typed flag validation error with a --print-schema hint
+			c.add(fmt.Errorf("%s", msg)) //nolint:forbidigo // intermediate error; validateFlagAgainstSchema wraps it into a typed flag validation error with a --print-schema hint
 		}
 	}
 
 	if len(schema.OneOf) > 0 {
 		matched := false
 		for _, sub := range schema.OneOf {
-			if validateAgainstSchema(value, sub, path) == nil {
+			probe := &schemaErrorCollector{}
+			collectSchemaErrors(value, sub, path, probe)
+			if len(probe.errs) == 0 {
 				matched = true
 				break
 			}
 		}
 		if !matched {
-			return fmt.Errorf("%svalue does not match any of oneOf alternatives", pathPrefix(path)) //nolint:forbidigo // intermediate error; validateFlagAgainstSchema wraps it into a typed flag validation error with a --print-schema hint
+			c.add(fmt.Errorf("%svalue does not match any of oneOf alternatives", pathPrefix(path))) //nolint:forbidigo // intermediate error; validateFlagAgainstSchema wraps it into a typed flag validation error with a --print-schema hint
 		}
 	}
 
@@ -316,8 +422,18 @@ func validateAgainstSchema(value interface{}, schema *schemaProperty, path strin
 	// the schema also describes their per-key shape via `properties`.
 	if obj, ok := value.(map[string]interface{}); ok {
 		for _, key := range schema.Required {
+			if c.full() {
+				return
+			}
 			if _, present := obj[key]; !present {
-				return fmt.Errorf("required property %q is missing at %s", key, pathOrRoot(path)) //nolint:forbidigo // intermediate error; validateFlagAgainstSchema wraps it into a typed flag validation error with a --print-schema hint
+				msg := fmt.Sprintf("required property %q is missing at %s", key, pathOrRoot(path))
+				// Inline the missing field's type / one-line description / enum so
+				// the agent supplies a correctly-shaped value on the first retry
+				// instead of fetching the full schema.
+				if hint := schemaFieldHint(schema.Properties[key]); hint != "" {
+					msg += "; expected " + hint
+				}
+				c.add(fmt.Errorf("%s", msg)) //nolint:forbidigo // intermediate error; validateFlagAgainstSchema wraps it into a typed flag validation error with a --print-schema hint
 			}
 		}
 		if schema.Properties != nil {
@@ -327,6 +443,9 @@ func validateAgainstSchema(value interface{}, schema *schemaProperty, path strin
 			}
 			sort.Strings(keys)
 			for _, key := range keys {
+				if c.full() {
+					return
+				}
 				sub := schema.Properties[key]
 				v, present := obj[key]
 				if !present {
@@ -350,14 +469,12 @@ func validateAgainstSchema(value interface{}, schema *schemaProperty, path strin
 				if path != "" {
 					child = path + "." + key
 				}
-				if err := validateAgainstSchema(v, sub, child); err != nil {
-					return err
-				}
+				collectSchemaErrors(v, sub, child, c)
 			}
 		}
 		// additionalProperties: enforce only when explicitly declared.
 		// Absent means lenient (matches the file header's stance). Sort
-		// extras so the first rejection is deterministic across runs.
+		// extras so rejection order is deterministic across runs.
 		if schema.AdditionalProperties != nil {
 			extras := make([]string, 0)
 			for key := range obj {
@@ -368,17 +485,29 @@ func validateAgainstSchema(value interface{}, schema *schemaProperty, path strin
 			}
 			sort.Strings(extras)
 			for _, key := range extras {
+				if c.full() {
+					return
+				}
 				if schema.AdditionalProperties.Strict {
-					return fmt.Errorf("%sunexpected property %q (not declared in schema)", pathPrefix(path), key) //nolint:forbidigo // intermediate error; validateFlagAgainstSchema wraps it into a typed flag validation error with a --print-schema hint
+					msg := fmt.Sprintf("%sunexpected property %q (not declared in schema)", pathPrefix(path), key)
+					// Inline the node's declared keys (and a did-you-mean when the
+					// unknown key is a near miss) so the agent renames it in one
+					// retry instead of a --print-schema round trip.
+					if legal := sortedSchemaPropertyKeys(schema.Properties); len(legal) > 0 {
+						if guess := suggest.Closest(key, legal, 1); len(guess) > 0 {
+							msg += fmt.Sprintf(` (did you mean %q?)`, guess[0])
+						}
+						msg += "; valid properties: " + formatPropertyKeyList(legal)
+					}
+					c.add(fmt.Errorf("%s", msg)) //nolint:forbidigo // intermediate error; validateFlagAgainstSchema wraps it into a typed flag validation error with a --print-schema hint
+					continue
 				}
 				if schema.AdditionalProperties.Schema != nil {
 					child := key
 					if path != "" {
 						child = path + "." + key
 					}
-					if err := validateAgainstSchema(obj[key], schema.AdditionalProperties.Schema, child); err != nil {
-						return err
-					}
+					collectSchemaErrors(obj[key], schema.AdditionalProperties.Schema, child, c)
 				}
 			}
 		}
@@ -387,33 +516,50 @@ func validateAgainstSchema(value interface{}, schema *schemaProperty, path strin
 	if schema.Type == "array" && schema.Items != nil {
 		arr, ok := value.([]interface{})
 		if !ok {
-			return nil // type mismatch already reported above.
+			return // type mismatch already reported above.
 		}
 		for i, item := range arr {
-			child := fmt.Sprintf("%s[%d]", path, i)
-			if err := validateAgainstSchema(item, schema.Items, child); err != nil {
-				return err
+			if c.full() {
+				return
 			}
+			child := fmt.Sprintf("%s[%d]", path, i)
+			collectSchemaErrors(item, schema.Items, child, c)
 		}
 	}
-
-	return nil
 }
 
 // typeMismatchError is the type-check branch of validateAgainstSchema
 // as a typed error, so validateValueAgainstSchema can recognize shape
 // confusion (vs. deep value errors) and inline a skeleton of the
-// expected shape. Error() keeps the exact legacy wording.
+// expected shape. Error() keeps the exact legacy wording; enum /
+// description ride alongside for the deep-mismatch hintSuffix, so they
+// never leak into the shallow-skeleton message.
 type typeMismatchError struct {
-	path     string
-	expected string
-	got      string
+	path        string
+	expected    string
+	got         string
+	enum        []interface{}
+	description string
 }
 
 func (e *typeMismatchError) Error() string {
 	return fmt.Sprintf("%sexpected type %q, got %q", pathPrefix(e.path), e.expected, e.got)
 }
 
+// hintSuffix renders the field's description / enum as a one-line tail for
+// the deep type-mismatch fallback (type is already stated by Error()).
+// Empty when the field declares neither.
+func (e *typeMismatchError) hintSuffix() string {
+	var parts []string
+	if d := oneLineDescription(e.description); d != "" {
+		parts = append(parts, "description: "+d)
+	}
+	if len(e.enum) > 0 {
+		parts = append(parts, "one of "+formatEnum(e.enum))
+	}
+	return strings.Join(parts, ", ")
+}
+
 // pathDepth counts how many levels below the flag root a JSON path
 // points at: "" → 0, "[0]" → 1, "[0][3]" → 2, "[0][3].value" → 3,
 // "legend" → 1, "snapshot.axes" → 2. Every "[" and "." starts a new
@@ -605,6 +751,70 @@ func joinFormatted(values []interface{}) string {
 	return strings.Join(parts, ", ")
 }
 
+// schemaFieldHint renders a compact one-line "type X, description: …, one of
+// […]" sketch of a single field's schema, used to enrich a required-missing
+// error so the agent supplies a correctly-shaped value without --print-schema.
+// Empty when the field declares none of type / description / enum.
+func schemaFieldHint(s *schemaProperty) string {
+	if s == nil {
+		return ""
+	}
+	var parts []string
+	if s.Type != "" {
+		parts = append(parts, fmt.Sprintf("type %q", s.Type))
+	}
+	if d := oneLineDescription(s.Description); d != "" {
+		parts = append(parts, "description: "+d)
+	}
+	if len(s.Enum) > 0 {
+		parts = append(parts, "one of "+formatEnum(s.Enum))
+	}
+	return strings.Join(parts, ", ")
+}
+
+// sortedSchemaPropertyKeys returns the declared property names in a stable
+// (sorted) order so the valid-property list in a strict unexpected-property
+// error is deterministic across runs.
+func sortedSchemaPropertyKeys(props map[string]*schemaProperty) []string {
+	keys := make([]string, 0, len(props))
+	for k := range props {
+		keys = append(keys, k)
+	}
+	sort.Strings(keys)
+	return keys
+}
+
+// propertyKeyDisplayLimit caps how many declared property names ride inline on
+// a strict unexpected-property error, so a wide object doesn't bury the actual
+// error under a wall of keys. Overflow is summarised as "(N more)".
+const propertyKeyDisplayLimit = 15
+
+func formatPropertyKeyList(keys []string) string {
+	if len(keys) <= propertyKeyDisplayLimit {
+		return "[" + strings.Join(keys, ", ") + "]"
+	}
+	shown := keys[:propertyKeyDisplayLimit]
+	return fmt.Sprintf("[%s, … (%d more)]", strings.Join(shown, ", "), len(keys)-propertyKeyDisplayLimit)
+}
+
+// descriptionMaxLen bounds an inlined field description to one reasonable line;
+// schema descriptions can run several sentences, which would swamp the error.
+const descriptionMaxLen = 120
+
+// oneLineDescription collapses a (possibly multi-line) schema description into
+// a single whitespace-normalised line, truncated to descriptionMaxLen runes.
+// Returns "" for an empty / whitespace-only description.
+func oneLineDescription(s string) string {
+	collapsed := strings.Join(strings.Fields(s), " ")
+	if collapsed == "" {
+		return ""
+	}
+	if r := []rune(collapsed); len(r) > descriptionMaxLen {
+		return string(r[:descriptionMaxLen]) + "…"
+	}
+	return collapsed
+}
+
 // suggestEnumMatch returns the canonical enum entry when the user's
 // value unambiguously means one — casing ("SUM" vs "sum", "True" vs
 // "true") or a cross-vocabulary alias (CSS "center" for Lark's vertical

shortcuts/sheets/flag_schema_validate_test.go

@@ -5,6 +5,8 @@ package sheets
 
 import (
 	"encoding/json"
+	"errors"
+	"fmt"
 	"strings"
 	"testing"
 )
@@ -438,6 +440,373 @@ func TestValidateValueAgainstSchema_ShapeSkeletonOnShallowTypeMismatch(t *testin
 	}
 }
 
+// TestValidateAgainstSchema_StrictUnexpectedPropertyListsKeys pins the strict
+// additionalProperties:false enhancement: the error lists the node's legal
+// property keys (sorted, capped at 15 with an "(N more)" overflow) and, when
+// the unknown key is a near miss, appends a did-you-mean.
+func TestValidateAgainstSchema_StrictUnexpectedPropertyListsKeys(t *testing.T) {
+	t.Parallel()
+
+	t.Run("lists legal keys and suggests a near miss", func(t *testing.T) {
+		t.Parallel()
+		schema := parseSchema(t, `{
+			"type":"object",
+			"additionalProperties":false,
+			"properties":{
+				"background_color":{"type":"string"},
+				"font_weight":{"type":"string"},
+				"font_size":{"type":"integer"}
+			}
+		}`)
+		err := validateAgainstSchema(map[string]interface{}{"background_colour": "#fff"}, schema, "")
+		if err == nil {
+			t.Fatal("unknown key under strict schema must fail")
+		}
+		msg := err.Error()
+		if !strings.Contains(msg, `unexpected property "background_colour"`) {
+			t.Errorf("want the offending key named; got %q", msg)
+		}
+		if !strings.Contains(msg, `did you mean "background_color"?`) {
+			t.Errorf("want a did-you-mean for the near miss; got %q", msg)
+		}
+		for _, want := range []string{"valid properties:", "background_color", "font_size", "font_weight"} {
+			if !strings.Contains(msg, want) {
+				t.Errorf("want valid-property list to contain %q; got %q", want, msg)
+			}
+		}
+	})
+
+	t.Run("no did-you-mean for an unrelated key", func(t *testing.T) {
+		t.Parallel()
+		schema := parseSchema(t, `{
+			"type":"object",
+			"additionalProperties":false,
+			"properties":{"background_color":{"type":"string"}}
+		}`)
+		err := validateAgainstSchema(map[string]interface{}{"zzzzzzzz": 1}, schema, "")
+		if err == nil {
+			t.Fatal("unknown key must fail")
+		}
+		if strings.Contains(err.Error(), "did you mean") {
+			t.Errorf("unrelated key should get no suggestion; got %q", err.Error())
+		}
+		if !strings.Contains(err.Error(), "valid properties: [background_color]") {
+			t.Errorf("want the valid-property list; got %q", err.Error())
+		}
+	})
+
+	t.Run("wide object truncates the key list with overflow", func(t *testing.T) {
+		t.Parallel()
+		props := make([]string, 0, 20)
+		for i := 0; i < 20; i++ {
+			props = append(props, fmt.Sprintf(`"k%02d":{"type":"string"}`, i))
+		}
+		schema := parseSchema(t, `{"type":"object","additionalProperties":false,"properties":{`+strings.Join(props, ",")+`}}`)
+		err := validateAgainstSchema(map[string]interface{}{"nope": 1}, schema, "")
+		if err == nil {
+			t.Fatal("unknown key must fail")
+		}
+		if !strings.Contains(err.Error(), "(5 more)") { // 20 keys, cap 15
+			t.Errorf("want overflow marker '(5 more)'; got %q", err.Error())
+		}
+	})
+}
+
+// TestValidateAgainstSchema_RequiredMissingInlinesFieldHint pins that a
+// required-property-missing error inlines the field's type / one-line
+// description / enum when the schema describes that field.
+func TestValidateAgainstSchema_RequiredMissingInlinesFieldHint(t *testing.T) {
+	t.Parallel()
+
+	schema := parseSchema(t, `{
+		"type":"object",
+		"required":["operation"],
+		"properties":{
+			"operation":{
+				"type":"string",
+				"description":"Which mutation to run.",
+				"enum":["insert","delete","move"]
+			}
+		}
+	}`)
+	err := validateAgainstSchema(map[string]interface{}{}, schema, "")
+	if err == nil {
+		t.Fatal("missing required property must fail")
+	}
+	msg := err.Error()
+	for _, want := range []string{
+		`required property "operation"`,
+		`type "string"`,
+		"description: Which mutation to run.",
+		`one of ["insert", "delete", "move"]`,
+	} {
+		if !strings.Contains(msg, want) {
+			t.Errorf("want %q in required-missing error; got %q", want, msg)
+		}
+	}
+}
+
+// TestValidateAgainstSchema_RequiredMissingNoSchemaStaysPlain pins that a
+// missing required key with no describing schema keeps the plain legacy
+// message (no trailing "expected ...").
+func TestValidateAgainstSchema_RequiredMissingNoSchemaStaysPlain(t *testing.T) {
+	t.Parallel()
+	schema := parseSchema(t, `{"type":"object","required":["a"]}`)
+	err := validateAgainstSchema(map[string]interface{}{}, schema, "")
+	if err == nil {
+		t.Fatal("missing required must fail")
+	}
+	if strings.Contains(err.Error(), "; expected") {
+		t.Errorf("no field schema → no inlined hint; got %q", err.Error())
+	}
+}
+
+// TestValidateValueAgainstSchema_DeepTypeMismatchAppendsEnum pins that a deep
+// type mismatch (past the skeleton depth limit) still gets no whole-shape
+// skeleton, but appends the field's enum / description one-liner.
+func TestValidateValueAgainstSchema_DeepTypeMismatchAppendsEnum(t *testing.T) {
+	t.Parallel()
+	// A wrong-typed value three levels deep where the field is an enum string.
+	schema := parseSchema(t, `{
+		"type":"array",
+		"items":{"type":"array","items":{"type":"object","properties":{
+			"align":{"type":"string","description":"Text alignment.","enum":["left","center","right"]}
+		}}}
+	}`)
+	deep := parseValue(t, `[[{"align":42}]]`)
+	err := validateAgainstSchema(deep, schema, "")
+	if err == nil {
+		t.Fatal("wrong type for align must fail")
+	}
+	var tm *typeMismatchError
+	if !errors.As(err, &tm) {
+		t.Fatalf("want *typeMismatchError, got %T", err)
+	}
+	suffix := tm.hintSuffix()
+	for _, want := range []string{"description: Text alignment.", `one of ["left", "center", "right"]`} {
+		if !strings.Contains(suffix, want) {
+			t.Errorf("want %q in hintSuffix; got %q", want, suffix)
+		}
+	}
+}
+
+// TestSchemaFieldHint covers the single-field sketch used by
+// required-missing errors: each of type / description / enum contributes
+// its own segment, absent parts are simply skipped, and a nil / empty
+// schema yields no hint at all.
+func TestSchemaFieldHint(t *testing.T) {
+	t.Parallel()
+	cases := []struct {
+		name   string
+		schema *schemaProperty
+		want   string
+	}{
+		{"nil schema", nil, ""},
+		{"empty schema", &schemaProperty{}, ""},
+		{"type only", &schemaProperty{Type: "string"}, `type "string"`},
+		{"description only", &schemaProperty{Description: "Cell note."}, "description: Cell note."},
+		{"enum only", &schemaProperty{Enum: []interface{}{"a", "b"}}, `one of ["a", "b"]`},
+		{
+			"all three",
+			&schemaProperty{Type: "string", Description: "段类型", Enum: []interface{}{"text", "link"}},
+			`type "string", description: 段类型, one of ["text", "link"]`,
+		},
+	}
+	for _, tc := range cases {
+		tc := tc
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			if got := schemaFieldHint(tc.schema); got != tc.want {
+				t.Errorf("schemaFieldHint = %q, want %q", got, tc.want)
+			}
+		})
+	}
+}
+
+// TestFormatPropertyKeyList_Boundaries pins the display cap edges: exactly
+// at the cap nothing is folded, one past the cap folds into "(1 more)".
+func TestFormatPropertyKeyList_Boundaries(t *testing.T) {
+	t.Parallel()
+	keys := make([]string, 0, propertyKeyDisplayLimit+1)
+	for i := 0; i < propertyKeyDisplayLimit; i++ {
+		keys = append(keys, fmt.Sprintf("k%02d", i))
+	}
+	if got := formatPropertyKeyList(keys); strings.Contains(got, "more)") {
+		t.Errorf("exactly %d keys must not fold, got %q", propertyKeyDisplayLimit, got)
+	}
+	keys = append(keys, "overflow")
+	if got := formatPropertyKeyList(keys); !strings.Contains(got, "(1 more)") {
+		t.Errorf("%d keys should fold into '(1 more)', got %q", propertyKeyDisplayLimit+1, got)
+	}
+}
+
+// TestTypeMismatchHintSuffix_EmptyWhenUndeclared pins that a field with
+// neither enum nor description adds no suffix — the deep-mismatch fallback
+// message must stay byte-identical to the legacy wording in that case.
+func TestTypeMismatchHintSuffix_EmptyWhenUndeclared(t *testing.T) {
+	t.Parallel()
+	tm := &typeMismatchError{path: "a.b", expected: "string", got: "number"}
+	if got := tm.hintSuffix(); got != "" {
+		t.Errorf("no enum/description → empty suffix, got %q", got)
+	}
+}
+
+// TestValidateAgainstSchema_StrictUnexpectedProperty_CaseOnlyTypo pins the
+// did-you-mean for a key that differs from a legal one only in casing /
+// underscore style — a high-frequency LLM slip.
+func TestValidateAgainstSchema_StrictUnexpectedProperty_CaseOnlyTypo(t *testing.T) {
+	t.Parallel()
+	schema := parseSchema(t, `{
+		"type":"object",
+		"additionalProperties":false,
+		"properties":{"background_color":{"type":"string"}}
+	}`)
+	err := validateAgainstSchema(map[string]interface{}{"Background_Color": "#fff"}, schema, "")
+	if err == nil {
+		t.Fatal("case-typo key under strict schema must fail")
+	}
+	if !strings.Contains(err.Error(), `did you mean "background_color"?`) {
+		t.Errorf("want case-insensitive did-you-mean; got %q", err.Error())
+	}
+}
+
+// TestValidateValueAgainstSchema_RequiredMissingRealSchema replays 场景3
+// of the doubao case against the real embedded flag-schemas.json: a
+// rich_text segment without "type" must inline the field's type, enum and
+// description while keeping the --print-schema pointer.
+func TestValidateValueAgainstSchema_RequiredMissingRealSchema(t *testing.T) {
+	t.Parallel()
+	fv := mapFlagView{command: "+cells-set"}
+	value := parseValue(t, `[[{"rich_text":[{"text":"x"}]}]]`)
+	err := validateValueAgainstSchema(fv, "cells", value)
+	if err == nil {
+		t.Fatal("rich_text without type must fail against the embedded schema")
+	}
+	msg := err.Error()
+	for _, want := range []string{
+		`required property "type" is missing`,
+		`expected type "string"`,
+		"one of [",
+		`"text"`,
+		"--print-schema",
+	} {
+		if !strings.Contains(msg, want) {
+			t.Errorf("want %q in real-schema required-missing error; got %q", want, msg)
+		}
+	}
+}
+
+// TestValidateValueAgainstSchema_DeepMismatchRealSchema replays 场景4: a
+// numeric rich_text "type" three levels deep gets the field's enum inline
+// (no whole-shape skeleton), still with the --print-schema pointer.
+func TestValidateValueAgainstSchema_DeepMismatchRealSchema(t *testing.T) {
+	t.Parallel()
+	fv := mapFlagView{command: "+cells-set"}
+	value := parseValue(t, `[[{"rich_text":[{"type":42,"text":"x"}]}]]`)
+	err := validateValueAgainstSchema(fv, "cells", value)
+	if err == nil {
+		t.Fatal("numeric rich_text type must fail against the embedded schema")
+	}
+	msg := err.Error()
+	for _, want := range []string{
+		`expected type "string", got "number"`,
+		"one of [",
+		`"text"`,
+		"--print-schema",
+	} {
+		if !strings.Contains(msg, want) {
+			t.Errorf("want %q in real-schema deep-mismatch error; got %q", want, msg)
+		}
+	}
+	if strings.Contains(msg, "expected shape:") {
+		t.Errorf("deep mismatch must not inline a skeleton; got %q", msg)
+	}
+}
+
+// TestValidateValueAgainstSchema_AggregatesMultipleErrors pins the
+// aggregate path: a payload with several independent problems reports them
+// all in one numbered reply (each with its own teaching hint) instead of
+// the fail-fast fix-one-retry-hit-the-next loop.
+func TestValidateValueAgainstSchema_AggregatesMultipleErrors(t *testing.T) {
+	t.Parallel()
+	fv := mapFlagView{command: "+cells-set"}
+	// Two independent problems in one --cells payload: cell[0][0].rich_text[0]
+	// misses required "type"; cell[0][1].note has the wrong type.
+	value := parseValue(t, `[[{"rich_text":[{"text":"x"}]},{"note":12.5}]]`)
+	err := validateValueAgainstSchema(fv, "cells", value)
+	if err == nil {
+		t.Fatal("payload with two problems must fail")
+	}
+	msg := err.Error()
+	for _, want := range []string{
+		"2 validation errors:",
+		`1) required property "type" is missing`,
+		`one of ["text"`, // teaching hint rides along in aggregate mode too
+		`2) [0][1].note: expected type "string"`,
+		"--print-schema",
+	} {
+		if !strings.Contains(msg, want) {
+			t.Errorf("want %q in aggregated error; got %q", want, msg)
+		}
+	}
+}
+
+// TestValidateValueAgainstSchema_AggregateCapTruncates pins the display
+// cap: a pathological payload reports schemaErrorDisplayLimit entries and
+// an explicit truncation tail, never the full flood.
+func TestValidateValueAgainstSchema_AggregateCapTruncates(t *testing.T) {
+	t.Parallel()
+	fv := mapFlagView{command: "+cells-set"}
+	// Seven cells all missing required rich_text "type" → 7 independent errors.
+	row := make([]string, 0, 7)
+	for i := 0; i < 7; i++ {
+		row = append(row, `{"rich_text":[{"text":"x"}]}`)
+	}
+	value := parseValue(t, `[[`+strings.Join(row, ",")+`]]`)
+	err := validateValueAgainstSchema(fv, "cells", value)
+	if err == nil {
+		t.Fatal("payload with seven problems must fail")
+	}
+	msg := err.Error()
+	if !strings.Contains(msg, "5+ validation errors:") {
+		t.Errorf("want capped header '5+ validation errors:'; got %q", msg)
+	}
+	if !strings.Contains(msg, "more errors not shown") {
+		t.Errorf("want truncation tail; got %q", msg)
+	}
+	if strings.Contains(msg, "6)") {
+		t.Errorf("must not render entries beyond the display limit; got %q", msg)
+	}
+}
+
+// TestCollectSchemaErrors_OneOfProbeDoesNotLeak pins that failed oneOf
+// alternatives don't leak probe errors into the caller's collector when a
+// later alternative matches.
+func TestCollectSchemaErrors_OneOfProbeDoesNotLeak(t *testing.T) {
+	t.Parallel()
+	schema := parseSchema(t, `{"oneOf":[{"type":"string"},{"type":"number"}]}`)
+	c := &schemaErrorCollector{}
+	collectSchemaErrors(42.0, schema, "", c)
+	if len(c.errs) != 0 {
+		t.Errorf("number matches the second oneOf alternative; want no errors, got %v", c.errs)
+	}
+}
+
+func TestOneLineDescription(t *testing.T) {
+	t.Parallel()
+	if got := oneLineDescription("   "); got != "" {
+		t.Errorf("whitespace-only → empty, got %q", got)
+	}
+	if got := oneLineDescription("line one\n  line two"); got != "line one line two" {
+		t.Errorf("multi-line collapse = %q", got)
+	}
+	long := strings.Repeat("x", 200)
+	got := oneLineDescription(long)
+	if !strings.HasSuffix(got, "…") || len([]rune(got)) != descriptionMaxLen+1 {
+		t.Errorf("long description should truncate to %d runes + ellipsis, got %d", descriptionMaxLen, len([]rune(got)))
+	}
+}
+
 func TestPathDepth(t *testing.T) {
 	t.Parallel()
 	cases := []struct {

shortcuts/sheets/flag_schemas_gen.go

@@ -34,6 +34,7 @@ var commandsWithSchema = map[string]struct{}{
 	"+rows-resize":           {},
 	"+sparkline-create":      {},
 	"+sparkline-update":      {},
+	"+styles-put":            {},
 	"+table-put":             {},
 	"+workbook-create":       {},
 }

shortcuts/sheets/helpers.go

@@ -11,7 +11,6 @@ import (
 	"context"
 	"encoding/json"
 	"errors"
-	"fmt"
 	neturl "net/url"
 	"strings"
 
@@ -407,6 +406,13 @@ func parseJSONFlag(runtime flagView, name string) (interface{}, error) {
 		}
 		return nil, sheetsValidationForFlag(name, "--%s: invalid JSON: %v", name, err).WithCause(err)
 	}
+	// Unambiguous habitual shapes are rewritten onto the wire contract
+	// before validation (see jsonFlagNormalizers). Runs on the parsed value,
+	// so both the standalone cobra path and +batch-update sub-ops (whose
+	// mapFlagView.Str re-encodes composites through here) get the rewrite.
+	if norm := jsonFlagNormalizers[runtime.Command()][name]; norm != nil {
+		out = norm(out)
+	}
 	// Schema-driven flag validation at the user-input boundary. Skips
 	// --properties (validated at the input-builder tail after enhance
 	// hooks fill in flat-flag-derived fields) and any flag without an
@@ -417,6 +423,92 @@ func parseJSONFlag(runtime flagView, name string) (interface{}, error) {
 	return out, nil
 }
 
+// jsonFlagNormalizers rewrites, per (command, flag), unambiguous habitual
+// input shapes onto the wire contract before schema validation — same
+// contract as enum normalization: only a shape whose meaning is beyond
+// doubt may be rewritten; anything ambiguous must fail with a prescription
+// instead. Applied to the parsed JSON value inside parseJSONFlag.
+var jsonFlagNormalizers = map[string]map[string]func(interface{}) interface{}{
+	"+cells-set":    {"cells": wrapLoneCellObject},
+	"+chart-create": {"properties": normalizeChartHexColors},
+	"+chart-update": {"properties": normalizeChartHexColors},
+}
+
+// normalizeChartHexColors walks a chart properties payload and prefixes bare
+// 6/8-digit hex values on color keys with '#' (4472C4 → #4472C4 — the
+// Excel-habit form the chart backend rejects with "expected rgba() or
+// #RRGGBB/#RRGGBBAA"). In-place, recursive; anything not unambiguously a
+// bare hex color is untouched.
+func normalizeChartHexColors(v interface{}) interface{} {
+	switch t := v.(type) {
+	case map[string]interface{}:
+		for k, val := range t {
+			if s, ok := val.(string); ok && isColorKey(k) && isBareHexColor(s) {
+				t[k] = "#" + s
+				continue
+			}
+			normalizeChartHexColors(val)
+		}
+	case []interface{}:
+		for _, e := range t {
+			normalizeChartHexColors(e)
+		}
+	}
+	return v
+}
+
+func isColorKey(k string) bool {
+	return k == "color" || strings.HasSuffix(k, "_color") || strings.HasSuffix(k, "Color")
+}
+
+func isBareHexColor(s string) bool {
+	if len(s) != 6 && len(s) != 8 {
+		return false
+	}
+	for _, r := range s {
+		switch {
+		case r >= '0' && r <= '9', r >= 'a' && r <= 'f', r >= 'A' && r <= 'F':
+		default:
+			return false
+		}
+	}
+	return true
+}
+
+// cellObjectKeys pins the property vocabulary of a single cell in the
+// +cells-set --cells schema ([[{…}]]). Drift against the embedded schema is
+// guarded by TestCellObjectKeys_MatchEmbeddedSchema.
+var cellObjectKeys = map[string]struct{}{
+	"border_styles":   {},
+	"cell_styles":     {},
+	"data_validation": {},
+	"formula":         {},
+	"multiple_values": {},
+	"note":            {},
+	"rich_text":       {},
+	"value":           {},
+}
+
+// wrapLoneCellObject rewrites a bare cell object into the [[cell]] the
+// --cells contract expects. Eval traces show agents writing a single cell
+// routinely pass {"value":…} without the two array layers; when every key
+// belongs to the cell vocabulary the meaning is a 1×1 write and the wrap is
+// safe. Anything else (unknown keys, arrays — one bracket layer could be a
+// row or a column) is returned untouched for the schema validator to
+// prescribe.
+func wrapLoneCellObject(v interface{}) interface{} {
+	obj, ok := v.(map[string]interface{})
+	if !ok || len(obj) == 0 {
+		return v
+	}
+	for k := range obj {
+		if _, known := cellObjectKeys[k]; !known {
+			return v
+		}
+	}
+	return []interface{}{[]interface{}{obj}}
+}
+
 // requireJSONObject is parseJSONFlag + a type assertion to map[string]interface{}.
 func requireJSONObject(runtime flagView, name string) (map[string]interface{}, error) {
 	v, err := parseJSONFlag(runtime, name)
@@ -448,146 +540,3 @@ func requireJSONArray(runtime flagView, name string) ([]interface{}, error) {
 	}
 	return a, nil
 }
-
-// ─── style flags (shared by +cells-set-style and +cells-batch-set-style) ─
-
-// buildCellStyleFromFlags reads the 12 flat style flags and returns the
-// cell_styles map expected by set_cell_range. Skips any flag the user
-// didn't set so partial styles work.
-func buildCellStyleFromFlags(runtime flagView) map[string]interface{} {
-	style := map[string]interface{}{}
-	if v := runtime.Str("background-color"); v != "" {
-		style["background_color"] = v
-	}
-	if v := runtime.Str("font-color"); v != "" {
-		style["font_color"] = v
-	}
-	if v := runtime.Str("font-family"); v != "" {
-		style["font_family"] = v
-	}
-	if runtime.Changed("font-size") && runtime.Float64("font-size") > 0 {
-		style["font_size"] = runtime.Float64("font-size")
-	}
-	if v := runtime.Str("font-style"); v != "" {
-		style["font_style"] = v
-	}
-	if v := runtime.Str("font-weight"); v != "" {
-		style["font_weight"] = v
-	}
-	if v := runtime.Str("font-line"); v != "" {
-		style["font_line"] = v
-	}
-	if v := runtime.Str("horizontal-alignment"); v != "" {
-		style["horizontal_alignment"] = v
-	}
-	if v := runtime.Str("vertical-alignment"); v != "" {
-		style["vertical_alignment"] = v
-	}
-	if v := runtime.Str("word-wrap"); v != "" {
-		style["word_wrap"] = v
-	}
-	if v := runtime.Str("number-format"); v != "" {
-		style["number_format"] = v
-	}
-	return style
-}
-
-// cellStyleAliases maps shorthand cell_styles field names that models commonly
-// hallucinate (Excel / openpyxl / CSS conventions) onto the canonical field
-// names the backend expects. Only the unambiguous alignment shorthands are
-// aliased — they are the high-frequency miss; ambiguous guesses (e.g. "color",
-// "bg_color", "text_align") are intentionally left out so a wrong guess still
-// surfaces as an error rather than being silently reinterpreted.
-var cellStyleAliases = []struct{ alias, canonical string }{
-	{"horizontal_align", "horizontal_alignment"},
-	{"halign", "horizontal_alignment"},
-	{"vertical_align", "vertical_alignment"},
-	{"valign", "vertical_alignment"},
-}
-
-// normalizeCellStyleAliases renames known shorthand keys in a single
-// cell_styles map to their canonical equivalents, in place, so a model that
-// writes e.g. "horizontal_align" instead of "horizontal_alignment" still
-// applies the style instead of hitting an "unsupported field" error (--styles)
-// or having the field silently dropped by the backend (typed --cells). If both
-// the shorthand and its canonical key are present it returns a validation error
-// rather than picking one. path labels the map for the error message.
-func normalizeCellStyleAliases(style map[string]interface{}, path string) error {
-	if len(style) == 0 {
-		return nil
-	}
-	for _, a := range cellStyleAliases {
-		v, ok := style[a.alias]
-		if !ok {
-			continue
-		}
-		if _, exists := style[a.canonical]; exists {
-			return common.ValidationErrorf("%s.%s conflicts with %s; pass only %s", path, a.alias, a.canonical, a.canonical)
-		}
-		style[a.canonical] = v
-		delete(style, a.alias)
-	}
-	return nil
-}
-
-// normalizeTypedCellsStyleAliases walks a typed --cells 2D array and applies
-// normalizeCellStyleAliases to every cell's inline cell_styles object, so the
-// alignment shorthands are accepted on +cells-set the same as on --styles.
-// Structure is checked leniently to match the pass-through contract: any
-// element that isn't the expected shape is skipped, not rejected.
-func normalizeTypedCellsStyleAliases(cells []interface{}, path string) error {
-	for r, rowRaw := range cells {
-		row, ok := rowRaw.([]interface{})
-		if !ok {
-			continue
-		}
-		for c, cellRaw := range row {
-			cell, ok := cellRaw.(map[string]interface{})
-			if !ok {
-				continue
-			}
-			st, ok := cell["cell_styles"].(map[string]interface{})
-			if !ok {
-				continue
-			}
-			if err := normalizeCellStyleAliases(st, fmt.Sprintf("%s[%d][%d].cell_styles", path, r, c)); err != nil {
-				return err
-			}
-		}
-	}
-	return nil
-}
-
-// borderStylesFromFlag parses --border-styles as a JSON object (top/bottom/
-// left/right with style sub-objects). Returns nil when the flag is empty.
-func borderStylesFromFlag(runtime flagView) (map[string]interface{}, error) {
-	if runtime.Str("border-styles") == "" {
-		return nil, nil
-	}
-	v, err := parseJSONFlag(runtime, "border-styles")
-	if err != nil {
-		return nil, err
-	}
-	m, ok := v.(map[string]interface{})
-	if !ok {
-		return nil, sheetsValidationForFlag("border-styles", "--border-styles must be a JSON object")
-	}
-	return m, nil
-}
-
-// requireAnyStyleFlag ensures at least one style-defining flag (style or
-// border) is set — otherwise the request would do nothing.
-func requireAnyStyleFlag(runtime flagView) error {
-	if len(buildCellStyleFromFlags(runtime)) > 0 {
-		return nil
-	}
-	if runtime.Str("border-styles") != "" {
-		return nil
-	}
-	return common.ValidationErrorf("at least one style flag is required (e.g. --background-color, --font-weight, --border-styles)").
-		WithParams(
-			sheetsInvalidParam("background-color", "required; specify at least one style flag"),
-			sheetsInvalidParam("font-weight", "required; specify at least one style flag"),
-			sheetsInvalidParam("border-styles", "required; specify at least one style flag"),
-		)
-}

shortcuts/sheets/json_flag_normalize_test.go

@@ -0,0 +1,209 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package sheets
+
+import (
+	"encoding/json"
+	"strings"
+	"testing"
+)
+
+// TestWrapLoneCellObject pins the auto-wrap contract: a bare cell object —
+// the classic missing-[[…]] shape agents produce for a 1×1 write — is
+// rewritten to [[cell]]; anything whose meaning is not beyond doubt stays
+// untouched for the schema validator to prescribe.
+func TestWrapLoneCellObject(t *testing.T) {
+	t.Parallel()
+	cases := []struct {
+		name    string
+		in      string
+		wrapped bool
+	}{
+		{"lone value cell", `{"value":"hi"}`, true},
+		{"lone formula cell with styles", `{"formula":"=SUM(A1:A3)","cell_styles":{"font_weight":"bold"}}`, true},
+		{"unknown key stays", `{"value":"hi","range":"A1"}`, false},
+		{"array of cells stays (row vs column ambiguous)", `[{"value":"a"},{"value":"b"}]`, false},
+		{"proper 2D array stays", `[[{"value":"a"}]]`, false},
+		{"empty object stays", `{}`, false},
+		{"scalar stays", `"hi"`, false},
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			var v interface{}
+			if err := json.Unmarshal([]byte(tc.in), &v); err != nil {
+				t.Fatalf("bad fixture: %v", err)
+			}
+			out := wrapLoneCellObject(v)
+			_, isWrapped := out.([]interface{})
+			_, wasArray := v.([]interface{})
+			if tc.wrapped && (!isWrapped || wasArray) {
+				t.Errorf("expected wrap to [[cell]], got %#v", out)
+			}
+			if !tc.wrapped && !wasArray && isWrapped {
+				t.Errorf("expected no wrap, got %#v", out)
+			}
+			if tc.wrapped {
+				rows, _ := out.([]interface{})
+				if len(rows) != 1 {
+					t.Fatalf("want 1 row, got %d", len(rows))
+				}
+				cells, _ := rows[0].([]interface{})
+				if len(cells) != 1 {
+					t.Fatalf("want 1 cell, got %d", len(cells))
+				}
+			}
+		})
+	}
+}
+
+// TestCellObjectKeys_MatchEmbeddedSchema drift-guards the hardcoded cell
+// vocabulary against the embedded +cells-set --cells schema: if the spec
+// repo adds or removes a cell property, this fails and cellObjectKeys must
+// be updated (an outdated set only narrows the auto-wrap, but silently
+// narrowing is still drift).
+func TestCellObjectKeys_MatchEmbeddedSchema(t *testing.T) {
+	t.Parallel()
+	idx, err := loadFlagSchemas()
+	if err != nil {
+		t.Fatalf("loadFlagSchemas: %v", err)
+	}
+	raw, ok := idx.Flags["+cells-set"]["cells"]
+	if !ok {
+		t.Fatal("embedded schema for +cells-set --cells missing")
+	}
+	var schema schemaProperty
+	if err := json.Unmarshal(raw, &schema); err != nil {
+		t.Fatalf("unmarshal schema: %v", err)
+	}
+	cell := schema.Items
+	if cell != nil && cell.Items != nil {
+		cell = cell.Items
+	}
+	if cell == nil || len(cell.Properties) == 0 {
+		t.Fatal("schema shape changed: expected array→array→object with properties")
+	}
+	for k := range cell.Properties {
+		if _, ok := cellObjectKeys[k]; !ok {
+			t.Errorf("schema property %q missing from cellObjectKeys", k)
+		}
+	}
+	for k := range cellObjectKeys {
+		if _, ok := cell.Properties[k]; !ok {
+			t.Errorf("cellObjectKeys has %q which the schema no longer declares", k)
+		}
+	}
+}
+
+// TestCellsSet_LoneCellObjectAutoWraps runs the mounted path end-to-end: the
+// eval-trace failure shape (--cells with a bare object) now dry-runs clean
+// instead of failing "expected type array, got object".
+func TestCellsSet_LoneCellObjectAutoWraps(t *testing.T) {
+	t.Parallel()
+	sc := shortcutFromRegistry(t, "+cells-set")
+	stdout, _, err := runShortcutCapturingErr(t, sc, []string{
+		"--url", testURL,
+		"--sheet-name", "s",
+		"--range", "A1",
+		"--cells", `{"value":"hello"}`,
+		"--dry-run",
+	})
+	if err != nil {
+		t.Fatalf("lone cell object should auto-wrap to [[cell]], got: %v", err)
+	}
+	if !strings.Contains(stdout, "hello") {
+		t.Errorf("dry-run body should carry the cell value, got %q", stdout)
+	}
+}
+
+// TestTablePut_SheetsDecodeHints pins the two decode-failure prescriptions:
+// wrong JSON kind inlines the expected shape; mangled JSON steers to
+// stdin/@file.
+func TestTablePut_SheetsDecodeHints(t *testing.T) {
+	t.Parallel()
+
+	t.Run("type mismatch inlines skeleton", func(t *testing.T) {
+		t.Parallel()
+		sc := shortcutFromRegistry(t, "+table-put")
+		_, _, err := runShortcutCapturingErr(t, sc, []string{
+			"--url", testURL,
+			"--sheets", `{"sheets":[{"name":"s","columns":[{"name":"a"}],"data":[]}]}`,
+			"--dry-run",
+		})
+		ve := requireValidation(t, err, "--sheets: invalid JSON")
+		for _, want := range []string{"expected shape:", `"columns":["City","Revenue"]`, `"dtypes":{"Revenue":"float64"}`} {
+			if !strings.Contains(ve.Hint, want) {
+				t.Errorf("hint should contain %q, got %q", want, ve.Hint)
+			}
+		}
+	})
+
+	t.Run("syntax error steers to stdin or @file", func(t *testing.T) {
+		t.Parallel()
+		sc := shortcutFromRegistry(t, "+table-put")
+		_, _, err := runShortcutCapturingErr(t, sc, []string{
+			"--url", testURL,
+			"--sheets", `{"sheets":[)`,
+			"--dry-run",
+		})
+		ve := requireValidation(t, err, "--sheets: invalid JSON")
+		for _, want := range []string{"stdin", "@./payload.json"} {
+			if !strings.Contains(ve.Hint, want) {
+				t.Errorf("hint should contain %q, got %q", want, ve.Hint)
+			}
+		}
+	})
+}
+
+// TestNormalizeChartHexColors pins the '#' prefixing on bare hex color
+// values (eval V2U024: bars.color "4472C4" rejected server-side) and the
+// pass-through of everything else, including the parseJSONFlag wiring for
+// the batch sub-op path.
+func TestNormalizeChartHexColors(t *testing.T) {
+	t.Parallel()
+	props := map[string]interface{}{
+		"plotArea": map[string]interface{}{
+			"plot": map[string]interface{}{
+				"series": []interface{}{
+					map[string]interface{}{"bars": map[string]interface{}{"color": "4472C4"}},
+					map[string]interface{}{"line": map[string]interface{}{"color": "#ED7D31"}},
+					map[string]interface{}{"area": map[string]interface{}{"color": "rgba(1,2,3,0.5)"}},
+					map[string]interface{}{"font_color": "ED7D31AA", "label": "not a color 4472C4"},
+				},
+			},
+		},
+	}
+	normalizeChartHexColors(props)
+	series := props["plotArea"].(map[string]interface{})["plot"].(map[string]interface{})["series"].([]interface{})
+	if got := series[0].(map[string]interface{})["bars"].(map[string]interface{})["color"]; got != "#4472C4" {
+		t.Errorf("bare hex should gain #, got %v", got)
+	}
+	if got := series[1].(map[string]interface{})["line"].(map[string]interface{})["color"]; got != "#ED7D31" {
+		t.Errorf("already-prefixed color must not change, got %v", got)
+	}
+	if got := series[2].(map[string]interface{})["area"].(map[string]interface{})["color"]; got != "rgba(1,2,3,0.5)" {
+		t.Errorf("rgba color must not change, got %v", got)
+	}
+	last := series[3].(map[string]interface{})
+	if got := last["font_color"]; got != "#ED7D31AA" {
+		t.Errorf("8-digit hex on a *_color key should gain #, got %v", got)
+	}
+	if got := last["label"]; got != "not a color 4472C4" {
+		t.Errorf("non-color key must not change, got %v", got)
+	}
+
+	// Wiring: a +chart-create sub-op style view routes through parseJSONFlag
+	// and picks up the normalizer.
+	fv := newMapFlagViewForCommand("+chart-create", map[string]interface{}{
+		"properties": map[string]interface{}{"title": map[string]interface{}{"font_color": "112233"}},
+	})
+	out, err := parseJSONFlag(fv, "properties")
+	if err != nil {
+		t.Fatalf("parseJSONFlag: %v", err)
+	}
+	title := out.(map[string]interface{})["title"].(map[string]interface{})
+	if title["font_color"] != "#112233" {
+		t.Errorf("parseJSONFlag should apply the chart color normalizer, got %v", title["font_color"])
+	}
+}

shortcuts/sheets/lark_sheet_batch_update.go

@@ -5,6 +5,7 @@ package sheets
 
 import (
 	"context"
+	"fmt"
 	"strings"
 
 	"github.com/larksuite/cli/shortcuts/common"
@@ -29,10 +30,14 @@ import (
 // The tool's contract (post-translation):
 //   { excel_id, operations: [{tool_name, input}, ...], continue_on_error? }
 //
-// continue_on_error defaults to false (strict transaction): any failure
-// rolls back the whole batch. CLI leaves the default in place for the
-// three "fan-out" shortcuts since they're meant to be all-or-nothing;
-// only +batch-update lets callers flip it via --continue-on-error.
+// continue_on_error defaults to false (fail-fast): execution stops at the
+// first failing sub-op, but sub-ops already applied are NOT rolled back —
+// the server reports "N succeeded, M failed" and the N stay in the sheet
+// (verified against live batches; earlier docs wrongly promised a rollback,
+// which made agents resend whole batches and double-apply the successes).
+// CLI leaves the default in place for the fan-out shortcuts since they're
+// idempotent stamps; only +batch-update lets callers flip it via
+// --continue-on-error.
 
 // BatchUpdate accepts a CLI-shape operations array (each item
 // {shortcut, input}); on Validate / DryRun / Execute we translate each
@@ -42,7 +47,7 @@ import (
 var BatchUpdate = common.Shortcut{
 	Service:     "sheets",
 	Command:     "+batch-update",
-	Description: "Execute a batch of write shortcuts as a single atomic request (rolls back on failure by default).",
+	Description: "Execute a batch of write shortcuts in one request; fail-fast on the first failing sub-op (already-applied sub-ops are NOT rolled back).",
 	Risk:        "high-risk-write",
 	Scopes:      []string{"sheets:spreadsheet:write_only"},
 	AuthTypes:   []string{"user", "bot"},
@@ -64,7 +69,11 @@ var BatchUpdate = common.Shortcut{
 	DryRun: func(ctx context.Context, runtime *common.RuntimeContext) *common.DryRunAPI {
 		token, _ := resolveSpreadsheetToken(runtime)
 		input, _ := batchUpdateInput(runtime, token)
-		return invokeToolDryRun(token, ToolKindWrite, "batch_update", input)
+		dr := invokeToolDryRun(token, ToolKindWrite, "batch_update", input)
+		if batchNeedsDimInsertBeforeStyleWarning(runtime) {
+			dr.Set("warning_message", dimInsertBeforeStyleWarning)
+		}
+		return dr
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
 		token, err := resolveSpreadsheetTokenExec(runtime)
@@ -75,6 +84,9 @@ var BatchUpdate = common.Shortcut{
 		if err != nil {
 			return err
 		}
+		if batchNeedsDimInsertBeforeStyleWarning(runtime) {
+			fmt.Fprintln(runtime.IO().ErrOut, dimInsertBeforeStyleWarning)
+		}
 		out, err := callTool(ctx, runtime, token, ToolKindWrite, "batch_update", input)
 		if err != nil {
 			return err
@@ -83,7 +95,8 @@ var BatchUpdate = common.Shortcut{
 		return nil
 	},
 	Tips: []string{
-		"Default is strict transaction — any sub-tool failure rolls the whole batch back. Pass --continue-on-error to keep partial successes.",
+		"high-risk-write: always pass --yes (or --dry-run to preview) — without it the call exits 10 asking for confirmation.",
+		"Execution is fail-fast, NOT transactional: on \"N succeeded, M failed\" the succeeded sub-ops stay applied (no rollback) — fix the failure and resend ONLY the operations from the first failed index onward; resending the whole batch re-applies the succeeded ones. Pass --continue-on-error to keep going past failures instead.",
 		"Each sub-op is {shortcut, input}. Do NOT pass input.operation (implied by shortcut name) or input.excel_id / input.url (set at the +batch-update top level).",
 	},
 }
@@ -124,6 +137,46 @@ func batchUpdateInput(runtime *common.RuntimeContext, token string) (map[string]
 	return input, nil
 }
 
+// batchNeedsDimInsertBeforeStyleWarning reports whether any +dim-insert sub-op
+// requests --inherit-style before at the first row/column, where the
+// preceding-side style cannot be copied (no preceding row/column exists).
+func batchNeedsDimInsertBeforeStyleWarning(runtime *common.RuntimeContext) bool {
+	rawOps, err := parseBatchOperationsFlag(runtime)
+	if err != nil {
+		return false
+	}
+	for _, raw := range rawOps {
+		op, ok := raw.(map[string]interface{})
+		if !ok {
+			continue
+		}
+		sc, _ := op["shortcut"].(string)
+		if sc != "+dim-insert" {
+			continue
+		}
+		input, _ := op["input"].(map[string]interface{})
+		isBefore := false
+		for _, key := range []string{"inherit-style", "inherit_style", "inheritStyle"} {
+			if v, _ := input[key].(string); strings.EqualFold(v, "before") {
+				isBefore = true
+				break
+			}
+		}
+		if !isBefore {
+			continue
+		}
+		posRaw, hasPos := input["position"]
+		if !hasPos {
+			continue
+		}
+		// Warn only at the first row/column (idx 0).
+		if _, idx, err := parseA1Position(strings.TrimSpace(fmt.Sprintf("%v", posRaw))); err == nil && idx == 0 {
+			return true
+		}
+	}
+	return false
+}
+
 // parseBatchOperationsFlag accepts --operations as either a JSON array (the
 // operations list directly) or an envelope object { operations, continue_on_error }
 // for back-compat with the legacy --data shape. Returns the operations array.
@@ -160,6 +213,11 @@ var CellsBatchSetStyle = common.Shortcut{
 	AuthTypes:   []string{"user", "bot"},
 	HasFormat:   true,
 	Flags:       flagsFor("+cells-batch-set-style"),
+	Tips: []string{
+		"DEPRECATED: superseded by +styles-put, whose one spec also covers merges, row/col sizes and freeze — prefer it for new work.",
+		`Example: lark-cli sheets +cells-batch-set-style --url <URL> --ranges '["Sheet1!A1:B2","汇总!C1:C9"]' --font-weight bold`,
+		"Every range carries its sheet-NAME prefix (Sheet1!A1:B2, not a sheet_id) — there is no --sheet-id / --sheet-name flag here.",
+	},
 	Validate: func(ctx context.Context, runtime *common.RuntimeContext) error {
 		if _, err := resolveSpreadsheetToken(runtime); err != nil {
 			return err
@@ -189,6 +247,10 @@ var CellsBatchSetStyle = common.Shortcut{
 		if err != nil {
 			return err
 		}
+		// Phase-1 deprecation (docs already point at +styles-put): keep the
+		// command working, steer new usage to the superset in-band.
+		fmt.Fprintln(runtime.IO().ErrOut,
+			"note: +cells-batch-set-style is superseded by +styles-put (one spec covers styles + merges + row/col sizes + freeze); prefer +styles-put for new work")
 		out, err := callTool(ctx, runtime, token, ToolKindWrite, "batch_update", input)
 		if err != nil {
 			return err

shortcuts/sheets/lark_sheet_batch_update_test.go

@@ -58,6 +58,39 @@ func TestBatchUpdate_TranslatesShortcutToToolName(t *testing.T) {
 	}
 }
 
+func TestBatchUpdate_DimInsertInheritAfterCopiesFollowingStyle(t *testing.T) {
+	t.Parallel()
+
+	body := parseDryRunBody(t, BatchUpdate, []string{
+		"--url", testURL,
+		"--operations", `[
+		  {"shortcut":"+dim-insert","input":{"sheet_id":"sh1","position":"D","count":1,"inherit_style":"after"}}
+		]`,
+		"--yes",
+	})
+	input := decodeToolInput(t, body, "batch_update")
+	ops, _ := input["operations"].([]interface{})
+	if len(ops) != 1 {
+		t.Fatalf("operations length = %d, want 1", len(ops))
+	}
+	op := ops[0].(map[string]interface{})
+	if op["tool_name"] != "modify_sheet_structure" {
+		t.Fatalf("tool_name = %v, want modify_sheet_structure", op["tool_name"])
+	}
+	in, _ := op["input"].(map[string]interface{})
+	// inherit_style=after copies the following column's style via a plain
+	// before-insert at the same position (the backend anchors on the following
+	// column), so position stays D with side=before.
+	assertInputEquals(t, in, map[string]interface{}{
+		"excel_id":  testToken,
+		"sheet_id":  "sh1",
+		"operation": "insert",
+		"position":  "D",
+		"count":     float64(1),
+		"side":      "before",
+	})
+}
+
 func TestBatchUpdate_HighRiskWriteRequiresYes(t *testing.T) {
 	t.Parallel()
 	stdout, stderr, err := runShortcutCapturingErr(t, BatchUpdate, []string{
@@ -405,6 +438,21 @@ func TestBatchUpdate_TranslatorRejects(t *testing.T) {
 			opsJSON:   `[{"shortcut":"+cells-set","input":"not-an-object"}]`,
 			wantMatch: "'input' must be a JSON object",
 		},
+		{
+			name:      "wrapped cell_styles structure",
+			opsJSON:   `[{"shortcut":"+cells-set-style","input":{"sheet_name":"s","range":"A1","cell_styles":{"background_color":"#EBF1F8"}}}]`,
+			wantMatch: "do not wrap in cell_styles",
+		},
+		{
+			name:      "wrapped styles structure",
+			opsJSON:   `[{"shortcut":"+cells-set-style","input":{"sheet_name":"s","range":"A1","styles":{"font_weight":"bold"}}}]`,
+			wantMatch: "do not wrap in styles",
+		},
+		{
+			name:      "wrapped cell_merges structure",
+			opsJSON:   `[{"shortcut":"+cells-set-style","input":{"sheet_name":"s","range":"A1","cell_merges":[{"range":"A1:B1"}]}}]`,
+			wantMatch: "do not wrap in cell_merges",
+		},
 	}
 	for _, tc := range cases {
 		t.Run(tc.name, func(t *testing.T) {
@@ -420,6 +468,99 @@ func TestBatchUpdate_TranslatorRejects(t *testing.T) {
 	}
 }
 
+// TestBatchUpdate_FlattenedStyleKeysNotMistakenForWrapper guards the
+// wrapped-structure rejection against overreach: the same style fields in
+// their correct flattened form must translate cleanly — only the wrapper
+// container keys (cell_styles / styles / cell_merges) are rejected.
+func TestBatchUpdate_FlattenedStyleKeysNotMistakenForWrapper(t *testing.T) {
+	t.Parallel()
+	got, err := translateBatchOp(map[string]interface{}{
+		"shortcut": "+cells-set-style",
+		"input": map[string]interface{}{
+			"sheet_name":       "s",
+			"range":            "A1",
+			"background_color": "#EBF1F8",
+			"font_weight":      "bold",
+		},
+	}, testToken, 0)
+	if err != nil {
+		t.Fatalf("flattened style keys must pass the wrapper check, got %v", err)
+	}
+	input := got["input"].(map[string]interface{})
+	cells := input["cells"].([][]interface{})
+	style := cells[0][0].(map[string]interface{})["cell_styles"].(map[string]interface{})
+	if style["background_color"] != "#EBF1F8" || style["font_weight"] != "bold" {
+		t.Fatalf("translated style = %#v", style)
+	}
+}
+
+// TestBatchUpdate_WrapperKeysDisjointFromSubOpFlags locks the static
+// assumption wrappedSubOpInputKeys relies on: no shortcut registered in
+// batchOpDispatch declares a flag named cell_styles / cell_merges / styles.
+// If a future dispatch-table addition (e.g. +table-put) carries one of these
+// flags, its legitimate input would be silently rejected by the wrapper
+// check — this test turns that silent breakage into a build-time failure.
+func TestBatchUpdate_WrapperKeysDisjointFromSubOpFlags(t *testing.T) {
+	t.Parallel()
+	wrapped := make(map[string]struct{}, len(wrappedSubOpInputKeys))
+	for _, k := range wrappedSubOpInputKeys {
+		wrapped[k] = struct{}{}
+	}
+	for shortcut := range batchOpDispatch {
+		for _, f := range flagsFor(shortcut) {
+			key := strings.ReplaceAll(f.Name, "-", "_")
+			if _, clash := wrapped[key]; clash {
+				t.Errorf("%s declares flag --%s which collides with wrappedSubOpInputKeys; "+
+					"exempt this shortcut from the wrapper check before adding it to batchOpDispatch",
+					shortcut, f.Name)
+			}
+		}
+	}
+}
+
+// TestBatchUpdate_AggregatesMultipleOpErrors pins op-level aggregation: when
+// several operations are invalid, one reply names them all (numbered, with
+// each op's own error) instead of failing on the first bad op only. A single
+// bad op keeps the historical single-error message (no aggregate wrapper).
+func TestBatchUpdate_AggregatesMultipleOpErrors(t *testing.T) {
+	t.Parallel()
+
+	t.Run("two bad ops reported together", func(t *testing.T) {
+		t.Parallel()
+		_, _, err := runShortcutCapturingErr(t, BatchUpdate, []string{
+			"--url", testURL,
+			"--operations", `[
+				{"shortcut":"+cells-set-magic","input":{}},
+				{"shortcut":"+cells-set","input":{"sheet_name":"s","range":"A1"}},
+				{"shortcut":"+cells-clear","input":{"sheet_name":"s","range":"A1"}}
+			]`,
+			"--yes", "--dry-run",
+		})
+		requireValidation(t, err, "2 of 3 operations failed validation")
+		for _, want := range []string{"1) ", "2) ", "operations[0]", "operations[1]"} {
+			if !strings.Contains(err.Error(), want) {
+				t.Errorf("aggregated op error should contain %q, got %q", want, err.Error())
+			}
+		}
+	})
+
+	t.Run("single bad op keeps plain message", func(t *testing.T) {
+		t.Parallel()
+		_, _, err := runShortcutCapturingErr(t, BatchUpdate, []string{
+			"--url", testURL,
+			"--operations", `[
+				{"shortcut":"+cells-set-magic","input":{}},
+				{"shortcut":"+cells-clear","input":{"sheet_name":"s","range":"A1"}}
+			]`,
+			"--yes", "--dry-run",
+		})
+		requireValidation(t, err, "not allowed in +batch-update")
+		if strings.Contains(err.Error(), "operations failed validation") {
+			t.Errorf("single bad op must not get the aggregate wrapper, got %q", err.Error())
+		}
+	})
+}
+
 // TestBatchUpdate_PrescriptiveHints pins the recovery hints that ride on the
 // highest-frequency batch failures, so an agent can repair its payload in a
 // single retry without --help / --print-schema round trips.

shortcuts/sheets/lark_sheet_range_operations.go

@@ -67,7 +67,7 @@ var CellsClear = common.Shortcut{
 		return nil
 	},
 	Tips: []string{
-		"high-risk-write — always preview with --dry-run; clear is not undoable.",
+		"high-risk-write — pass --yes to confirm (exit 10 without it), or preview with --dry-run first; clear is not undoable.",
 		"Can't delete an embedded pivot/chart by clearing cells — remove the object itself with +pivot-delete / +chart-delete.",
 	},
 }
@@ -266,9 +266,13 @@ var ColsResize = common.Shortcut{
 	AuthTypes:   []string{"user", "bot"},
 	HasFormat:   true,
 	Flags:       flagsFor("+cols-resize"),
-	Validate:    validateViaResize("column"),
-	DryRun:      resizeDryRun("column"),
-	Execute:     resizeExecute("column"),
+	Tips: []string{
+		"Example: lark-cli sheets +cols-resize --url <URL> --sheet-name Sheet1 --range A:C --width 120",
+		`Different widths per column in one atomic call: --widths '{"A":80,"C:E":120}'. Widths are pixels (px ≈ chars × 8 + 16), not Excel character units.`,
+	},
+	Validate: validateViaResize("column"),
+	DryRun:   resizeDryRun("column"),
+	Execute:  resizeExecute("column"),
 }
 
 // resizeDryRun / resizeExecute route a resize shortcut through resizeToolCall

shortcuts/sheets/lark_sheet_read_data.go

@@ -69,8 +69,7 @@ var CellsGet = common.Shortcut{
 		if err != nil {
 			return err
 		}
-		runtime.Out(out, nil)
-		return nil
+		return emitReadResult(runtime, out)
 	},
 }
 
@@ -84,21 +83,28 @@ func cellsGetInput(runtime *common.RuntimeContext, token, sheetID, sheetName str
 	if runtime.Bool("skip-hidden") {
 		input["skip_hidden"] = true
 	}
+	// Preserve omission so the tool can keep the legacy fallback where
+	// skip_filter inherits skip_hidden. An explicit false must still be sent.
+	if runtime.Changed("skip-filter") {
+		input["skip_filter"] = runtime.Bool("skip-filter")
+	}
 	// --cell-limit was removed from the CLI surface; --max-chars is the single
 	// read cap. Pin cell_limit very high so the tool's own default never binds
 	// before max_chars.
 	input["cell_limit"] = unboundedReadLimit
-	if n := runtime.Int("max-chars"); n > 0 {
+	if n, ok := maxCharsInput(runtime); ok {
 		input["max_chars"] = n
 	}
 	return input
 }
 
 // applyIncludeToCellsGet maps the fine-grained --include vocabulary to the
-// tool's two coarse switches:
+// tool's switches:
 //
 //   - include_styles (bool) — toggled by "style" presence
 //   - value_render_option (enum) — "formula" → formula; otherwise omitted
+//   - include_truncation_info (bool) — toggled by "truncation" presence; makes
+//     the tool estimate and return per-cell isRowTruncated / isColTruncated
 //
 // "value", "comment", and "data_validation" are always returned by the tool
 // per the schema; they have no dedicated knob today but are accepted in
@@ -119,6 +125,9 @@ func applyIncludeToCellsGet(input map[string]interface{}, include []string) {
 	if want["formula"] {
 		input["value_render_option"] = "formula"
 	}
+	if want["truncation"] {
+		input["include_truncation_info"] = true
+	}
 }
 
 // CsvGet wraps get_range_as_csv: pull one range as RFC 4180 CSV with optional
@@ -139,9 +148,6 @@ var CsvGet = common.Shortcut{
 		if _, _, err := resolveSheetSelector(runtime); err != nil {
 			return err
 		}
-		if strings.TrimSpace(runtime.Str("range")) == "" {
-			return sheetsValidationForFlag("range", "--range is required")
-		}
 		return nil
 	},
 	DryRun: func(ctx context.Context, runtime *common.RuntimeContext) *common.DryRunAPI {
@@ -165,16 +171,25 @@ var CsvGet = common.Shortcut{
 		if !runtime.Bool("include-row-prefix") {
 			out = stripRowPrefixFromCsvOutput(out)
 		}
-		runtime.Out(out, nil)
-		return nil
+		return emitReadResult(runtime, out)
 	},
 }
 
+// csvGetFullSheetRange is the range sent when --range is omitted: the tool
+// requires one, but clips anything past the grid bounds and reports the clip
+// in actual_range — so an over-wide whole-columns range reads the entire
+// sheet in one call, with no workbook-info pre-flight. Eval traces show
+// "read the whole sheet" as a recurring intent (--range was the single most
+// missed required flag once the rest of the surface was fixed).
+const csvGetFullSheetRange = "A:ZZZ"
+
 func csvGetInput(runtime *common.RuntimeContext, token, sheetID, sheetName string) map[string]interface{} {
 	input := map[string]interface{}{"excel_id": token}
 	sheetSelectorForToolInput(input, sheetID, sheetName)
 	if r := strings.TrimSpace(runtime.Str("range")); r != "" {
 		input["range"] = r
+	} else {
+		input["range"] = csvGetFullSheetRange
 	}
 	if runtime.Bool("skip-hidden") {
 		input["skip_hidden"] = true
@@ -183,7 +198,7 @@ func csvGetInput(runtime *common.RuntimeContext, token, sheetID, sheetName strin
 	// read cap. Pin max_rows very high so the tool's own default never binds
 	// before max_chars.
 	input["max_rows"] = unboundedReadLimit
-	if n := runtime.Int("max-chars"); n > 0 {
+	if n, ok := maxCharsInput(runtime); ok {
 		input["max_chars"] = n
 	}
 	return input

shortcuts/sheets/lark_sheet_read_data_test.go

@@ -34,6 +34,41 @@ func TestReadDataShortcuts_DryRun(t *testing.T) {
 				"cell_limit":          float64(unboundedReadLimit), // pinned high; --max-chars is the only cap
 			},
 		},
+		{
+			name:     "+cells-get skip filtered rows only",
+			sc:       CellsGet,
+			args:     []string{"--url", testURL, "--sheet-id", testSheetID, "--range", "A1:B2", "--skip-filter"},
+			toolName: "get_cell_ranges",
+			wantInput: map[string]interface{}{
+				"skip_filter": true,
+			},
+		},
+		{
+			name:     "+cells-get skip hidden but keep filtered rows",
+			sc:       CellsGet,
+			args:     []string{"--url", testURL, "--sheet-id", testSheetID, "--range", "A1:B2", "--skip-hidden", "--skip-filter=false"},
+			toolName: "get_cell_ranges",
+			wantInput: map[string]interface{}{
+				"skip_hidden": true,
+				"skip_filter": false,
+			},
+		},
+		{
+			// --include truncation toggles include_truncation_info so the tool
+			// estimates and returns per-cell isRowTruncated / isColTruncated.
+			name:     "+cells-get include=truncation",
+			sc:       CellsGet,
+			args:     []string{"--url", testURL, "--sheet-id", testSheetID, "--range", "A1:B2", "--include", "truncation"},
+			toolName: "get_cell_ranges",
+			wantInput: map[string]interface{}{
+				"excel_id":                testToken,
+				"sheet_id":                testSheetID,
+				"ranges":                  []interface{}{"A1:B2"},
+				"include_styles":          false,
+				"include_truncation_info": true,
+				"cell_limit":              float64(unboundedReadLimit),
+			},
+		},
 		{
 			// Canonical form: --sheet-id + bare --range. Aligned with
 			// +cells-get / +csv-get; before the e2e BUG-019 fix this
@@ -74,6 +109,17 @@ func TestReadDataShortcuts_DryRun(t *testing.T) {
 	}
 }
 
+func TestCellsGet_OmitsSkipFilterWhenUnset(t *testing.T) {
+	t.Parallel()
+	body := parseDryRunBody(t, CellsGet, []string{
+		"--url", testURL, "--sheet-id", testSheetID, "--range", "A1:B2",
+	})
+	got := decodeToolInput(t, body, "get_cell_ranges")
+	if _, ok := got["skip_filter"]; ok {
+		t.Fatalf("skip_filter must be omitted when --skip-filter is unset: %#v", got)
+	}
+}
+
 // TestDropdownGet_RequiresSheetSelector locks the +cells-get-style
 // selector contract: at least one of --sheet-id / --sheet-name must be
 // supplied. Before BUG-019 fix this shortcut required a "Sheet!A1"
@@ -92,7 +138,9 @@ func TestDropdownGet_RequiresSheetSelector(t *testing.T) {
 
 // TestReadData_RequiresRange covers the trim-based --range guard on the
 // single-range readers (--range "" slips past cobra's MarkFlagRequired but
-// must still be rejected by Validate).
+// must still be rejected by Validate). +csv-get is deliberately absent:
+// its --range is optional — omitted/blank means a whole-sheet read (see
+// TestCsvGet_RangeOptionalDefaultsToFullSheet).
 func TestReadData_RequiresRange(t *testing.T) {
 	t.Parallel()
 	cases := []struct {
@@ -100,7 +148,6 @@ func TestReadData_RequiresRange(t *testing.T) {
 		sc   common.Shortcut
 	}{
 		{"+cells-get", CellsGet},
-		{"+csv-get", CsvGet},
 		{"+dropdown-get", DropdownGet},
 	}
 	for _, c := range cases {
@@ -114,6 +161,23 @@ func TestReadData_RequiresRange(t *testing.T) {
 	}
 }
 
+// TestCsvGet_RangeOptionalDefaultsToFullSheet pins the whole-sheet default:
+// with --range omitted the request carries the over-wide clip range, so a
+// full read needs no workbook-info pre-flight (eval: --range was the most
+// missed required flag on +csv-get once the rest of the surface settled).
+func TestCsvGet_RangeOptionalDefaultsToFullSheet(t *testing.T) {
+	t.Parallel()
+	stdout, _, err := runShortcutCapturingErr(t, CsvGet, []string{
+		"--url", testURL, "--sheet-id", testSheetID, "--dry-run",
+	})
+	if err != nil {
+		t.Fatalf("rangeless +csv-get must pass validation, got: %v", err)
+	}
+	if !strings.Contains(stdout, csvGetFullSheetRange) {
+		t.Fatalf("dry-run body should carry the full-sheet range %q, got %q", csvGetFullSheetRange, stdout)
+	}
+}
+
 // TestInfoTypeFromInclude exercises the fine-grained → coarse mapping
 // directly (white-box).
 func TestInfoTypeFromInclude(t *testing.T) {

shortcuts/sheets/lark_sheet_sheet_structure.go

@@ -6,6 +6,7 @@ package sheets
 import (
 	"context"
 	"fmt"
+	"sort"
 	"strconv"
 	"strings"
 
@@ -128,12 +129,20 @@ var DimInsert = common.Shortcut{
 	AuthTypes:   []string{"user", "bot"},
 	HasFormat:   true,
 	Flags:       flagsFor("+dim-insert"),
-	Validate:    validateViaInput(dimInsertInput),
+	Tips: []string{
+		"Example: lark-cli sheets +dim-insert --url <URL> --sheet-name Sheet1 --position 3 --count 2 --inherit-style before",
+		"Rows vs columns comes from --position alone: a row number (3) inserts rows, a column letter (C) inserts columns — there is no --dimension flag.",
+	},
+	Validate: validateViaInput(dimInsertInput),
 	DryRun: func(ctx context.Context, runtime *common.RuntimeContext) *common.DryRunAPI {
 		token, _ := resolveSpreadsheetToken(runtime)
 		sheetID, sheetName, _ := resolveSheetSelector(runtime)
 		input, _ := dimInsertInput(runtime, token, sheetID, sheetName)
-		return invokeToolDryRun(token, ToolKindWrite, "modify_sheet_structure", input)
+		dr := invokeToolDryRun(token, ToolKindWrite, "modify_sheet_structure", input)
+		if dimInsertNeedsBeforeStyleWarning(runtime) {
+			dr.Set("warning_message", dimInsertBeforeStyleWarning)
+		}
+		return dr
 	},
 	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
 		token, err := resolveSpreadsheetTokenExec(runtime)
@@ -148,6 +157,9 @@ var DimInsert = common.Shortcut{
 		if err != nil {
 			return err
 		}
+		if dimInsertNeedsBeforeStyleWarning(runtime) {
+			fmt.Fprintln(runtime.IO().ErrOut, dimInsertBeforeStyleWarning)
+		}
 		out, err := callTool(ctx, runtime, token, ToolKindWrite, "modify_sheet_structure", input)
 		if err != nil {
 			return err
@@ -157,8 +169,31 @@ var DimInsert = common.Shortcut{
 	},
 }
 
+// dimInsertBeforeStyleWarning fires only when the preceding-side style cannot
+// be copied: --inherit-style before at the first row/column, where no
+// preceding row/column exists. The row/column is still inserted before
+// --position, just without style inheritance. (--inherit-style after has no
+// such edge — a plain before-insert always has a following row/column.)
+const dimInsertBeforeStyleWarning = "warning: --inherit-style before cannot copy the preceding row/column's style at the first row/column (no preceding row/column exists); inserting before --position without style inheritance. Copy styles separately if needed."
+
+func dimInsertNeedsBeforeStyleWarning(runtime flagView) bool {
+	if !runtime.Changed("inherit-style") || runtime.Str("inherit-style") != "before" {
+		return false
+	}
+	// Only the first row/column (idx 0) has no preceding row/column.
+	_, idx, err := parseA1Position(strings.TrimSpace(runtime.Str("position")))
+	return err == nil && idx == 0
+}
+
 // dimInsertInput passes --position (1-based row number "3" or column letter
-// "C") straight to the tool's `position` field; --count maps to `count`.
+// "C") to the tool's `position` field; --count maps to `count`.
+//
+// +dim-insert's public contract is always "insert before --position";
+// --inherit-style only selects which side's style the new row/column copies,
+// never the insertion side. The sheet-ai tool always copies the *anchor*
+// column's style (the target passed as position), regardless of side — so
+// --inherit-style before is emulated by anchoring one unit earlier. See the
+// switch below.
 func dimInsertInput(runtime flagView, token, sheetID, sheetName string) (map[string]interface{}, error) {
 	if err := requireSheetSelector(sheetID, sheetName); err != nil {
 		return nil, err
@@ -184,11 +219,27 @@ func dimInsertInput(runtime flagView, token, sheetID, sheetName string) (map[str
 		"count":     count,
 	}
 	sheetSelectorForToolInput(input, sheetID, sheetName)
+	// --inherit-style selects which side's style the blank row/column copies;
+	// the insertion always lands *before* --position. Empirically the addCol
+	// backend copies the *anchor* column's style (the target passed as
+	// position), regardless of side — side only decides whether the blank lands
+	// before or after that anchor (verified live, see
+	// TestDimInsertInheritStyleSideMapping):
+	//   after  → side=before at P: the blank lands at P and anchor P becomes the
+	//            *following* neighbour, so the blank copies it. Position unchanged.
+	//   before → side=after at P-1: the blank still lands at P (insert-after-(P-1)
+	//            == insert-before-P) and anchor P-1 becomes the *preceding*
+	//            neighbour, so the blank copies it.
 	switch runtime.Str("inherit-style") {
-	case "before":
-		input["side"] = "before"
 	case "after":
-		input["side"] = "after"
+		input["side"] = "before"
+	case "before":
+		if prev, ok := a1PositionBefore(position); ok {
+			input["side"] = "after"
+			input["position"] = prev
+		}
+		// First row/column: no preceding row/column exists, so fall back to a
+		// plain before-insert (dimInsertNeedsBeforeStyleWarning surfaces this).
 	}
 	return input, nil
 }
@@ -203,10 +254,34 @@ var DimDelete = common.Shortcut{
 	AuthTypes:   []string{"user", "bot"},
 	HasFormat:   true,
 	Flags:       flagsFor("+dim-delete"),
-	Validate:    validateDimRangeOp("delete"),
+	Validate: func(ctx context.Context, runtime *common.RuntimeContext) error {
+		if runtime.Changed("ranges") {
+			if runtime.Changed("range") {
+				return sheetsValidationForFlag("ranges", "--range and --ranges are mutually exclusive; put every range into --ranges")
+			}
+			token, err := resolveSpreadsheetToken(runtime)
+			if err != nil {
+				return err
+			}
+			sheetID, sheetName, err := resolveSheetSelector(runtime)
+			if err != nil {
+				return err
+			}
+			_, err = dimDeleteRangesOps(runtime, token, sheetID, sheetName)
+			return err
+		}
+		return validateDimRangeOp("delete")(ctx, runtime)
+	},
 	DryRun: func(ctx context.Context, runtime *common.RuntimeContext) *common.DryRunAPI {
 		token, _ := resolveSpreadsheetToken(runtime)
 		sheetID, sheetName, _ := resolveSheetSelector(runtime)
+		if runtime.Changed("ranges") {
+			ops, _ := dimDeleteRangesOps(runtime, token, sheetID, sheetName)
+			return invokeToolDryRun(token, ToolKindWrite, "batch_update", map[string]interface{}{
+				"excel_id":   token,
+				"operations": ops,
+			})
+		}
 		input, _ := dimRangeOpInput(runtime, token, sheetID, sheetName, "delete")
 		return invokeToolDryRun(token, ToolKindWrite, "modify_sheet_structure", input)
 	},
@@ -219,6 +294,21 @@ var DimDelete = common.Shortcut{
 		if err != nil {
 			return err
 		}
+		if runtime.Changed("ranges") {
+			ops, err := dimDeleteRangesOps(runtime, token, sheetID, sheetName)
+			if err != nil {
+				return err
+			}
+			out, err := callTool(ctx, runtime, token, ToolKindWrite, "batch_update", map[string]interface{}{
+				"excel_id":   token,
+				"operations": ops,
+			})
+			if err != nil {
+				return err
+			}
+			runtime.Out(out, nil)
+			return nil
+		}
 		input, err := dimRangeOpInput(runtime, token, sheetID, sheetName, "delete")
 		if err != nil {
 			return err
@@ -232,9 +322,76 @@ var DimDelete = common.Shortcut{
 	},
 	Tips: []string{
 		"Row/column deletion is irreversible. Always preview with --dry-run first.",
+		`Scattered ranges: --ranges '["5:5","8:8","11:13"]' deletes them in one atomic call — the CLI orders positions descending, so indexes never shift under you.`,
 	},
 }
 
+// dimDeleteRangesOps parses --ranges into one atomic batch of
+// modify_sheet_structure delete ops, ordered DESCENDING by start position:
+// deleting an earlier row shifts every later index up, so ascending
+// execution deletes the wrong rows — the recurring failure of hand-built
+// dim-delete batches in eval traces. Same-dimension and non-overlap are
+// enforced up front.
+func dimDeleteRangesOps(runtime flagView, token, sheetID, sheetName string) ([]interface{}, error) {
+	if err := requireSheetSelector(sheetID, sheetName); err != nil {
+		return nil, err
+	}
+	raw, err := requireJSONArray(runtime, "ranges")
+	if err != nil {
+		return nil, err
+	}
+	if len(raw) == 0 {
+		return nil, sheetsValidationForFlag("ranges", "--ranges must be a non-empty JSON array")
+	}
+	if len(raw) > maxBatchRanges {
+		return nil, sheetsValidationForFlag("ranges", "--ranges accepts at most %d entries; got %d", maxBatchRanges, len(raw))
+	}
+	type span struct {
+		raw        string
+		start, end int
+	}
+	spans := make([]span, 0, len(raw))
+	dimension := ""
+	for i, v := range raw {
+		s, ok := v.(string)
+		if !ok {
+			return nil, sheetsValidationForFlag("ranges", "--ranges[%d] must be a string", i)
+		}
+		dim, start, end, err := parseA1Range(s)
+		if err != nil {
+			return nil, sheetsValidationForFlag("ranges", "--ranges[%d] %q: %v", i, s, err)
+		}
+		if dimension == "" {
+			dimension = dim
+		} else if dim != dimension {
+			return nil, sheetsValidationForFlag("ranges", "--ranges[%d] %q is a %s range but earlier entries are %s ranges; one call deletes rows OR columns, not both", i, s, dim, dimension)
+		}
+		spans = append(spans, span{raw: strings.TrimSpace(s), start: start, end: end})
+	}
+	sort.Slice(spans, func(i, j int) bool { return spans[i].start > spans[j].start })
+	for i := 1; i < len(spans); i++ {
+		// Descending order: spans[i-1] starts at or after spans[i]. Overlap
+		// (or duplicate) makes the later delete hit already-shifted positions.
+		if spans[i].end >= spans[i-1].start {
+			return nil, sheetsValidationForFlag("ranges", "--ranges entries %q and %q overlap; merge them into one range", spans[i].raw, spans[i-1].raw)
+		}
+	}
+	ops := make([]interface{}, 0, len(spans))
+	for _, sp := range spans {
+		input := map[string]interface{}{
+			"excel_id":  token,
+			"operation": "delete",
+			"range":     sp.raw,
+		}
+		sheetSelectorForToolInput(input, sheetID, sheetName)
+		ops = append(ops, map[string]interface{}{
+			"tool_name": "modify_sheet_structure",
+			"input":     input,
+		})
+	}
+	return ops, nil
+}
+
 // validateDimRangeOp returns a Validate closure that delegates to
 // dimRangeOpInput for shortcuts (delete/hide/unhide) whose builder takes an
 // extra `op` argument. Token check happens here; the rest is the builder.
@@ -292,7 +449,10 @@ var DimFreeze = common.Shortcut{
 	AuthTypes:   []string{"user", "bot"},
 	HasFormat:   true,
 	Flags:       flagsFor("+dim-freeze"),
-	Validate:    validateViaInput(dimFreezeInput),
+	Tips: []string{
+		"Example: lark-cli sheets +dim-freeze --url <URL> --sheet-name Sheet1 --dimension row --count 2 (freezes the first 2 rows; --count 0 unfreezes)",
+	},
+	Validate: validateViaInput(dimFreezeInput),
 	DryRun: func(ctx context.Context, runtime *common.RuntimeContext) *common.DryRunAPI {
 		token, _ := resolveSpreadsheetToken(runtime)
 		sheetID, sheetName, _ := resolveSheetSelector(runtime)
@@ -557,6 +717,23 @@ func columnIndexToLetter(idx int) string {
 	return string(out)
 }
 
+// a1PositionBefore returns the A1 position one unit before s ("6" → "5",
+// "C" → "B"), preserving row/column form. ok is false when s is the first
+// row/column (row 1 / column A) — no earlier position — or is not a valid A1
+// position. Callers validate via parseA1Position first, so in practice ok is
+// false only at the first row/column.
+func a1PositionBefore(s string) (pos string, ok bool) {
+	dimension, idx, err := parseA1Position(s)
+	if err != nil || idx == 0 {
+		return "", false
+	}
+	if dimension == "row" {
+		// idx is 0-based; the 1-based number one row earlier is idx itself.
+		return strconv.Itoa(idx), true
+	}
+	return columnIndexToLetter(idx - 1), true
+}
+
 // ─── +dim-move (native v3 move_dimension, cli_status: cli-only) ──────
 //
 // Moves a contiguous block of rows or columns to a new index in the same

shortcuts/sheets/lark_sheet_sheet_structure_test.go

@@ -48,6 +48,8 @@ func TestSheetStructureShortcuts_DryRun(t *testing.T) {
 			},
 		},
 		{
+			// --inherit-style before copies the preceding row: anchor row 5 and
+			// insert after it (side=after), so the blank still lands before row 6.
 			name:     "+dim-insert row position=6 count=3 inherit-before",
 			sc:       DimInsert,
 			args:     []string{"--url", testURL, "--sheet-id", testSheetID, "--position", "6", "--count", "3", "--inherit-style", "before"},
@@ -56,9 +58,9 @@ func TestSheetStructureShortcuts_DryRun(t *testing.T) {
 				"excel_id":  testToken,
 				"operation": "insert",
 				"sheet_id":  testSheetID,
-				"position":  "6",
+				"position":  "5",
 				"count":     float64(3),
-				"side":      "before",
+				"side":      "after",
 			},
 		},
 		{
@@ -169,6 +171,93 @@ func TestSheetStructureShortcuts_DryRun(t *testing.T) {
 	}
 }
 
+func TestDimInsertInheritStyleSideMapping(t *testing.T) {
+	t.Parallel()
+
+	cases := []struct {
+		name         string
+		position     string
+		inherit      string
+		wantPosition string
+		wantSide     string
+		wantSideSet  bool
+	}{
+		{
+			name:         "after copies the following style with a plain before-insert, position unchanged",
+			position:     "D",
+			inherit:      "after",
+			wantPosition: "D",
+			wantSide:     "before",
+			wantSideSet:  true,
+		},
+		{
+			name:         "before anchors one column earlier (side=after) to copy the preceding style",
+			position:     "D",
+			inherit:      "before",
+			wantPosition: "C",
+			wantSide:     "after",
+			wantSideSet:  true,
+		},
+		{
+			name:         "before on a row anchors one row earlier",
+			position:     "6",
+			inherit:      "before",
+			wantPosition: "5",
+			wantSide:     "after",
+			wantSideSet:  true,
+		},
+		{
+			name:         "before at the first column falls back to a plain before-insert",
+			position:     "A",
+			inherit:      "before",
+			wantPosition: "A",
+			wantSideSet:  false,
+		},
+		{
+			name:         "after at the first column still works (before-insert anchors the following)",
+			position:     "A",
+			inherit:      "after",
+			wantPosition: "A",
+			wantSide:     "before",
+			wantSideSet:  true,
+		},
+		{
+			name:         "default (flag omitted) omits side, backend inherits the following row/column",
+			position:     "D",
+			wantPosition: "D",
+			wantSideSet:  false,
+		},
+	}
+
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
+			args := []string{"--url", testURL, "--sheet-id", testSheetID, "--position", tc.position, "--count", "1"}
+			if tc.inherit != "" {
+				args = append(args, "--inherit-style", tc.inherit)
+			}
+			body := parseDryRunBody(t, DimInsert, args)
+			got := decodeToolInput(t, body, "modify_sheet_structure")
+			assertInputEquals(t, got, map[string]interface{}{
+				"excel_id":  testToken,
+				"operation": "insert",
+				"sheet_id":  testSheetID,
+				"position":  tc.wantPosition,
+				"count":     float64(1),
+			})
+
+			gv, ok := got["side"]
+			if ok != tc.wantSideSet {
+				t.Fatalf("side presence = %v, want %v (input=%#v)", ok, tc.wantSideSet, got)
+			}
+			if ok && gv != tc.wantSide {
+				t.Fatalf("side = %v, want %q", gv, tc.wantSide)
+			}
+		})
+	}
+}
+
 // TestDimRange_Validation covers the A1 range parser's edge cases routed
 // through +dim-hide (any --range shortcut works; we just need to exercise
 // the validator).

shortcuts/sheets/lark_sheet_styles_put.go

@@ -0,0 +1,272 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package sheets
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"strings"
+
+	"github.com/larksuite/cli/shortcuts/common"
+)
+
+// ─── +styles-put ──────────────────────────────────────────────────────
+//
+// Declarative visual spec for EXISTING spreadsheets. Eval attribution
+// showed ~73% of real +batch-update calls were pure formatting finishers
+// (style stamps + merges + resizes + freeze) hand-assembled as imperative
+// operations arrays — the top error surface. +styles-put replaces that
+// with the {styles:[...]} protocol already shared by +workbook-create /
+// +table-put --styles (identical vocabulary, parsed by the same
+// parseWorkbookCreateStyleItem), applied to a live workbook and expanded
+// client-side into ONE atomic batch_update.
+//
+// Per-sheet expansion order (server behavior verified live: style stamps
+// over merged regions are allowed — the top-left-only restriction applies
+// to value writes, not styles):
+//
+//	cell_merges → cell_styles → row_sizes → col_sizes → freeze
+var StylesPut = common.Shortcut{
+	Service:     "sheets",
+	Command:     "+styles-put",
+	Description: "Apply one declarative visual spec (styles/merges/row-col sizes/freeze) to existing sheets in one atomic batch.",
+	Risk:        "write",
+	Scopes:      []string{"sheets:spreadsheet:write_only"},
+	AuthTypes:   []string{"user", "bot"},
+	HasFormat:   true,
+	Flags:       flagsFor("+styles-put"),
+	Tips: []string{
+		`Example: lark-cli sheets +styles-put --url <URL> --styles '{"styles":[{"name":"Sheet1","cell_styles":[{"range":"A1:F1","font_weight":"bold"}],"freeze":{"rows":1}}]}'`,
+		"Same --styles vocabulary as +workbook-create / +table-put; one item per target sheet, name = the real sheet name.",
+		"Style stamps are safe to re-run; the whole spec goes out as one atomic batch.",
+	},
+	Validate: func(ctx context.Context, runtime *common.RuntimeContext) error {
+		token, err := resolveSpreadsheetToken(runtime)
+		if err != nil {
+			return err
+		}
+		_, err = stylesPutOperations(runtime, token)
+		return err
+	},
+	DryRun: func(ctx context.Context, runtime *common.RuntimeContext) *common.DryRunAPI {
+		token, _ := resolveSpreadsheetToken(runtime)
+		ops, _ := stylesPutOperations(runtime, token)
+		return invokeToolDryRun(token, ToolKindWrite, "batch_update", map[string]interface{}{
+			"excel_id":   token,
+			"operations": ops,
+		})
+	},
+	Execute: func(ctx context.Context, runtime *common.RuntimeContext) error {
+		token, err := resolveSpreadsheetTokenExec(runtime)
+		if err != nil {
+			return err
+		}
+		ops, err := stylesPutOperations(runtime, token)
+		if err != nil {
+			return err
+		}
+		out, err := callTool(ctx, runtime, token, ToolKindWrite, "batch_update", map[string]interface{}{
+			"excel_id":   token,
+			"operations": ops,
+		})
+		if err != nil {
+			return err
+		}
+		runtime.Out(out, nil)
+		return nil
+	},
+}
+
+// stylesPutOperations parses --styles ({styles:[...]}, one item per target
+// sheet) and expands it into the MCP batch_update operations array. Reuses
+// the shared workbook-create style item parser, so field validation, alias
+// normalization (border "all" shorthand, style vocabulary) and the
+// aggregate-all-issues error shape are identical across the three --styles
+// carriers.
+func stylesPutOperations(runtime flagView, token string) ([]interface{}, error) {
+	if strings.TrimSpace(runtime.Str("styles")) == "" {
+		return nil, sheetsValidationForFlag("styles", "--styles is required")
+	}
+	v, err := parseJSONFlag(runtime, "styles")
+	if err != nil {
+		return nil, err
+	}
+	items, err := parseWorkbookCreateStylesItems(v)
+	if err != nil {
+		return nil, err
+	}
+	if len(items) == 0 {
+		return nil, sheetsValidationForFlag("styles", "--styles.styles must be a non-empty array (one item per target sheet)")
+	}
+	var probs []error
+	type sheetSpec struct {
+		name    string
+		payload *workbookCreateStylePayload
+	}
+	specs := make([]sheetSpec, 0, len(items))
+	seenName := map[string]bool{}
+	for i, item := range items {
+		path := fmt.Sprintf("--styles.styles[%d]", i)
+		name, _ := item["name"].(string)
+		name = strings.TrimSpace(name)
+		if name == "" {
+			probs = append(probs, common.ValidationErrorf("%s.name is required (the real sheet name; check +workbook-info)", path))
+			continue
+		}
+		if seenName[name] {
+			probs = append(probs, common.ValidationErrorf("%s.name %q appears twice; merge the two items", path, name))
+			continue
+		}
+		seenName[name] = true
+		payload, itemProbs := parseWorkbookCreateStyleItem(item, path)
+		if len(itemProbs) > 0 {
+			probs = append(probs, itemProbs...)
+			continue
+		}
+		specs = append(specs, sheetSpec{name: name, payload: payload})
+	}
+	if err := joinStyleValidationErrors(probs); err != nil {
+		return nil, err
+	}
+
+	ops := make([]interface{}, 0, len(specs)*4)
+	var totalCells int64
+	appendVisual := func(name string, op workbookCreateStyleOp) {
+		input, toolName := workbookCreateVisualOpInput(token, "", name, op)
+		if toolName == "" {
+			return
+		}
+		ops = append(ops, map[string]interface{}{"tool_name": toolName, "input": input})
+	}
+	for _, spec := range specs {
+		// merges first so subsequent style stamps see the final grid.
+		for _, m := range spec.payload.CellMerges {
+			appendVisual(spec.name, workbookCreateStyleOp{Kind: "cell_merge", Range: m.Range, MergeType: m.MergeType})
+		}
+		for _, cs := range coalesceStyleStamps(spec.payload.CellStyles) {
+			rows, cols, err := rangeDimensions(cs.Range)
+			if err != nil {
+				return nil, sheetsValidationForFlag("styles", "cell_styles range %q: %v", cs.Range, err)
+			}
+			if err := checkStampMatrixBudget("styles", cs.Range, rows, cols); err != nil {
+				return nil, err
+			}
+			totalCells += int64(rows) * int64(cols)
+			if err := checkBatchStampBudget(totalCells); err != nil {
+				return nil, err
+			}
+			ops = append(ops, map[string]interface{}{
+				"tool_name": "set_cell_range",
+				"input": map[string]interface{}{
+					"excel_id":   token,
+					"sheet_name": spec.name,
+					"range":      stripSheetPrefix(cs.Range),
+					"cells":      fillCellsMatrix(rows, cols, cs.Style),
+				},
+			})
+		}
+		for _, rs := range spec.payload.RowSizes {
+			appendVisual(spec.name, workbookCreateStyleOp{Kind: "row_size", Range: rs.Range, ResizeType: rs.ResizeType, Size: rs.Size})
+		}
+		for _, csz := range spec.payload.ColSizes {
+			appendVisual(spec.name, workbookCreateStyleOp{Kind: "col_size", Range: csz.Range, ResizeType: csz.ResizeType, Size: csz.Size})
+		}
+		if f := spec.payload.Freeze; f != nil {
+			if f.Rows > 0 {
+				appendVisual(spec.name, workbookCreateStyleOp{Kind: "freeze_rows", Size: f.Rows})
+			}
+			if f.Cols > 0 {
+				appendVisual(spec.name, workbookCreateStyleOp{Kind: "freeze_cols", Size: f.Cols})
+			}
+		}
+	}
+	if len(ops) > maxBatchOperations {
+		return nil, sheetsValidationForFlag("styles",
+			"--styles expands to %d operations even after merging adjacent same-style ranges, over the %d cap; split the spec into several +styles-put calls — and for alternating-row banding or value-dependent coloring use +cond-format-create instead of per-row stamps",
+			len(ops), maxBatchOperations)
+	}
+	return ops, nil
+}
+
+// coalesceStyleStamps merges cell_styles entries that carry the IDENTICAL
+// style into larger rectangles: same column span + contiguous/overlapping
+// rows fuse vertically, same row span + contiguous columns fuse
+// horizontally, iterated to a fixpoint. Models routinely emit one entry per
+// row (07-21 rerun: specs expanding to 184/203/861 operations against the
+// 100-op cap); a declarative spec describes intent, so execution shape is
+// the CLI's to optimize. Entries with unparsable ranges pass through
+// untouched (the per-op validation reports them with proper context).
+func coalesceStyleStamps(ops []workbookCreateCellStyleOp) []workbookCreateCellStyleOp {
+	if len(ops) < 2 {
+		return ops
+	}
+	type rect struct{ c1, r1, c2, r2 int }
+	type group struct {
+		style map[string]interface{}
+		rects []rect
+	}
+	var order []string
+	groups := map[string]*group{}
+	out := make([]workbookCreateCellStyleOp, 0, len(ops))
+	for _, op := range ops {
+		c1, r1, c2, r2, err := workbookCreateStyleRangeBounds(op.Range)
+		key, jerr := json.Marshal(op.Style) // map keys marshal sorted → canonical
+		if err != nil || jerr != nil {
+			out = append(out, op)
+			continue
+		}
+		g, ok := groups[string(key)]
+		if !ok {
+			g = &group{style: op.Style}
+			groups[string(key)] = g
+			order = append(order, string(key))
+		}
+		g.rects = append(g.rects, rect{c1, r1, c2, r2})
+	}
+	for _, key := range order {
+		g := groups[key]
+		rects := g.rects
+		for changed := true; changed; {
+			changed = false
+			for i := 0; i < len(rects) && !changed; i++ {
+				for j := i + 1; j < len(rects); j++ {
+					a, b := rects[i], rects[j]
+					var merged rect
+					switch {
+					case a.c1 == b.c1 && a.c2 == b.c2 && b.r1 <= a.r2+1 && a.r1 <= b.r2+1:
+						merged = rect{a.c1, min(a.r1, b.r1), a.c2, max(a.r2, b.r2)}
+					case a.r1 == b.r1 && a.r2 == b.r2 && b.c1 <= a.c2+1 && a.c1 <= b.c2+1:
+						merged = rect{min(a.c1, b.c1), a.r1, max(a.c2, b.c2), a.r2}
+					default:
+						continue
+					}
+					rects[i] = merged
+					rects = append(rects[:j], rects[j+1:]...)
+					changed = true
+					break
+				}
+			}
+		}
+		for _, rc := range rects {
+			out = append(out, workbookCreateCellStyleOp{
+				Range: fmt.Sprintf("%s%d:%s%d",
+					columnIndexToLetter(rc.c1), rc.r1+1,
+					columnIndexToLetter(rc.c2), rc.r2+1),
+				Style: g.style,
+			})
+		}
+	}
+	return out
+}
+
+// stripSheetPrefix drops an optional "Sheet!"-style prefix from an A1 range:
+// the target sheet is already carried by the spec item's name, and the
+// batch sub-op input names the sheet separately.
+func stripSheetPrefix(rangeStr string) string {
+	if idx := strings.Index(rangeStr, "!"); idx >= 0 {
+		return strings.TrimSpace(rangeStr[idx+1:])
+	}
+	return strings.TrimSpace(rangeStr)
+}

shortcuts/sheets/lark_sheet_styles_put_test.go

@@ -0,0 +1,347 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package sheets
+
+import (
+	"strings"
+	"testing"
+)
+
+func stylesPutView(spec map[string]interface{}) mapFlagView {
+	return newMapFlagViewForCommand("+styles-put", map[string]interface{}{"styles": spec})
+}
+
+// TestStylesPutOperations_ExpansionOrder pins the per-sheet expansion:
+// cell_merges → cell_styles → row_sizes → col_sizes → freeze, all inside one
+// batch_update operations array (server-side order dependence verified live).
+func TestStylesPutOperations_ExpansionOrder(t *testing.T) {
+	t.Parallel()
+	ops, err := stylesPutOperations(stylesPutView(map[string]interface{}{
+		"styles": []interface{}{map[string]interface{}{
+			"name":        "Sheet1",
+			"cell_merges": []interface{}{map[string]interface{}{"range": "A5:A8"}},
+			"cell_styles": []interface{}{map[string]interface{}{"range": "A1:B1", "font_weight": "bold"}},
+			"row_sizes":   []interface{}{map[string]interface{}{"range": "1:1", "type": "pixel", "size": float64(36)}},
+			"col_sizes":   []interface{}{map[string]interface{}{"range": "A:B", "type": "pixel", "size": float64(120)}},
+			"freeze":      map[string]interface{}{"rows": float64(1), "cols": float64(2)},
+		}},
+	}), testToken)
+	if err != nil {
+		t.Fatalf("unexpected error: %v", err)
+	}
+	wantTools := []string{"merge_cells", "set_cell_range", "resize_range", "resize_range", "modify_sheet_structure", "modify_sheet_structure"}
+	if len(ops) != len(wantTools) {
+		t.Fatalf("got %d ops, want %d", len(ops), len(wantTools))
+	}
+	for i, want := range wantTools {
+		op := ops[i].(map[string]interface{})
+		if op["tool_name"] != want {
+			t.Fatalf("ops[%d].tool_name = %v, want %s", i, op["tool_name"], want)
+		}
+		input := op["input"].(map[string]interface{})
+		if input["sheet_name"] != "Sheet1" {
+			t.Fatalf("ops[%d] missing sheet_name: %v", i, input)
+		}
+		if input["excel_id"] != testToken {
+			t.Fatalf("ops[%d] missing excel_id", i)
+		}
+	}
+	// The style stamp carries a cells matrix matching the range (1×2).
+	stamp := ops[1].(map[string]interface{})["input"].(map[string]interface{})
+	cells := stamp["cells"].([][]interface{})
+	if len(cells) != 1 || len(cells[0]) != 2 {
+		t.Fatalf("style stamp matrix = %dx%d, want 1x2", len(cells), len(cells[0]))
+	}
+	// Freeze ops carry the freeze counts.
+	fr := ops[4].(map[string]interface{})["input"].(map[string]interface{})
+	if fr["operation"] != "freeze" || fr["freeze_rows"] != 1 {
+		t.Fatalf("freeze rows op = %v", fr)
+	}
+	fc := ops[5].(map[string]interface{})["input"].(map[string]interface{})
+	if fc["freeze_columns"] != 2 {
+		t.Fatalf("freeze cols op = %v", fc)
+	}
+}
+
+// TestStylesPutOperations_Validation pins the aggregate error shape and the
+// section/name requirements.
+func TestStylesPutOperations_Validation(t *testing.T) {
+	t.Parallel()
+
+	t.Run("missing name and empty item aggregate", func(t *testing.T) {
+		t.Parallel()
+		_, err := stylesPutOperations(stylesPutView(map[string]interface{}{
+			"styles": []interface{}{
+				map[string]interface{}{"cell_styles": []interface{}{map[string]interface{}{"range": "A1", "font_weight": "bold"}}},
+				map[string]interface{}{"name": "S2"},
+			},
+		}), testToken)
+		ve := requireValidation(t, err, "name is required")
+		if !strings.Contains(ve.Message, "at least one of cell_styles/row_sizes/col_sizes/cell_merges/freeze") {
+			t.Fatalf("message %q missing empty-item issue", ve.Message)
+		}
+	})
+
+	t.Run("duplicate sheet name rejected", func(t *testing.T) {
+		t.Parallel()
+		item := map[string]interface{}{"name": "S1", "freeze": map[string]interface{}{"rows": float64(1)}}
+		item2 := map[string]interface{}{"name": "S1", "freeze": map[string]interface{}{"rows": float64(2)}}
+		_, err := stylesPutOperations(stylesPutView(map[string]interface{}{
+			"styles": []interface{}{item, item2},
+		}), testToken)
+		requireValidation(t, err, "appears twice")
+	})
+
+	t.Run("freeze-only item is valid", func(t *testing.T) {
+		t.Parallel()
+		ops, err := stylesPutOperations(stylesPutView(map[string]interface{}{
+			"styles": []interface{}{map[string]interface{}{"name": "S1", "freeze": map[string]interface{}{"rows": float64(1)}}},
+		}), testToken)
+		if err != nil || len(ops) != 1 {
+			t.Fatalf("ops=%d err=%v", len(ops), err)
+		}
+	})
+
+	t.Run("all-zero freeze rejected", func(t *testing.T) {
+		t.Parallel()
+		_, err := stylesPutOperations(stylesPutView(map[string]interface{}{
+			"styles": []interface{}{map[string]interface{}{"name": "S1", "freeze": map[string]interface{}{"rows": float64(0)}}},
+		}), testToken)
+		requireValidation(t, err, "at least one dimension")
+	})
+}
+
+// TestStylesPayloadVocabularyForgiveness pins the 07-20 rerun fixes: the
+// payload path (--styles cell_styles objects) accepts the same habitual
+// vocabulary the flag path already normalized — border family folding, wrap
+// aliases, and enum VALUE canonicalization (CSS center → Lark middle etc.).
+func TestStylesPayloadVocabularyForgiveness(t *testing.T) {
+	t.Parallel()
+
+	stamp := func(styleFields map[string]interface{}) ([]interface{}, error) {
+		item := map[string]interface{}{"range": "A1:B1"}
+		for k, v := range styleFields {
+			item[k] = v
+		}
+		return stylesPutOperations(stylesPutView(map[string]interface{}{
+			"styles": []interface{}{map[string]interface{}{
+				"name":        "S1",
+				"cell_styles": []interface{}{item},
+			}},
+		}), testToken)
+	}
+	cellProto := func(t *testing.T, ops []interface{}) map[string]interface{} {
+		t.Helper()
+		input := ops[0].(map[string]interface{})["input"].(map[string]interface{})
+		cells := input["cells"].([][]interface{})
+		return cells[0][0].(map[string]interface{})
+	}
+
+	t.Run("vertical_alignment center canonicalizes to middle", func(t *testing.T) {
+		t.Parallel()
+		ops, err := stamp(map[string]interface{}{"vertical_alignment": "center", "font_weight": "BOLD"})
+		if err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+		cs := cellProto(t, ops)["cell_styles"].(map[string]interface{})
+		if cs["vertical_alignment"] != "middle" || cs["font_weight"] != "bold" {
+			t.Fatalf("cell_styles = %v, want middle/bold", cs)
+		}
+	})
+
+	t.Run("off-enum value rejected client-side with did-you-mean", func(t *testing.T) {
+		t.Parallel()
+		_, err := stamp(map[string]interface{}{"vertical_alignment": "botom"})
+		requireValidation(t, err, `did you mean "bottom"`)
+	})
+
+	t.Run("boolean wrap_text folds to word_wrap auto-wrap", func(t *testing.T) {
+		t.Parallel()
+		ops, err := stamp(map[string]interface{}{"wrap_text": true})
+		if err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+		cs := cellProto(t, ops)["cell_styles"].(map[string]interface{})
+		if cs["word_wrap"] != "auto-wrap" {
+			t.Fatalf("word_wrap = %v, want auto-wrap", cs["word_wrap"])
+		}
+	})
+
+	t.Run("borders object folds into border_styles", func(t *testing.T) {
+		t.Parallel()
+		ops, err := stamp(map[string]interface{}{
+			"borders": map[string]interface{}{"style": "solid", "color": "#000000"},
+		})
+		if err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+		bs := cellProto(t, ops)["border_styles"].(map[string]interface{})
+		top, _ := bs["top"].(map[string]interface{})
+		if top == nil || top["style"] != "solid" {
+			t.Fatalf("border_styles = %v, want all-sides solid", bs)
+		}
+	})
+
+	t.Run("flattened border_bottom and border_top_color fold per side", func(t *testing.T) {
+		t.Parallel()
+		ops, err := stamp(map[string]interface{}{
+			"border_bottom":    map[string]interface{}{"style": "solid"},
+			"border_top_color": "#FF0000",
+		})
+		if err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+		bs := cellProto(t, ops)["border_styles"].(map[string]interface{})
+		bottom, _ := bs["bottom"].(map[string]interface{})
+		topSide, _ := bs["top"].(map[string]interface{})
+		if bottom["style"] != "solid" || topSide["color"] != "#FF0000" {
+			t.Fatalf("border_styles = %v", bs)
+		}
+	})
+
+	t.Run("border_style thin means thin solid line", func(t *testing.T) {
+		t.Parallel()
+		ops, err := stamp(map[string]interface{}{"border_style": "thin"})
+		if err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+		bs := cellProto(t, ops)["border_styles"].(map[string]interface{})
+		top, _ := bs["top"].(map[string]interface{})
+		if top["weight"] != "thin" || top["style"] != "solid" {
+			t.Fatalf("border_styles.top = %v, want thin solid", top)
+		}
+	})
+
+	t.Run("fore_color prescribes instead of guessing", func(t *testing.T) {
+		t.Parallel()
+		_, err := stamp(map[string]interface{}{"fore_color": "#FF0000"})
+		requireValidation(t, err, "fore_color is ambiguous")
+	})
+
+	t.Run("bare string cell_merges accepted", func(t *testing.T) {
+		t.Parallel()
+		ops, err := stylesPutOperations(stylesPutView(map[string]interface{}{
+			"styles": []interface{}{map[string]interface{}{
+				"name":        "S1",
+				"cell_merges": []interface{}{"A5:B6"},
+			}},
+		}), testToken)
+		if err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+		input := ops[0].(map[string]interface{})["input"].(map[string]interface{})
+		if input["range"] != "A5:B6" || input["merge_type"] != "all" {
+			t.Fatalf("merge op = %v", input)
+		}
+	})
+}
+
+// TestStylesResizeSizeAliases pins the one-way Excel-vocabulary aliases on
+// the shared styles resize parser: height in row_sizes / width in col_sizes
+// resolve to size silently; the wrong dimension's word is a targeted error.
+func TestStylesResizeSizeAliases(t *testing.T) {
+	t.Parallel()
+
+	t.Run("height aliases to size in row_sizes", func(t *testing.T) {
+		t.Parallel()
+		ops, err := stylesPutOperations(stylesPutView(map[string]interface{}{
+			"styles": []interface{}{map[string]interface{}{
+				"name":      "S1",
+				"row_sizes": []interface{}{map[string]interface{}{"range": "1:1", "type": "pixel", "height": float64(36)}},
+			}},
+		}), testToken)
+		if err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+		input := ops[0].(map[string]interface{})["input"].(map[string]interface{})
+		block := input["resize_height"].(map[string]interface{})
+		if block["value"] != 36 {
+			t.Fatalf("resize_height = %v, want value 36", block)
+		}
+	})
+
+	t.Run("width aliases to size in col_sizes", func(t *testing.T) {
+		t.Parallel()
+		_, err := stylesPutOperations(stylesPutView(map[string]interface{}{
+			"styles": []interface{}{map[string]interface{}{
+				"name":      "S1",
+				"col_sizes": []interface{}{map[string]interface{}{"range": "A:C", "type": "pixel", "width": float64(120)}},
+			}},
+		}), testToken)
+		if err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+	})
+
+	t.Run("wrong-dimension word is a targeted error", func(t *testing.T) {
+		t.Parallel()
+		_, err := stylesPutOperations(stylesPutView(map[string]interface{}{
+			"styles": []interface{}{map[string]interface{}{
+				"name":      "S1",
+				"row_sizes": []interface{}{map[string]interface{}{"range": "1:1", "type": "pixel", "width": float64(36)}},
+			}},
+		}), testToken)
+		requireValidation(t, err, "does not apply to this array")
+	})
+
+	t.Run("size plus alias together rejected", func(t *testing.T) {
+		t.Parallel()
+		_, err := stylesPutOperations(stylesPutView(map[string]interface{}{
+			"styles": []interface{}{map[string]interface{}{
+				"name":      "S1",
+				"row_sizes": []interface{}{map[string]interface{}{"range": "1:1", "type": "pixel", "size": float64(36), "height": float64(40)}},
+			}},
+		}), testToken)
+		requireValidation(t, err, "either size or height")
+	})
+}
+
+// TestDimDeleteRangesOps pins the descending-order expansion and the
+// same-dimension / non-overlap guards.
+func TestDimDeleteRangesOps(t *testing.T) {
+	t.Parallel()
+
+	view := func(ranges ...interface{}) mapFlagView {
+		return newMapFlagViewForCommand("+dim-delete", map[string]interface{}{"ranges": ranges})
+	}
+
+	t.Run("rows execute descending", func(t *testing.T) {
+		t.Parallel()
+		ops, err := dimDeleteRangesOps(view("5:5", "11:13", "8:8"), testToken, "", "S1")
+		if err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+		var got []string
+		for _, op := range ops {
+			got = append(got, op.(map[string]interface{})["input"].(map[string]interface{})["range"].(string))
+		}
+		want := []string{"11:13", "8:8", "5:5"}
+		for i := range want {
+			if got[i] != want[i] {
+				t.Fatalf("order = %v, want %v", got, want)
+			}
+		}
+	})
+
+	t.Run("mixed dimensions rejected", func(t *testing.T) {
+		t.Parallel()
+		_, err := dimDeleteRangesOps(view("5:5", "C:C"), testToken, "", "S1")
+		requireValidation(t, err, "rows OR columns")
+	})
+
+	t.Run("overlap rejected", func(t *testing.T) {
+		t.Parallel()
+		_, err := dimDeleteRangesOps(view("5:8", "7:9"), testToken, "", "S1")
+		requireValidation(t, err, "overlap")
+	})
+
+	t.Run("ranges cannot nest inside batch", func(t *testing.T) {
+		t.Parallel()
+		_, err := translateBatchOp(subOp("+dim-delete", map[string]interface{}{
+			"sheet_name": "S1",
+			"ranges":     []interface{}{"5:5", "8:8"},
+		}), testToken, 0)
+		requireValidation(t, err, "not supported inside +batch-update")
+	})
+}

shortcuts/sheets/lark_sheet_table_io.go

@@ -88,6 +88,7 @@ var TablePut = common.Shortcut{
 		return tablePutWrite(ctx, runtime, token, payload, styles)
 	},
 	Tips: []string{
+		`Example: lark-cli sheets +table-put --url <URL> --sheets '{"sheets":[{"name":"S1","columns":["City","Rev"],"dtypes":{"Rev":"float64"},"data":[["SH",1234.5]]}]}'`,
 		"Writes into an existing spreadsheet — pass --url or --spreadsheet-token. To create a new workbook first, use +workbook-create, then point --spreadsheet-token here.",
 		"Payload sheets are matched to existing sub-sheets by name (created when absent). Date columns take ISO yyyy-mm-dd strings — converted to real dates (serial + date format).",
 		"--styles applies number formats, colors, merges, and row/col sizes in the same call (same shape as +workbook-create's --styles): one styles item per written sheet, name-matched. Skips the separate +cells-set-style round-trip.",
@@ -241,6 +242,11 @@ func decoderExpectEOF(dec *json.Decoder) error {
 	return nil
 }
 
+// tablePutSheetsSkeleton is the one-line --sheets shape inlined on a decode
+// error, so the retry needs no --print-schema round trip. Field vocabulary
+// mirrors tableSheetIn.
+const tablePutSheetsSkeleton = `{"sheets":[{"name":"Sheet1","columns":["City","Revenue"],"dtypes":{"Revenue":"float64"},"data":[["SH",123.4],["BJ",56.7]],"start_cell":"A1"}]}`
+
 // parseTablePutPayload reads --sheets (JSON, supports @file / stdin) into a
 // validated payload. UseNumber keeps numeric cells as json.Number so large
 // integers (order IDs, etc.) survive without precision loss or scientific
@@ -259,7 +265,19 @@ func parseTablePutPayload(runtime flagView) (*tablePayload, error) {
 		Sheets []tableSheetIn `json:"sheets"`
 	}
 	if err := dec.Decode(&wire); err != nil {
-		return nil, common.ValidationErrorf("--sheets: invalid JSON: %v", err).WithCause(err)
+		// Eval traces show two distinct decode failures that each burned
+		// retries: a field with the wrong JSON kind (columns as objects,
+		// dtypes as an array) — fixed by seeing the expected shape once —
+		// and shell-mangled JSON, fixed by moving the payload to stdin/@file.
+		verr := common.ValidationErrorf("--sheets: invalid JSON: %v", err).WithCause(err)
+		var ute *json.UnmarshalTypeError
+		if errors.As(err, &ute) {
+			return nil, verr.WithHint(
+				"expected shape: %s (columns is a flat string array; dtypes/formats are column-name-keyed maps; data is row-major)",
+				tablePutSheetsSkeleton)
+		}
+		return nil, verr.WithHint(
+			"if the payload contains formulas / quotes / commas, pass it via stdin (`--sheets - < file`) or a relative @file (`--sheets @./payload.json`)")
 	}
 	// Reject trailing non-whitespace after the first JSON value: json.Decoder
 	// accepts it silently (unlike json.Unmarshal), so e.g. `--sheets '{...} oops'`
@@ -1208,12 +1226,11 @@ var TableGet = common.Shortcut{
 			}
 			sheets = append(sheets, spec)
 		}
-		runtime.Out(map[string]interface{}{"sheets": sheets}, nil)
-		return nil
+		return emitReadResult(runtime, map[string]interface{}{"sheets": sheets})
 	},
 	Tips: []string{
 		"Output is the same shape +table-put consumes — pipe it back in, or load sheets[].rows into a DataFrame keyed by columns[].name.",
-		"Column types are inferred per column, but only when every non-empty cell agrees; a column mixing types (e.g. numbers + \"暂无\") degrades to string — lossless and round-trips cleanly. Numeric coercion of dirty cells is the caller's job (pandas to_numeric(errors=\"coerce\") on the string column).",
+		"Column types are inferred per column, but only when every non-empty cell agrees; a column mixing types (e.g. numbers + \"N/A\") degrades to string — lossless and round-trips cleanly. Numeric coercion of dirty cells is the caller's job (pandas to_numeric(errors=\"coerce\") on the string column).",
 	},
 }
 
@@ -1354,11 +1371,18 @@ func readSheetAsSpec(ctx context.Context, runtime *common.RuntimeContext, token
 		"value_render_option": "raw_value",
 		"cell_limit":          unboundedReadLimit,
 	}
+	// --max-chars binds the char budget (default 500000); --output-path lifts it
+	// to unbounded. Without this the tool applied its own ~50000 default and
+	// silently dropped rows past it with no signal in the +table-get output.
+	if n, ok := maxCharsInput(runtime); ok {
+		input["max_chars"] = n
+	}
 	sheetSelectorForToolInput(input, t.id, t.name)
 	out, err := callTool(ctx, runtime, token, ToolKindRead, "get_cell_ranges", input)
 	if err != nil {
 		return nil, err
 	}
+	truncated := cellRangesTruncated(out)
 	grid := extractCellGrid(out)
 	if len(grid) == 0 {
 		return emptySpec(), nil
@@ -1433,9 +1457,38 @@ func readSheetAsSpec(ctx context.Context, runtime *common.RuntimeContext, token
 	if len(formats) > 0 {
 		spec["formats"] = formats
 	}
+	// The tool clipped the read at max_chars: rows past the cap are missing from
+	// data. Surface it so the caller doesn't mistake a partial read for the whole
+	// sheet — re-run with --output-path (unlimited) or a higher --max-chars.
+	if truncated {
+		spec["truncated"] = true
+		spec["truncation_warning"] = "Result truncated by max_chars; rows past the cap were not returned. Best: re-run with --output-path to dump the whole sheet in one lossless pass (no cap). Alternatively raise --max-chars, or continue-read the remaining rows by passing --range for them — but that needs --no-header and you must reattach the header row and reconcile per-chunk dtypes yourself (this chunk's types were inferred from the rows returned here)."
+	}
 	return spec, nil
 }
 
+// cellRangesTruncated reports whether a get_cell_ranges response was clipped by
+// max_chars — either the top-level has_more flag or the first range's truncated
+// flag. Used by +table-get, whose spec output otherwise drops both signals.
+func cellRangesTruncated(out interface{}) bool {
+	m, ok := out.(map[string]interface{})
+	if !ok {
+		return false
+	}
+	if hm, ok := m["has_more"].(bool); ok && hm {
+		return true
+	}
+	ranges, _ := m["ranges"].([]interface{})
+	if len(ranges) > 0 {
+		if r0, ok := ranges[0].(map[string]interface{}); ok {
+			if t, ok := r0["truncated"].(bool); ok {
+				return t
+			}
+		}
+	}
+	return false
+}
+
 // sheetCurrentRegion returns the A1 range covering the sheet's existing data,
 // or "" for an empty sheet.
 //
@@ -1522,7 +1575,7 @@ func readCellFormat(cell map[string]interface{}) string {
 // inferColumnType decides a column's type from its data cells: a date
 // number_format guides each cell's type, but a column is given a non-string type
 // only when EVERY non-empty cell agrees. Real sheet columns often mix types (a
-// number column with a stray "暂无", a date column with a bare count); declaring
+// number column with a stray "N/A", a date column with a bare count); declaring
 // number/date while a string value rides along makes the output inconsistent —
 // it breaks round-trip back into +table-put (which rejects a string in a number
 // column) and crashes pandas astype. So a mixed column degrades to string

shortcuts/sheets/lark_sheet_table_io_test.go

@@ -1140,7 +1140,7 @@ func TestTableGet_InferColumnType(t *testing.T) {
 	// Mixed number+text degrades to string (self-consistent: every value is then
 	// a string), so the column round-trips and pandas doesn't choke. Numeric
 	// coercion of the dirty cells is left to the caller (pandas to_numeric).
-	if typ, _ := inferColumnType(col(mk(100.0, ""), mk("暂无", ""), mk(200.0, "")), 0); typ != "string" {
+	if typ, _ := inferColumnType(col(mk(100.0, ""), mk("N/A", ""), mk(200.0, "")), 0); typ != "string" {
 		t.Errorf("mixed number+text col → %s, want string", typ)
 	}
 	// A bare number mixed into a date column must NOT stay date (would serial-

shortcuts/sheets/lark_sheet_workbook.go

@@ -13,6 +13,7 @@ import (
 
 	"github.com/larksuite/cli/errs"
 	"github.com/larksuite/cli/extension/fileio"
+	"github.com/larksuite/cli/internal/suggest"
 	"github.com/larksuite/cli/internal/util"
 	"github.com/larksuite/cli/shortcuts/common"
 	"github.com/larksuite/cli/shortcuts/drive"
@@ -405,7 +406,11 @@ var SheetCopy = common.Shortcut{
 	AuthTypes:   []string{"user", "bot"},
 	HasFormat:   true,
 	Flags:       flagsFor("+sheet-copy"),
-	Validate:    validateViaInput(sheetCopyInput),
+	Tips: []string{
+		"Example: lark-cli sheets +sheet-copy --url <URL> --sheet-name 数据源 --title 数据源-副本",
+		"--sheet-name / --sheet-id selects the SOURCE sheet; the copy's new name goes in --title.",
+	},
+	Validate: validateViaInput(sheetCopyInput),
 	DryRun: func(ctx context.Context, runtime *common.RuntimeContext) *common.DryRunAPI {
 		token, _ := resolveSpreadsheetToken(runtime)
 		sheetID, sheetName, _ := resolveSheetSelector(runtime)
@@ -914,6 +919,14 @@ type workbookCreateStylePayload struct {
 	RowSizes   []workbookCreateResizeOp
 	ColSizes   []workbookCreateResizeOp
 	CellMerges []workbookCreateMergeOp
+	Freeze     *workbookCreateFreezeOp
+}
+
+// workbookCreateFreezeOp freezes the first Rows rows / Cols columns.
+// Zero means "leave that dimension alone".
+type workbookCreateFreezeOp struct {
+	Rows int
+	Cols int
 }
 
 type workbookCreateCellStyleOp struct {
@@ -965,7 +978,11 @@ func parseWorkbookCreateStyles(runtime flagView) (*workbookCreateStylePayload, e
 	if len(items) != 1 {
 		return nil, common.ValidationErrorf("--styles.styles must contain exactly one item when using --values")
 	}
-	return parseWorkbookCreateStyleItem(items[0], "--styles.styles[0]")
+	payload, probs := parseWorkbookCreateStyleItem(items[0], "--styles.styles[0]")
+	if err := joinStyleValidationErrors(probs); err != nil {
+		return nil, err
+	}
+	return payload, nil
 }
 
 // parseWorkbookCreateSheetStyles parses --styles for the typed --sheets path.
@@ -988,21 +1005,28 @@ func parseWorkbookCreateSheetStyles(runtime flagView, payload *tablePayload) (*w
 	}
 	out := &workbookCreateSheetStyles{ByName: map[string]*workbookCreateStylePayload{}}
 	out.ByIndex = make([]*workbookCreateStylePayload, len(payload.Sheets))
+	var probs []error
 	for i, item := range items {
 		name, _ := item["name"].(string)
 		if strings.TrimSpace(name) == "" {
-			return nil, common.ValidationErrorf("--styles.styles[%d].name is required", i)
+			probs = append(probs, common.ValidationErrorf("--styles.styles[%d].name is required", i))
+			continue
 		}
 		if name != payload.Sheets[i].Name {
-			return nil, common.ValidationErrorf("--styles.styles[%d].name %q must match --sheets.sheets[%d].name %q", i, name, i, payload.Sheets[i].Name)
+			probs = append(probs, common.ValidationErrorf("--styles.styles[%d].name %q must match --sheets.sheets[%d].name %q", i, name, i, payload.Sheets[i].Name))
+			continue
 		}
-		style, err := parseWorkbookCreateStyleItem(item, fmt.Sprintf("--styles.styles[%d]", i))
-		if err != nil {
-			return nil, err
+		style, itemProbs := parseWorkbookCreateStyleItem(item, fmt.Sprintf("--styles.styles[%d]", i))
+		if len(itemProbs) > 0 {
+			probs = append(probs, itemProbs...)
+			continue
 		}
 		out.ByIndex[i] = style
 		out.ByName[name] = style
 	}
+	if err := joinStyleValidationErrors(probs); err != nil {
+		return nil, err
+	}
 	return out, nil
 }
 
@@ -1030,182 +1054,337 @@ func parseWorkbookCreateStylesItems(v interface{}) ([]map[string]interface{}, er
 	return items, nil
 }
 
-func parseWorkbookCreateStyleItem(item map[string]interface{}, path string) (*workbookCreateStylePayload, error) {
+// parseWorkbookCreateStyleItem parses one --styles item. All four sections
+// are validated even after one fails, and every issue is returned in the
+// slice: eval traces show agents fixing --styles errors one round trip per
+// error (border side, then row_sizes.type, then size…) because only the
+// first was ever reported.
+func parseWorkbookCreateStyleItem(item map[string]interface{}, path string) (*workbookCreateStylePayload, []error) {
 	payload := &workbookCreateStylePayload{}
-	var err error
+	var probs []error
 	if raw, ok := item["cell_styles"]; ok {
-		payload.CellStyles, err = parseWorkbookCreateCellStyleOps(raw, path+".cell_styles")
-		if err != nil {
-			return nil, err
-		}
+		var errsHere []error
+		payload.CellStyles, errsHere = parseWorkbookCreateCellStyleOps(raw, path+".cell_styles")
+		probs = append(probs, errsHere...)
 	}
 	if raw, ok := item["row_sizes"]; ok {
-		payload.RowSizes, err = parseWorkbookCreateResizeOps(raw, path+".row_sizes", "row")
-		if err != nil {
-			return nil, err
-		}
+		var errsHere []error
+		payload.RowSizes, errsHere = parseWorkbookCreateResizeOps(raw, path+".row_sizes", "row")
+		probs = append(probs, errsHere...)
 	}
 	if raw, ok := item["col_sizes"]; ok {
-		payload.ColSizes, err = parseWorkbookCreateResizeOps(raw, path+".col_sizes", "column")
-		if err != nil {
-			return nil, err
-		}
+		var errsHere []error
+		payload.ColSizes, errsHere = parseWorkbookCreateResizeOps(raw, path+".col_sizes", "column")
+		probs = append(probs, errsHere...)
 	}
 	if raw, ok := item["cell_merges"]; ok {
-		payload.CellMerges, err = parseWorkbookCreateMergeOps(raw, path+".cell_merges")
+		var errsHere []error
+		payload.CellMerges, errsHere = parseWorkbookCreateMergeOps(raw, path+".cell_merges")
+		probs = append(probs, errsHere...)
+	}
+	if raw, ok := item["freeze"]; ok {
+		freeze, err := parseWorkbookCreateFreezeOp(raw, path+".freeze")
 		if err != nil {
-			return nil, err
+			probs = append(probs, err)
+		} else {
+			payload.Freeze = freeze
 		}
 	}
-	if len(payload.CellStyles) == 0 && len(payload.RowSizes) == 0 && len(payload.ColSizes) == 0 && len(payload.CellMerges) == 0 {
-		return nil, common.ValidationErrorf("%s must include at least one of cell_styles/row_sizes/col_sizes/cell_merges", path)
+	if len(probs) > 0 {
+		return nil, probs
+	}
+	if len(payload.CellStyles) == 0 && len(payload.RowSizes) == 0 && len(payload.ColSizes) == 0 && len(payload.CellMerges) == 0 && payload.Freeze == nil {
+		return nil, []error{common.ValidationErrorf("%s must include at least one of cell_styles/row_sizes/col_sizes/cell_merges/freeze", path)}
 	}
 	return payload, nil
 }
 
-func parseWorkbookCreateCellStyleOps(v interface{}, path string) ([]workbookCreateCellStyleOp, error) {
+// parseWorkbookCreateFreezeOp parses a {rows, cols} freeze section. At least
+// one dimension must be positive — an all-zero freeze is a no-op the caller
+// almost certainly didn't mean.
+func parseWorkbookCreateFreezeOp(raw interface{}, path string) (*workbookCreateFreezeOp, error) {
+	obj, ok := raw.(map[string]interface{})
+	if !ok {
+		return nil, common.ValidationErrorf("%s must be an object like {\"rows\":1} or {\"rows\":1,\"cols\":2}", path)
+	}
+	out := &workbookCreateFreezeOp{}
+	for k, v := range obj {
+		n, isNum := v.(float64)
+		if !isNum || n != float64(int(n)) || n < 0 {
+			return nil, common.ValidationErrorf("%s.%s must be a non-negative integer", path, k)
+		}
+		switch k {
+		case "rows":
+			out.Rows = int(n)
+		case "cols", "columns":
+			out.Cols = int(n)
+		default:
+			return nil, common.ValidationErrorf("%s.%s is not a supported field (want rows/cols)", path, k)
+		}
+	}
+	if out.Rows == 0 && out.Cols == 0 {
+		return nil, common.ValidationErrorf("%s must freeze at least one dimension (rows or cols > 0)", path)
+	}
+	return out, nil
+}
+
+// joinStyleValidationErrors folds the issues collected across one --styles
+// parse into a single typed error that lists them all, so the caller can fix
+// the whole payload in one retry instead of one error per round trip.
+func joinStyleValidationErrors(probs []error) error {
+	switch len(probs) {
+	case 0:
+		return nil
+	case 1:
+		return probs[0]
+	}
+	const maxShown = 8
+	msgs := make([]string, 0, len(probs))
+	for _, e := range probs {
+		if p, ok := errs.ProblemOf(e); ok {
+			msgs = append(msgs, p.Message)
+			continue
+		}
+		msgs = append(msgs, e.Error())
+	}
+	suffix := ""
+	if len(msgs) > maxShown {
+		suffix = fmt.Sprintf(" (+%d more)", len(msgs)-maxShown)
+		msgs = msgs[:maxShown]
+	}
+	return common.ValidationErrorf("--styles has %d issues: %s%s", len(probs), strings.Join(msgs, " | "), suffix)
+}
+
+func parseWorkbookCreateCellStyleOps(v interface{}, path string) ([]workbookCreateCellStyleOp, []error) {
 	arr, ok := v.([]interface{})
 	if !ok {
-		return nil, common.ValidationErrorf("%s must be an array", path)
+		return nil, []error{common.ValidationErrorf("%s must be an array", path)}
 	}
 	ops := make([]workbookCreateCellStyleOp, 0, len(arr))
+	var probs []error
 	for i, raw := range arr {
-		op, ok := raw.(map[string]interface{})
-		if !ok {
-			return nil, common.ValidationErrorf("%s[%d] must be an object", path, i)
-		}
-		rangeStr, err := requireWorkbookCreateRange(op, fmt.Sprintf("%s[%d]", path, i))
+		op, err := parseWorkbookCreateCellStyleOp(raw, fmt.Sprintf("%s[%d]", path, i))
 		if err != nil {
-			return nil, err
+			probs = append(probs, err)
+			continue
 		}
-		if _, _, _, _, err := workbookCreateStyleRangeBounds(rangeStr); err != nil {
-			return nil, common.ValidationErrorf("%s[%d].range %q: %v", path, i, rangeStr, err)
-		}
-		styleObj := make(map[string]interface{}, len(op)-1)
-		for k, v := range op {
-			if k == "range" {
-				continue
-			}
-			styleObj[k] = v
-		}
-		style, err := normalizeWorkbookCreateStyleObject(styleObj, fmt.Sprintf("%s[%d]", path, i))
-		if err != nil {
-			return nil, err
-		}
-		if len(style) == 0 {
-			return nil, common.ValidationErrorf("%s[%d] must include at least one style field", path, i)
-		}
-		ops = append(ops, workbookCreateCellStyleOp{Range: rangeStr, Style: style})
+		ops = append(ops, op)
 	}
-	return ops, nil
+	return ops, probs
 }
 
-func parseWorkbookCreateMergeOps(v interface{}, path string) ([]workbookCreateMergeOp, error) {
+func parseWorkbookCreateCellStyleOp(raw interface{}, path string) (workbookCreateCellStyleOp, error) {
+	op, ok := raw.(map[string]interface{})
+	if !ok {
+		return workbookCreateCellStyleOp{}, common.ValidationErrorf("%s must be an object", path)
+	}
+	rangeStr, err := requireWorkbookCreateRange(op, path)
+	if err != nil {
+		return workbookCreateCellStyleOp{}, err
+	}
+	if _, _, _, _, err := workbookCreateStyleRangeBounds(rangeStr); err != nil {
+		return workbookCreateCellStyleOp{}, common.ValidationErrorf("%s.range %q: %v", path, rangeStr, err)
+	}
+	styleObj := make(map[string]interface{}, len(op)-1)
+	for k, v := range op {
+		if k == "range" {
+			continue
+		}
+		styleObj[k] = v
+	}
+	style, err := normalizeWorkbookCreateStyleObject(styleObj, path)
+	if err != nil {
+		return workbookCreateCellStyleOp{}, err
+	}
+	if len(style) == 0 {
+		return workbookCreateCellStyleOp{}, common.ValidationErrorf("%s must include at least one style field", path)
+	}
+	return workbookCreateCellStyleOp{Range: rangeStr, Style: style}, nil
+}
+
+func parseWorkbookCreateMergeOps(v interface{}, path string) ([]workbookCreateMergeOp, []error) {
 	arr, ok := v.([]interface{})
 	if !ok {
-		return nil, common.ValidationErrorf("%s must be an array", path)
+		return nil, []error{common.ValidationErrorf("%s must be an array", path)}
 	}
 	ops := make([]workbookCreateMergeOp, 0, len(arr))
+	var probs []error
 	for i, raw := range arr {
-		op, ok := raw.(map[string]interface{})
-		if !ok {
-			return nil, common.ValidationErrorf("%s[%d] must be an object", path, i)
-		}
-		rangeStr, err := requireWorkbookCreateRange(op, fmt.Sprintf("%s[%d]", path, i))
+		op, err := parseWorkbookCreateMergeOp(raw, fmt.Sprintf("%s[%d]", path, i))
 		if err != nil {
-			return nil, err
+			probs = append(probs, err)
+			continue
 		}
-		if _, _, _, _, err := workbookCreateStyleRangeBounds(rangeStr); err != nil {
-			return nil, common.ValidationErrorf("%s[%d].range %q: %v", path, i, rangeStr, err)
-		}
-		mergeType := "all"
-		if raw, ok := op["merge_type"]; ok {
-			v, ok := raw.(string)
-			if !ok || strings.TrimSpace(v) == "" {
-				return nil, common.ValidationErrorf("%s[%d].merge_type must be a non-empty string", path, i)
-			}
-			mergeType = strings.TrimSpace(v)
-		}
-		switch mergeType {
-		case "all", "rows", "columns":
-		default:
-			return nil, common.ValidationErrorf("%s[%d].merge_type %q is invalid (want all/rows/columns)", path, i, mergeType)
-		}
-		if err := rejectUnexpectedWorkbookStyleFields(op, fmt.Sprintf("%s[%d]", path, i), "range", "merge_type"); err != nil {
-			return nil, err
-		}
-		ops = append(ops, workbookCreateMergeOp{Range: rangeStr, MergeType: mergeType})
+		ops = append(ops, op)
 	}
-	return ops, nil
+	return ops, probs
 }
 
-func parseWorkbookCreateResizeOps(v interface{}, path, dimension string) ([]workbookCreateResizeOp, error) {
+func parseWorkbookCreateMergeOp(raw interface{}, path string) (workbookCreateMergeOp, error) {
+	// A bare range string means {range: s, merge_type: all} — the only
+	// possible reading (07-20 eval hit).
+	if s, ok := raw.(string); ok && strings.TrimSpace(s) != "" {
+		raw = map[string]interface{}{"range": strings.TrimSpace(s)}
+	}
+	op, ok := raw.(map[string]interface{})
+	if !ok {
+		return workbookCreateMergeOp{}, common.ValidationErrorf("%s must be an object", path)
+	}
+	rangeStr, err := requireWorkbookCreateRange(op, path)
+	if err != nil {
+		return workbookCreateMergeOp{}, err
+	}
+	if _, _, _, _, err := workbookCreateStyleRangeBounds(rangeStr); err != nil {
+		return workbookCreateMergeOp{}, common.ValidationErrorf("%s.range %q: %v", path, rangeStr, err)
+	}
+	mergeType := "all"
+	if raw, ok := op["merge_type"]; ok {
+		v, ok := raw.(string)
+		if !ok || strings.TrimSpace(v) == "" {
+			return workbookCreateMergeOp{}, common.ValidationErrorf("%s.merge_type must be a non-empty string", path)
+		}
+		mergeType = normalizeMergeType(strings.TrimSpace(v))
+	}
+	switch mergeType {
+	case "all", "rows", "columns":
+	default:
+		return workbookCreateMergeOp{}, common.ValidationErrorf("%s.merge_type %q is invalid (want all/rows/columns)", path, mergeType)
+	}
+	if err := rejectUnexpectedWorkbookStyleFields(op, path, "range", "merge_type"); err != nil {
+		return workbookCreateMergeOp{}, err
+	}
+	return workbookCreateMergeOp{Range: rangeStr, MergeType: mergeType}, nil
+}
+
+// normalizeMergeType maps the raw OpenAPI merge vocabulary (MERGE_ALL /
+// MERGE_ROWS / MERGE_COLUMNS — which agents reproduce from the Lark API
+// docs) onto the CLI's all/rows/columns. Unknown values pass through for
+// the caller's enum check to reject.
+func normalizeMergeType(v string) string {
+	lower := strings.ToLower(v)
+	lower = strings.TrimPrefix(lower, "merge_")
+	switch lower {
+	case "all", "rows", "columns":
+		return lower
+	}
+	return v
+}
+
+func parseWorkbookCreateResizeOps(v interface{}, path, dimension string) ([]workbookCreateResizeOp, []error) {
 	arr, ok := v.([]interface{})
 	if !ok {
-		return nil, common.ValidationErrorf("%s must be an array", path)
+		return nil, []error{common.ValidationErrorf("%s must be an array", path)}
 	}
 	ops := make([]workbookCreateResizeOp, 0, len(arr))
+	var probs []error
 	for i, raw := range arr {
-		op, ok := raw.(map[string]interface{})
-		if !ok {
-			return nil, common.ValidationErrorf("%s[%d] must be an object", path, i)
-		}
-		rangeStr, err := requireWorkbookCreateRange(op, fmt.Sprintf("%s[%d]", path, i))
+		op, err := parseWorkbookCreateResizeOp(raw, fmt.Sprintf("%s[%d]", path, i), dimension)
 		if err != nil {
-			return nil, err
+			probs = append(probs, err)
+			continue
 		}
-		parsedDim, _, _, err := parseA1Range(rangeStr)
-		if err != nil {
-			want := "row numbers like 2:10"
-			if dimension == "column" {
-				want = "column letters like A:E"
-			}
-			return nil, common.ValidationErrorf("%s[%d].range %q must use %s: %v", path, i, rangeStr, want, err)
+		ops = append(ops, op)
+	}
+	return ops, probs
+}
+
+// resizeOpExample renders a complete valid op for the dimension, inlined on
+// every type/size error: eval traces show the field errors chaining (type
+// "custom" → fixed to pixel → "pixel requires size"), each costing a round
+// trip, because no error ever showed a whole valid op at once.
+func resizeOpExample(dimension string) string {
+	if dimension == "column" {
+		return `{"range":"A:C","type":"pixel","size":120} (or {"range":"A:C","type":"standard"} to reset)`
+	}
+	return `{"range":"2:10","type":"pixel","size":32} (or "type":"auto" to fit content)`
+}
+
+func parseWorkbookCreateResizeOp(raw interface{}, path, dimension string) (workbookCreateResizeOp, error) {
+	op, ok := raw.(map[string]interface{})
+	if !ok {
+		return workbookCreateResizeOp{}, common.ValidationErrorf("%s must be an object", path)
+	}
+	rangeStr, err := requireWorkbookCreateRange(op, path)
+	if err != nil {
+		return workbookCreateResizeOp{}, err
+	}
+	parsedDim, _, _, err := parseA1Range(rangeStr)
+	if err != nil {
+		want := "row numbers like 2:10"
+		if dimension == "column" {
+			want = "column letters like A:E"
 		}
-		if parsedDim != dimension {
-			want := "row numbers like 2:10"
-			if dimension == "column" {
-				want = "column letters like A:E"
-			}
-			return nil, common.ValidationErrorf("%s[%d].range %q must use %s", path, i, rangeStr, want)
-		}
-		typeHint := "pixel/standard"
-		if dimension == "row" {
-			typeHint = "pixel/standard/auto"
-		}
-		resizeType, _ := op["type"].(string)
-		resizeType = strings.TrimSpace(resizeType)
-		if resizeType == "" {
-			return nil, common.ValidationErrorf("%s[%d].type is required (%s)", path, i, typeHint)
+		return workbookCreateResizeOp{}, common.ValidationErrorf("%s.range %q must use %s: %v", path, rangeStr, want, err)
+	}
+	if parsedDim != dimension {
+		want := "row numbers like 2:10"
+		if dimension == "column" {
+			want = "column letters like A:E"
 		}
+		return workbookCreateResizeOp{}, common.ValidationErrorf("%s.range %q must use %s", path, rangeStr, want)
+	}
+	typeHint := "pixel/standard"
+	if dimension == "row" {
+		typeHint = "pixel/standard/auto"
+	}
+	resizeType, _ := op["type"].(string)
+	resizeType = strings.TrimSpace(resizeType)
+	if resizeType != "" {
 		if dimension == "column" && resizeType == "auto" {
-			return nil, common.ValidationErrorf("%s[%d].type auto is rows-only", path, i)
+			return workbookCreateResizeOp{}, common.ValidationErrorf("%s.type auto is rows-only", path)
 		}
 		switch resizeType {
 		case "pixel", "standard", "auto":
 		default:
-			return nil, common.ValidationErrorf("%s[%d].type %q is invalid (want %s)", path, i, resizeType, typeHint)
+			return workbookCreateResizeOp{}, common.ValidationErrorf("%s.type %q is invalid (want %s), e.g. %s", path, resizeType, typeHint, resizeOpExample(dimension))
 		}
-		size := 0
-		if raw, ok := op["size"]; ok {
-			n, ok := util.ToFloat64(raw)
-			if !ok || n <= 0 {
-				return nil, common.ValidationErrorf("%s[%d].size must be a positive number", path, i)
-			}
-			size = int(n)
-		}
-		if resizeType == "pixel" && size <= 0 {
-			return nil, common.ValidationErrorf("%s[%d].type pixel requires size", path, i)
-		}
-		if resizeType != "pixel" && size > 0 {
-			return nil, common.ValidationErrorf("%s[%d].size is only valid with type pixel", path, i)
-		}
-		if err := rejectUnexpectedWorkbookStyleFields(op, fmt.Sprintf("%s[%d]", path, i), "range", "type", "size"); err != nil {
-			return nil, err
-		}
-		ops = append(ops, workbookCreateResizeOp{Range: normalizeWorkbookResizeRange(rangeStr), ResizeType: resizeType, Size: size})
 	}
-	return ops, nil
+	// size is the canonical dimension key (uniform across row_sizes and
+	// col_sizes — the array name already carries the dimension). The Excel-
+	// vocabulary alias (height on rows, width on columns) is accepted
+	// silently; the WRONG dimension's word is a targeted error, never a
+	// silent rewrite.
+	alias, wrongDim := "height", "width"
+	if dimension == "column" {
+		alias, wrongDim = "width", "height"
+	}
+	if _, has := op[wrongDim]; has {
+		return workbookCreateResizeOp{}, common.ValidationErrorf("%s.%s does not apply to this array (the array name carries the dimension); use size, e.g. %s", path, wrongDim, resizeOpExample(dimension))
+	}
+	sizeRaw, hasSize := op["size"]
+	if aliasRaw, hasAlias := op[alias]; hasAlias {
+		if hasSize {
+			return workbookCreateResizeOp{}, common.ValidationErrorf("%s: give either size or %s, not both", path, alias)
+		}
+		sizeRaw, hasSize = aliasRaw, true
+	}
+	size := 0
+	if hasSize {
+		n, ok := util.ToFloat64(sizeRaw)
+		if !ok || n <= 0 {
+			return workbookCreateResizeOp{}, common.ValidationErrorf("%s.size must be a positive number", path)
+		}
+		size = int(n)
+	}
+	// type is optional ceremony when a pixel size is given: {range, size}
+	// means a pixel resize, exactly as --width/--height without --type does
+	// on the flag path. Explicit standard/auto still needs type.
+	if resizeType == "" {
+		if size <= 0 {
+			return workbookCreateResizeOp{}, common.ValidationErrorf("%s needs size (px) or type (%s), e.g. %s", path, typeHint, resizeOpExample(dimension))
+		}
+		resizeType = "pixel"
+	}
+	if resizeType == "pixel" && size <= 0 {
+		return workbookCreateResizeOp{}, common.ValidationErrorf("%s.type pixel requires size, e.g. %s", path, resizeOpExample(dimension))
+	}
+	if resizeType != "pixel" && size > 0 {
+		return workbookCreateResizeOp{}, common.ValidationErrorf("%s.size is only valid with type pixel", path)
+	}
+	if err := rejectUnexpectedWorkbookStyleFields(op, path, "range", "type", "size", alias); err != nil {
+		return workbookCreateResizeOp{}, err
+	}
+	return workbookCreateResizeOp{Range: normalizeWorkbookResizeRange(rangeStr), ResizeType: resizeType, Size: size}, nil
 }
 
 func requireWorkbookCreateRange(op map[string]interface{}, path string) (string, error) {
@@ -1245,6 +1424,9 @@ func normalizeWorkbookCreateStyleObject(in map[string]interface{}, path string)
 	if len(in) == 0 {
 		return nil, nil
 	}
+	if err := foldBorderFamilyAliases(in, path); err != nil {
+		return nil, err
+	}
 	if err := normalizeCellStyleAliases(in, path); err != nil {
 		return nil, err
 	}
@@ -1259,15 +1441,26 @@ func normalizeWorkbookCreateStyleObject(in map[string]interface{}, path string)
 			if !ok {
 				return nil, common.ValidationErrorf("%s.border_styles must be a JSON object", path)
 			}
+			expandBorderAllShorthand(m)
 			if err := validateWorkbookBorderStyles(m, path); err != nil {
 				return nil, err
 			}
 			out["border_styles"] = m
 		case "value", "formula", "rich_text", "multiple_values", "note", "data_validation":
-			return nil, common.ValidationErrorf("%s is for styles only; put content in --values or use --sheets for typed cell objects", path)
+			return nil, common.ValidationErrorf("%s.%s is a content field — a styles spec carries no cell content; write values/formulas via +cells-set or +table-put", path, k)
 		default:
 			if !workbookCreateCellStyleField(k) {
-				return nil, common.ValidationErrorf("%s.%s is not a supported style field", path, k)
+				// Universal rejection with did-you-mean + the full field list:
+				// this is the mechanism that absorbs the infinite tail of
+				// spelling permutations at a fixed one-retry cost — silent
+				// aliases are reserved for high-frequency words from real
+				// external vocabularies (see the style_vocab.go contract).
+				msg := fmt.Sprintf("%s.%s is not a supported style field", path, k)
+				if match := suggest.Closest(strings.ToLower(k), workbookCreateCellStyleFieldList, 1); len(match) > 0 {
+					msg += fmt.Sprintf(" — did you mean %q?", match[0])
+				}
+				msg += "; supported: " + strings.Join(workbookCreateCellStyleFieldList, ", ")
+				return nil, common.ValidationErrorf("%s", msg)
 			}
 			cellStyle[k] = v
 		}
@@ -1278,6 +1471,14 @@ func normalizeWorkbookCreateStyleObject(in map[string]interface{}, path string)
 	return out, nil
 }
 
+// workbookCreateCellStyleFieldList is the canonical style vocabulary plus the
+// two border carriers, in display order for the unknown-field hint.
+var workbookCreateCellStyleFieldList = []string{
+	"font_color", "font_family", "font_size", "font_weight", "font_style", "font_line",
+	"background_color", "horizontal_alignment", "vertical_alignment",
+	"number_format", "word_wrap", "border", "border_styles",
+}
+
 func workbookCreateCellStyleField(name string) bool {
 	switch name {
 	case "font_color", "font_family", "font_size", "font_weight", "font_style", "font_line",
@@ -1299,7 +1500,7 @@ func validateWorkbookBorderStyles(m map[string]interface{}, path string) error {
 		switch side {
 		case "top", "bottom", "left", "right":
 		default:
-			return common.ValidationErrorf("%s.border_styles.%s is not a valid side (want top/bottom/left/right)", path, side)
+			return common.ValidationErrorf("%s.border_styles.%s is not a valid side (want top/bottom/left/right; a horizontal line is the top/bottom side of its range, a vertical line is left/right)", path, side)
 		}
 		spec, ok := raw.(map[string]interface{})
 		if !ok {
@@ -1516,7 +1717,7 @@ func workbookCreateVisualOps(styles *workbookCreateStylePayload) []workbookCreat
 	if styles == nil {
 		return nil
 	}
-	ops := make([]workbookCreateStyleOp, 0, len(styles.CellMerges)+len(styles.RowSizes)+len(styles.ColSizes))
+	ops := make([]workbookCreateStyleOp, 0, len(styles.CellMerges)+len(styles.RowSizes)+len(styles.ColSizes)+2)
 	for _, op := range styles.CellMerges {
 		ops = append(ops, workbookCreateStyleOp{Kind: "cell_merge", Range: op.Range, MergeType: op.MergeType})
 	}
@@ -1526,6 +1727,14 @@ func workbookCreateVisualOps(styles *workbookCreateStylePayload) []workbookCreat
 	for _, op := range styles.ColSizes {
 		ops = append(ops, workbookCreateStyleOp{Kind: "col_size", Range: op.Range, ResizeType: op.ResizeType, Size: op.Size})
 	}
+	if styles.Freeze != nil {
+		if styles.Freeze.Rows > 0 {
+			ops = append(ops, workbookCreateStyleOp{Kind: "freeze_rows", Size: styles.Freeze.Rows})
+		}
+		if styles.Freeze.Cols > 0 {
+			ops = append(ops, workbookCreateStyleOp{Kind: "freeze_cols", Size: styles.Freeze.Cols})
+		}
+	}
 	return ops
 }
 
@@ -1564,6 +1773,18 @@ func workbookCreateVisualOpInput(token, sheetID, sheetName string, op workbookCr
 			input["resize_width"] = block
 		}
 		return input, "resize_range"
+	case "freeze_rows", "freeze_cols":
+		input := map[string]interface{}{
+			"excel_id":  token,
+			"operation": "freeze",
+		}
+		sheetSelectorForToolInput(input, sheetID, sheetName)
+		if op.Kind == "freeze_rows" {
+			input["freeze_rows"] = op.Size
+		} else {
+			input["freeze_columns"] = op.Size
+		}
+		return input, "modify_sheet_structure"
 	default:
 		return nil, ""
 	}

shortcuts/sheets/lark_sheet_write_cells.go

@@ -38,7 +38,11 @@ import (
 
 // CellsSet wraps set_cell_range: caller provides the cells matrix via --cells
 // (JSON), with an optional --copy-to-range to replicate the written block
-// across a larger area (formula refs auto-shift).
+// across a larger area (formula refs auto-shift). The plural form --writes
+// ([{sheet_name, range, cells}, …]) fans scattered regions — cross-sheet
+// allowed — into ONE atomic batch_update: eval traces show "fix all broken
+// formulas across ranges/sheets" as the dominant homogeneous scenario still
+// hand-assembled as +batch-update operations arrays.
 var CellsSet = common.Shortcut{
 	Service:     "sheets",
 	Command:     "+cells-set",
@@ -48,9 +52,31 @@ var CellsSet = common.Shortcut{
 	AuthTypes:   []string{"user", "bot"},
 	HasFormat:   true,
 	Flags:       flagsFor("+cells-set"),
-	Validate:    validateViaInput(cellsSetInput),
+	Tips: []string{
+		`Example: lark-cli sheets +cells-set --url <URL> --sheet-name Sheet1 --range A1:B1 --cells '[[{"value":"名称"},{"formula":"=SUM(B2:B9)"}]]'`,
+		`--cells is always a 2D array (rows × cells), even for one cell: [[{"value":…}]].`,
+		`Scattered regions (e.g. fixing formulas across ranges/sheets): --writes '[{"sheet_name":…,"range":…,"cells":[[…]]}, …]' — one atomic call, sheet selector inside each item.`,
+	},
+	Validate: func(ctx context.Context, runtime *common.RuntimeContext) error {
+		if runtime.Changed("writes") {
+			token, err := resolveSpreadsheetToken(runtime)
+			if err != nil {
+				return err
+			}
+			_, err = cellsSetWritesOps(runtime, token)
+			return err
+		}
+		return validateViaInput(cellsSetInput)(ctx, runtime)
+	},
 	DryRun: func(ctx context.Context, runtime *common.RuntimeContext) *common.DryRunAPI {
 		token, _ := resolveSpreadsheetToken(runtime)
+		if runtime.Changed("writes") {
+			ops, _ := cellsSetWritesOps(runtime, token)
+			return invokeToolDryRun(token, ToolKindWrite, "batch_update", map[string]interface{}{
+				"excel_id":   token,
+				"operations": ops,
+			})
+		}
 		sheetID, sheetName, _ := resolveSheetSelector(runtime)
 		input, _ := cellsSetInput(runtime, token, sheetID, sheetName)
 		return invokeToolDryRun(token, ToolKindWrite, "set_cell_range", input)
@@ -60,6 +86,21 @@ var CellsSet = common.Shortcut{
 		if err != nil {
 			return err
 		}
+		if runtime.Changed("writes") {
+			ops, err := cellsSetWritesOps(runtime, token)
+			if err != nil {
+				return err
+			}
+			out, err := callTool(ctx, runtime, token, ToolKindWrite, "batch_update", map[string]interface{}{
+				"excel_id":   token,
+				"operations": ops,
+			})
+			if err != nil {
+				return err
+			}
+			runtime.Out(out, nil)
+			return nil
+		}
 		sheetID, sheetName, err := resolveSheetSelector(runtime)
 		if err != nil {
 			return err
@@ -77,6 +118,108 @@ var CellsSet = common.Shortcut{
 	},
 }
 
+// cellsSetWritesOps parses --writes ([{sheet_name|sheet_id, range, cells}, …])
+// and expands it into set_cell_range operations for ONE atomic batch_update.
+// Single source of truth per item: the sheet selector LIVES IN THE ITEM (same
+// convention as +batch-update sub-ops and +styles-put items — no top-level
+// fallback, no precedence table to remember). Every item runs through the
+// exact standalone pipeline (key vocabulary, style acceptance layer, matrix
+// precheck, schema validation) via a per-item flag view, and item errors are
+// aggregated so one retry fixes them all.
+func cellsSetWritesOps(runtime *common.RuntimeContext, token string) ([]interface{}, error) {
+	for _, conflicting := range []string{"range", "cells", "copy-to-range"} {
+		if runtime.Changed(conflicting) {
+			return nil, sheetsValidationForFlag("writes", "--writes and --%s are mutually exclusive: single region → --range + --cells; multiple regions → --writes alone", conflicting)
+		}
+	}
+	if strings.TrimSpace(runtime.Str("sheet-name")) != "" || strings.TrimSpace(runtime.Str("sheet-id")) != "" {
+		return nil, sheetsValidationForFlag("writes", "--writes does not accept a top-level sheet selector — put sheet_name (or sheet_id) inside each writes item, same as +batch-update sub-ops")
+	}
+	raw, err := requireJSONArray(runtime, "writes")
+	if err != nil {
+		return nil, err
+	}
+	if len(raw) == 0 {
+		return nil, sheetsValidationForFlag("writes", "--writes must be a non-empty JSON array of {sheet_name, range, cells} items")
+	}
+	if len(raw) > maxBatchOperations {
+		return nil, sheetsValidationForFlag("writes", "--writes accepts at most %d items; got %d — merge adjacent regions or split into several calls", maxBatchOperations, len(raw))
+	}
+	topLevelOverwrite := runtime.Bool("allow-overwrite")
+	ops := make([]interface{}, 0, len(raw))
+	var probs []error
+	var totalCells int64
+	for i, v := range raw {
+		item, ok := v.(map[string]interface{})
+		if !ok {
+			probs = append(probs, common.ValidationErrorf("--writes[%d] must be an object like {\"sheet_name\":…,\"range\":…,\"cells\":[[…]]}", i))
+			continue
+		}
+		if err := normalizeSubOpInputKeys("+cells-set", item); err != nil {
+			probs = append(probs, common.ValidationErrorf("--writes[%d]: %v", i, err))
+			continue
+		}
+		if topLevelOverwrite {
+			if _, has := item["allow_overwrite"]; !has {
+				item["allow_overwrite"] = true
+			}
+		}
+		fv := newMapFlagViewForCommand("+cells-set", item)
+		sheetID := strings.TrimSpace(fv.Str("sheet-id"))
+		sheetName := strings.TrimSpace(fv.Str("sheet-name"))
+		input, err := cellsSetInput(fv, token, sheetID, sheetName)
+		if err != nil {
+			probs = append(probs, common.ValidationErrorf("--writes[%d]: %v", i, err))
+			continue
+		}
+		if cells, ok := input["cells"].([]interface{}); ok {
+			for _, row := range cells {
+				if r, ok := row.([]interface{}); ok {
+					totalCells += int64(len(r))
+				}
+			}
+		}
+		if err := checkBatchStampBudget(totalCells); err != nil {
+			return nil, err
+		}
+		ops = append(ops, map[string]interface{}{
+			"tool_name": "set_cell_range",
+			"input":     input,
+		})
+	}
+	if err := joinWritesValidationErrors(probs); err != nil {
+		return nil, err
+	}
+	return ops, nil
+}
+
+// joinWritesValidationErrors mirrors joinStyleValidationErrors for --writes:
+// every item's first error in one message, so the whole payload is fixed in
+// a single retry.
+func joinWritesValidationErrors(probs []error) error {
+	switch len(probs) {
+	case 0:
+		return nil
+	case 1:
+		return probs[0]
+	}
+	const maxShown = 8
+	msgs := make([]string, 0, len(probs))
+	for _, e := range probs {
+		if p, ok := errs.ProblemOf(e); ok {
+			msgs = append(msgs, p.Message)
+			continue
+		}
+		msgs = append(msgs, e.Error())
+	}
+	suffix := ""
+	if len(msgs) > maxShown {
+		suffix = fmt.Sprintf(" (+%d more)", len(msgs)-maxShown)
+		msgs = msgs[:maxShown]
+	}
+	return common.ValidationErrorf("--writes has %d issues: %s%s", len(probs), strings.Join(msgs, " | "), suffix)
+}
+
 func cellsSetInput(runtime flagView, token, sheetID, sheetName string) (map[string]interface{}, error) {
 	if err := requireSheetSelector(sheetID, sheetName); err != nil {
 		return nil, err
@@ -91,9 +234,13 @@ func cellsSetInput(runtime flagView, token, sheetID, sheetName string) (map[stri
 	if err := normalizeTypedCellsStyleAliases(cells, "--cells"); err != nil {
 		return nil, err
 	}
+	rangeStr := strings.TrimSpace(runtime.Str("range"))
+	if err := checkCellsMatchRange(cells, rangeStr); err != nil {
+		return nil, err
+	}
 	input := map[string]interface{}{
 		"excel_id": token,
-		"range":    strings.TrimSpace(runtime.Str("range")),
+		"range":    rangeStr,
 		"cells":    cells,
 	}
 	sheetSelectorForToolInput(input, sheetID, sheetName)
@@ -124,7 +271,11 @@ var CellsSetStyle = common.Shortcut{
 	AuthTypes:   []string{"user", "bot"},
 	HasFormat:   true,
 	Flags:       flagsFor("+cells-set-style"),
-	Validate:    validateViaInput(cellsSetStyleInput),
+	Tips: []string{
+		`Example: lark-cli sheets +cells-set-style --url <URL> --sheet-name Sheet1 --range A1:D1 --font-weight bold --background-color "#F0F0F0" --horizontal-alignment center`,
+		`Borders take JSON: --border-styles '{"top":{"style":"solid","weight":"thin","color":"#000000"}}' (sides: top/bottom/left/right).`,
+	},
+	Validate: validateViaInput(cellsSetStyleInput),
 	DryRun: func(ctx context.Context, runtime *common.RuntimeContext) *common.DryRunAPI {
 		token, _ := resolveSpreadsheetToken(runtime)
 		sheetID, sheetName, _ := resolveSheetSelector(runtime)
@@ -625,6 +776,43 @@ func warnDropdownSourceRangeHighlight(runtime *common.RuntimeContext) {
 // and returns its row / column counts. Errors on non-rectangular forms like
 // "A:C" (whole-column) or "3:6" (whole-row) — those need a row/col total
 // from get_sheet_structure, outside the scope of pure local parsing.
+// checkCellsMatchRange rejects, before any network call, the cells-vs-range
+// mismatches the server would otherwise fail mid-batch ("cells row count (N)
+// does not match range row count (M)" — a recurring server-side error cluster
+// in eval traces, and the failure leaves earlier batch sub-ops applied).
+// Single-cell ranges are checked too: the server enforces the same strict
+// match on a bare "A1" (07-21 rerun, 12 rows against range row count 1) —
+// there is no anchor semantics on +cells-set. An unparsable range is the
+// range validator's job, not ours.
+func checkCellsMatchRange(cells []interface{}, rangeStr string) error {
+	if len(cells) == 0 {
+		return sheetsValidationForFlag("cells",
+			"--cells is empty; to clear values use +cells-clear --scope content (needs --yes), or pass a non-empty 2D array")
+	}
+	rows, cols, err := rangeDimensions(rangeStr)
+	if err != nil {
+		return nil //nolint:nilerr // an unparsable range is reported by the range validation path with proper context
+	}
+	if len(cells) != rows {
+		return sheetsValidationForFlag("cells",
+			"--cells has %d rows but --range %q spans %d rows; make them equal (e.g. write N rows to an N-row range)",
+			len(cells), rangeStr, rows)
+	}
+	for r, rowRaw := range cells {
+		row, ok := rowRaw.([]interface{})
+		if !ok {
+			return sheetsValidationForFlag("cells",
+				"--cells[%d] must be an array (one row of cells) — --cells is always a 2D array, a single cell is [[{…}]]", r)
+		}
+		if len(row) != cols {
+			return sheetsValidationForFlag("cells",
+				"--cells[%d] has %d columns but --range %q spans %d columns; every row must match the range width",
+				r, len(row), rangeStr, cols)
+		}
+	}
+	return nil
+}
+
 func rangeDimensions(rangeStr string) (rows, cols int, err error) {
 	if idx := strings.Index(rangeStr, "!"); idx >= 0 {
 		rangeStr = rangeStr[idx+1:]

shortcuts/sheets/read_output.go

@@ -0,0 +1,71 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package sheets
+
+import (
+	"bytes"
+	"encoding/json"
+	"strings"
+
+	"github.com/larksuite/cli/extension/fileio"
+	"github.com/larksuite/cli/shortcuts/common"
+)
+
+// ─── lark_sheet read → file offload ───────────────────────────────────
+//
+// Shared plumbing for +cells-get / +csv-get / +table-get behind the
+// --output-path flag: when a caller redirects a read to a file, the char cap
+// should default to unlimited so the whole sheet lands on disk instead of being
+// clipped by the stdout-oriented max_chars safety cap.
+
+// readOutputPath returns the trimmed --output-path flag value ("" when unset).
+func readOutputPath(runtime *common.RuntimeContext) string {
+	return strings.TrimSpace(runtime.Str("output-path"))
+}
+
+// maxCharsInput resolves the max_chars value to send to the underlying read
+// tool. With --output-path set the cap is lifted (unbounded sentinel) so the
+// full result is written to the file; otherwise the --max-chars value binds.
+// The second return is false when nothing should be sent (max-chars <= 0), in
+// which case the tool's own default applies. Note the tool truncates at ~50000
+// even when max_chars is omitted, so callers that want an explicit cap should
+// pass a positive default.
+func maxCharsInput(runtime *common.RuntimeContext) (int, bool) {
+	if readOutputPath(runtime) != "" {
+		return unboundedReadLimit, true
+	}
+	if n := runtime.Int("max-chars"); n > 0 {
+		return n, true
+	}
+	return 0, false
+}
+
+// emitReadResult delivers a read shortcut's result. When --output-path is set it
+// writes the data payload to that path as pretty JSON and prints a small
+// confirmation envelope to stdout (path + byte count); otherwise it prints the
+// full result envelope to stdout as usual.
+func emitReadResult(runtime *common.RuntimeContext, out interface{}) error {
+	path := readOutputPath(runtime)
+	if path == "" {
+		runtime.Out(out, nil)
+		return nil
+	}
+	b, err := json.MarshalIndent(out, "", "  ")
+	if err != nil {
+		return err
+	}
+	b = append(b, '\n')
+	if _, err := runtime.FileIO().Save(path, fileio.SaveOptions{}, bytes.NewReader(b)); err != nil {
+		return err
+	}
+	resolved, err := runtime.FileIO().ResolvePath(path)
+	if err != nil {
+		resolved = path
+	}
+	runtime.Out(map[string]interface{}{
+		"output_path":   resolved,
+		"bytes_written": len(b),
+	}, nil)
+	return nil
+}

shortcuts/sheets/sheet_ai_api.go

@@ -7,6 +7,7 @@ import (
 	"context"
 	"encoding/json"
 	"fmt"
+	"strings"
 
 	"github.com/larksuite/cli/errs"
 	"github.com/larksuite/cli/internal/util"
@@ -83,7 +84,7 @@ func callTool(
 	code, _ := util.ToFloat64(envelope["code"])
 	if code != 0 {
 		msg, _ := envelope["msg"].(string)
-		return nil, errs.NewAPIError(errs.SubtypeServerError, "tool %q failed: [%d] %s", toolName, int(code), msg).
+		return nil, errs.NewAPIError(errs.SubtypeServerError, "tool %q failed: [%d] %s", toolName, int(code), flattenToolErrorMsg(msg)).
 			WithCode(int(code))
 	}
 	data, _ := envelope["data"].(map[string]interface{})
@@ -100,6 +101,66 @@ func callTool(
 	return out, nil
 }
 
+// flattenToolErrorMsg unwraps the nested-escaped-JSON error payload some
+// sheet-ai tools put in msg — batch_update in particular wraps its result as
+// {"error":"{\"message\":\"batch_update: N succeeded, M failed\",
+// \"failures\":[…]}","errorType":…,"data":{…}} — into one readable line
+// naming each failed operation. Eval traces show agents (and even the eval
+// aggregator) failing to extract the real cause from the double-escaped
+// form. Anything that doesn't match the nested shape passes through
+// untouched.
+func flattenToolErrorMsg(msg string) string {
+	trimmed := strings.TrimSpace(msg)
+	if !strings.HasPrefix(trimmed, "{") {
+		return msg
+	}
+	var outer struct {
+		Error string `json:"error"`
+	}
+	if json.Unmarshal([]byte(trimmed), &outer) != nil || strings.TrimSpace(outer.Error) == "" {
+		return msg
+	}
+	inner := strings.TrimSpace(outer.Error)
+	var detail struct {
+		Message  string `json:"message"`
+		Failures []struct {
+			Index    int    `json:"index"`
+			ToolName string `json:"tool_name"`
+			Error    string `json:"error"`
+		} `json:"failures"`
+	}
+	if strings.HasPrefix(inner, "{") && json.Unmarshal([]byte(inner), &detail) == nil && detail.Message != "" {
+		if len(detail.Failures) == 0 {
+			return detail.Message
+		}
+		parts := make([]string, 0, len(detail.Failures))
+		firstFailed := detail.Failures[0].Index
+		for _, f := range detail.Failures {
+			parts = append(parts, fmt.Sprintf("operations[%d] (%s): %s", f.Index, f.ToolName, f.Error))
+			if f.Index < firstFailed {
+				firstFailed = f.Index
+			}
+		}
+		out := detail.Message + " — " + strings.Join(parts, "; ")
+		// Partial failure is NOT rolled back server-side: the succeeded sub-ops
+		// stay applied. Spell out the recovery so agents don't resend the whole
+		// batch and double-apply the successes (observed in eval traces). With
+		// one failure (fail-fast) everything before it succeeded and nothing
+		// after it ran — resend from that index; with several (continue-on-error)
+		// only the listed failures need resending.
+		if strings.Contains(detail.Message, "succeeded") &&
+			!strings.Contains(detail.Message, " 0 succeeded") {
+			if len(detail.Failures) == 1 {
+				out += fmt.Sprintf("; note: succeeded operations stay applied (no rollback) — fix the failure and resend only operations[%d:] onward, do not resend the whole batch", firstFailed)
+			} else {
+				out += "; note: succeeded operations stay applied (no rollback) — fix and resend only the failed operations listed above, do not resend the whole batch"
+			}
+		}
+		return out
+	}
+	return inner
+}
+
 // invokeToolDryRun renders the One-OpenAPI request the shortcut would send.
 // The wire-format body (with input serialized to a JSON string) is preserved
 // for fidelity, and a decoded tool_input map is surfaced alongside so humans

shortcuts/sheets/sheet_ai_api_flatten_test.go

@@ -0,0 +1,57 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package sheets
+
+import (
+	"strings"
+	"testing"
+)
+
+// TestFlattenToolErrorMsg pins the unwrap of batch_update's double-escaped
+// error payload (the exact shape from eval V2U038/V2U013 traces) and the
+// pass-through of everything else.
+func TestFlattenToolErrorMsg(t *testing.T) {
+	t.Parallel()
+
+	t.Run("batch failures flatten to one line", func(t *testing.T) {
+		t.Parallel()
+		msg := `{"error":"{\"message\":\"batch_update: 0 succeeded, 1 failed\",\"succeeded\":0,\"failed\":1,\"failures\":[{\"index\":0,\"tool_name\":\"manage_chart_object\",\"error\":\"invalid snapshot.data.dim1.serie.index: 0, must be >= 1 (index is 1-based)\",\"errorType\":\"param_error\"}]}","errorType":"param_error","data":{"total":2,"succeeded":0,"failed":1}}`
+		got := flattenToolErrorMsg(msg)
+		for _, want := range []string{
+			"batch_update: 0 succeeded, 1 failed",
+			"operations[0] (manage_chart_object): invalid snapshot.data.dim1.serie.index",
+		} {
+			if !strings.Contains(got, want) {
+				t.Errorf("flattened msg should contain %q, got %q", want, got)
+			}
+		}
+		if strings.Contains(got, `\"`) {
+			t.Errorf("flattened msg must not carry escaped JSON, got %q", got)
+		}
+	})
+
+	t.Run("plain-string inner error unwraps", func(t *testing.T) {
+		t.Parallel()
+		got := flattenToolErrorMsg(`{"error":"sheet \"s\" not found","errorType":"param_error"}`)
+		if got != `sheet "s" not found` {
+			t.Errorf("got %q", got)
+		}
+	})
+
+	t.Run("non-JSON msg passes through", func(t *testing.T) {
+		t.Parallel()
+		msg := `cell at row 0, col 1 is inside a merged region (top-left: A1)`
+		if got := flattenToolErrorMsg(msg); got != msg {
+			t.Errorf("got %q", got)
+		}
+	})
+
+	t.Run("JSON without error field passes through", func(t *testing.T) {
+		t.Parallel()
+		msg := `{"detail":"x"}`
+		if got := flattenToolErrorMsg(msg); got != msg {
+			t.Errorf("got %q", got)
+		}
+	})
+}

shortcuts/sheets/shortcuts.go

@@ -35,6 +35,11 @@ func Shortcuts() []common.Shortcut {
 		if hasFlag(all[i].Flags, "spreadsheet-token") {
 			all[i].PostMount = withTokenAlias(all[i].PostMount)
 		}
+		// +chart-create grows --print-example (minimal per-type --properties
+		// templates) — the biggest --print-schema consumer in eval traces.
+		if all[i].Command == "+chart-create" {
+			all[i].PostMount = withChartPrintExample(all[i].PostMount)
+		}
 		// Sheets-scoped flag ergonomics (unknown-flag hints with the valid
 		// flags inlined, enum vocabulary normalization) ride the same
 		// PostMount composition, so no other domain's behavior shifts.
@@ -153,6 +158,9 @@ func shortcutList() []common.Shortcut {
 		SparklineCreate, SparklineUpdate, SparklineDelete,
 		FloatImageCreate, FloatImageUpdate, FloatImageDelete,
 
+		// lark_sheet_styles_put
+		StylesPut,
+
 		// lark_sheet_batch_update
 		BatchUpdate,
 		CellsBatchSetStyle,

shortcuts/sheets/style_vocab.go

@@ -0,0 +1,456 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package sheets
+
+import (
+	"fmt"
+	"slices"
+	"strings"
+
+	"github.com/larksuite/cli/shortcuts/common"
+)
+
+// ─── style vocabulary acceptance layer ────────────────────────────────
+//
+// The single home for how the sheets domain ACCEPTS style vocabulary, across
+// all three carrier paths that end in set_cell_range bodies:
+//
+//   flag path       +cells-set-style / +cells-batch-set-style flat flags
+//   typed cells     +cells-set --cells cell objects (incl. batch sub-ops)
+//   styles payload  --styles on +workbook-create / +table-put / +styles-put
+//
+// Design contract (established in the 2026-07 batch-update overhaul; see the
+// acceptance tests in styles_acceptance_test.go):
+//
+//   - ONE canonical form, documented; a WIDE acceptance layer, undocumented.
+//     Model priors are divergent (one eval batch produced six different
+//     border spellings), so no canonical structure can make first tries
+//     succeed — acceptance is normalized here instead, never per-call-site.
+//   - Every rewrite must be unambiguous; ambiguous guesses (fore_color) get
+//     a targeted prescription, never a silent pick. Silent ignoring and
+//     bare rejection are both bugs.
+//   - SILENT-ALIAS ADMISSION BAR (2026-07-21): only words from REAL external
+//     vocabularies (Excel/openpyxl, CSS, Google Sheets API), recurring
+//     across batches or ≥3 tasks in one, with zero semantic ambiguity.
+//     Spelling/word-order permutations do NOT get aliases — they are
+//     absorbed by the universal did-you-mean rejection (one self-healing
+//     retry, zero per-variant code). Real vocabularies are a finite set;
+//     permutations are not. Earlier permutation aliases are grandfathered.
+//   - Closure is enforced by two test properties: vocabulary parity (every
+//     flag-path style must be accepted on the payload paths) and the prior
+//     corpus (every observed model spelling either normalizes or
+//     prescribes). New eval finding → corpus row → fix HERE → locked.
+
+// ─── style flags (shared by +cells-set-style and +cells-batch-set-style) ─
+
+// buildCellStyleFromFlags reads the 12 flat style flags and returns the
+// cell_styles map expected by set_cell_range. Skips any flag the user
+// didn't set so partial styles work.
+func buildCellStyleFromFlags(runtime flagView) map[string]interface{} {
+	style := map[string]interface{}{}
+	if v := runtime.Str("background-color"); v != "" {
+		style["background_color"] = v
+	}
+	if v := runtime.Str("font-color"); v != "" {
+		style["font_color"] = v
+	}
+	if v := runtime.Str("font-family"); v != "" {
+		style["font_family"] = v
+	}
+	if runtime.Changed("font-size") && runtime.Float64("font-size") > 0 {
+		style["font_size"] = runtime.Float64("font-size")
+	}
+	if v := runtime.Str("font-style"); v != "" {
+		style["font_style"] = v
+	}
+	if v := runtime.Str("font-weight"); v != "" {
+		style["font_weight"] = v
+	}
+	if v := runtime.Str("font-line"); v != "" {
+		style["font_line"] = v
+	}
+	if v := runtime.Str("horizontal-alignment"); v != "" {
+		style["horizontal_alignment"] = v
+	}
+	if v := runtime.Str("vertical-alignment"); v != "" {
+		style["vertical_alignment"] = v
+	}
+	if v := runtime.Str("word-wrap"); v != "" {
+		style["word_wrap"] = v
+	}
+	if v := runtime.Str("number-format"); v != "" {
+		style["number_format"] = v
+	}
+	return style
+}
+
+// cellStyleAliases maps shorthand cell_styles field names that models commonly
+// hallucinate (Excel / openpyxl / CSS conventions) onto the canonical field
+// names the backend expects. Only the unambiguous alignment shorthands are
+// aliased — they are the high-frequency miss; ambiguous guesses (e.g. "color",
+// "bg_color", "text_align") are intentionally left out so a wrong guess still
+// surfaces as an error rather than being silently reinterpreted.
+var cellStyleAliases = []struct{ alias, canonical string }{
+	{"horizontal_align", "horizontal_alignment"},
+	{"halign", "horizontal_alignment"},
+	{"vertical_align", "vertical_alignment"},
+	{"valign", "vertical_alignment"},
+	// wrap family: word_wrap is the sole wrap concept, no ambiguity. 07-20
+	// eval: wrap_text alone produced an 88-issue retry loop on --styles;
+	// wrap_strategy (the Google Sheets API word) followed on 07-21.
+	{"wrap_text", "word_wrap"},
+	{"text_wrap", "word_wrap"},
+	{"wrap_strategy", "word_wrap"},
+}
+
+// cellStyleEnumFields sources the enum vocabulary for enum-bearing
+// cell_styles fields from the +cells-set-style flag-defs, so the payload path
+// (--styles / typed --cells) validates and canonicalizes values the same way
+// the cobra flag path does. 07-20 eval: "vertical_alignment":"center" (CSS
+// vocabulary; Lark spells it "middle") passed the CLI and burned a
+// server-side round trip ~10 times — the flag path had normalized it since
+// round 2, the payload path never did.
+func cellStyleEnumFields() map[string][]string {
+	defs, err := loadFlagDefs()
+	if err != nil {
+		return nil
+	}
+	spec, ok := defs["+cells-set-style"]
+	if !ok {
+		return nil
+	}
+	out := map[string][]string{}
+	for _, df := range spec.Flags {
+		if df.Kind != "own" || df.Type != "string" || len(df.Enum) == 0 {
+			continue
+		}
+		out[strings.ReplaceAll(df.Name, "-", "_")] = df.Enum
+	}
+	return out
+}
+
+// normalizeCellStyleAliases renames known shorthand keys in a single
+// cell_styles map to their canonical equivalents, in place, so a model that
+// writes e.g. "horizontal_align" instead of "horizontal_alignment" still
+// applies the style instead of hitting an "unsupported field" error (--styles)
+// or having the field silently dropped by the backend (typed --cells). If both
+// the shorthand and its canonical key are present it returns a validation error
+// rather than picking one. It then canonicalizes enum VALUES (casing + known
+// cross-vocabulary aliases like CSS "center" → Lark "middle"; boolean
+// word_wrap → the enum) and rejects off-enum values client-side instead of
+// letting the server fail the whole batch. path labels the map for errors.
+func normalizeCellStyleAliases(style map[string]interface{}, path string) error {
+	if len(style) == 0 {
+		return nil
+	}
+	for _, a := range cellStyleAliases {
+		v, ok := style[a.alias]
+		if !ok {
+			continue
+		}
+		if _, exists := style[a.canonical]; exists {
+			return common.ValidationErrorf("%s.%s conflicts with %s; pass only %s", path, a.alias, a.canonical, a.canonical)
+		}
+		style[a.canonical] = v
+		delete(style, a.alias)
+	}
+	// fore_color is deliberately NOT aliased: in openpyxl vocabulary fgColor
+	// is the FILL color while a plain reading suggests the font color — a
+	// silent pick could color the wrong thing. Prescribe both options.
+	if _, has := style["fore_color"]; has {
+		return common.ValidationErrorf("%s.fore_color is ambiguous — use font_color for text color or background_color for the cell fill", path)
+	}
+	// Boolean wrap habit: true unambiguously means wrap on, false means off.
+	if b, isBool := style["word_wrap"].(bool); isBool {
+		if b {
+			style["word_wrap"] = "auto-wrap"
+		} else {
+			style["word_wrap"] = "overflow"
+		}
+	}
+	for field, enum := range cellStyleEnumFields() {
+		raw, has := style[field]
+		if !has {
+			continue
+		}
+		val, isStr := raw.(string)
+		if !isStr || val == "" || slices.Contains(enum, val) {
+			continue
+		}
+		if canon := canonicalEnumValue(val, enum); canon != "" {
+			style[field] = canon
+			continue
+		}
+		msg := fmt.Sprintf("%s.%s value %q is invalid (allowed: %s)", path, field, val, strings.Join(enum, ", "))
+		if match := closestEnumValue(val, enum); match != "" {
+			msg += fmt.Sprintf("; did you mean %q?", match)
+		}
+		return common.ValidationErrorf("%s", msg)
+	}
+	return nil
+}
+
+// normalizeTypedCellsStyleAliases walks a typed --cells 2D array and applies
+// normalizeCellStyleAliases to every cell's inline cell_styles object, so the
+// alignment shorthands are accepted on +cells-set the same as on --styles.
+// It also expands the border "all" shorthand and intercepts border_styles
+// mis-nested inside cell_styles — both server-rejected shapes that eval
+// traces show surviving CLI validation and costing a full network round
+// trip. Structure is checked leniently to match the pass-through contract:
+// any element that isn't the expected shape is skipped, not rejected.
+func normalizeTypedCellsStyleAliases(cells []interface{}, path string) error {
+	for r, rowRaw := range cells {
+		row, ok := rowRaw.([]interface{})
+		if !ok {
+			continue
+		}
+		for c, cellRaw := range row {
+			cell, ok := cellRaw.(map[string]interface{})
+			if !ok {
+				continue
+			}
+			// cells[][].style is the habitual spelling of cell_styles (recurring
+			// server-side 900015206 in eval traces) — rewrite when unambiguous.
+			if styleObj, isObj := cell["style"].(map[string]interface{}); isObj {
+				if _, has := cell["cell_styles"]; has {
+					return common.ValidationErrorf("%s[%d][%d].style conflicts with cell_styles; pass only cell_styles", path, r, c)
+				}
+				cell["cell_styles"] = styleObj
+				delete(cell, "style")
+			}
+			// cells[][].type is not a cell field; the value type is whatever the
+			// JSON value is. Reject with the fix instead of a server round trip.
+			if _, has := cell["type"]; has {
+				return common.ValidationErrorf("%s[%d][%d].type is not a cell field — the value type is inferred from the JSON value; control display format via cell_styles.number_format", path, r, c)
+			}
+			if bs, ok := cell["border_styles"].(map[string]interface{}); ok {
+				expandBorderAllShorthand(bs)
+			}
+			st, ok := cell["cell_styles"].(map[string]interface{})
+			if !ok {
+				continue
+			}
+			if _, misNested := st["border_styles"]; misNested {
+				return common.ValidationErrorf(
+					"%s[%d][%d].cell_styles.border_styles is not valid — border_styles is a top-level cell field, a sibling of cell_styles; move it up one level",
+					path, r, c)
+			}
+			if err := normalizeCellStyleAliases(st, fmt.Sprintf("%s[%d][%d].cell_styles", path, r, c)); err != nil {
+				return err
+			}
+		}
+	}
+	return nil
+}
+
+// expandBorderAllShorthand rewrites the "all" side shorthand — habitual from
+// Excel / openpyxl vocabulary, rejected by the backend — into the four
+// explicit sides, in place. An explicitly set side wins over the shorthand.
+// Applied on both the typed --cells path and the --styles path, so batch
+// sub-ops get the same rewrite as standalone calls.
+func expandBorderAllShorthand(border map[string]interface{}) {
+	if all, ok := border["all"]; ok {
+		for _, side := range []string{"top", "bottom", "left", "right"} {
+			if _, exists := border[side]; !exists {
+				border[side] = all
+			}
+		}
+		delete(border, "all")
+	}
+	// Weight vocabulary in the style slot ("thin"/"medium"/"thick" are the
+	// habitual Excel words; the largest residual styles cluster in the 07-21
+	// rerun wrote them into border_styles.<side>.style of the FULL nested
+	// form). A thin border always means a thin solid line: move the word to
+	// weight and default style to solid. Only when weight is absent — an
+	// explicit conflicting weight keeps the enum error path.
+	for _, raw := range border {
+		side, ok := raw.(map[string]interface{})
+		if !ok {
+			continue
+		}
+		s, _ := side["style"].(string)
+		switch strings.ToLower(s) {
+		case "thin", "medium", "thick":
+			if _, hasWeight := side["weight"]; !hasWeight {
+				side["weight"] = strings.ToLower(s)
+				side["style"] = "solid"
+			}
+		}
+	}
+}
+
+// borderStylesFromFlag parses --border-styles as a JSON object (top/bottom/
+// left/right with style sub-objects), expanding the "all" side shorthand the
+// same as the typed --cells and --styles paths so +cells-set-style /
+// +cells-batch-set-style don't ship {"all":…} for the backend to reject.
+// Returns nil when the flag is empty.
+func borderStylesFromFlag(runtime flagView) (map[string]interface{}, error) {
+	if runtime.Str("border-styles") == "" {
+		return nil, nil
+	}
+	v, err := parseJSONFlag(runtime, "border-styles")
+	if err != nil {
+		return nil, err
+	}
+	m, ok := v.(map[string]interface{})
+	if !ok {
+		return nil, sheetsValidationForFlag("border-styles", "--border-styles must be a JSON object")
+	}
+	expandBorderAllShorthand(m)
+	return m, nil
+}
+
+// requireAnyStyleFlag ensures at least one style-defining flag (style or
+// border) is set — otherwise the request would do nothing.
+func requireAnyStyleFlag(runtime flagView) error {
+	if len(buildCellStyleFromFlags(runtime)) > 0 {
+		return nil
+	}
+	if runtime.Str("border-styles") != "" {
+		return nil
+	}
+	return common.ValidationErrorf("at least one style flag is required (e.g. --background-color, --font-weight, --border-styles)").
+		WithParams(
+			sheetsInvalidParam("background-color", "required; specify at least one style flag"),
+			sheetsInvalidParam("font-weight", "required; specify at least one style flag"),
+			sheetsInvalidParam("border-styles", "required; specify at least one style flag"),
+		)
+}
+
+// foldBorderFamilyAliases rewrites the habitual flattened border vocabulary
+// (Excel / openpyxl conventions) into the canonical nested border_styles
+// object, in place. 07-20 eval: the border family alone accounted for the
+// largest --styles error cluster (borders / border / border_bottom /
+// border_style / border_top_color / …), each burning a full payload retry.
+// Accepted rewrites, all unambiguous:
+//
+//	borders / border   (object) → border_styles (side-keyed) or border_styles.all (attr-keyed)
+//	border_top|bottom|left|right (object) → border_styles.<side>
+//	border_style|color|weight  (scalar)  → border_styles.all.<attr>
+//	border_<side>_<style|color|weight> (scalar) → border_styles.<side>.<attr>
+//
+// A border_style value from the WEIGHT vocabulary (thin/medium/thick — the
+// habitual Excel word) sets weight and defaults style to solid: a "thin
+// border" always means a thin solid line. Conflicts with an explicitly given
+// border_styles error out instead of picking a side.
+func foldBorderFamilyAliases(in map[string]interface{}, path string) error {
+	sides := map[string]bool{"top": true, "bottom": true, "left": true, "right": true, "all": true}
+	attrs := map[string]bool{"style": true, "color": true, "weight": true}
+	borderWeights := map[string]bool{"thin": true, "medium": true, "thick": true}
+
+	ensureBorder := func() map[string]interface{} {
+		bs, ok := in["border_styles"].(map[string]interface{})
+		if !ok {
+			bs = map[string]interface{}{}
+			in["border_styles"] = bs
+		}
+		return bs
+	}
+	setSideAttr := func(side, attr string, v interface{}, from string) error {
+		bs := ensureBorder()
+		sideObj, ok := bs[side].(map[string]interface{})
+		if !ok {
+			if _, exists := bs[side]; exists {
+				return common.ValidationErrorf("%s.%s conflicts with border_styles.%s; keep one form", path, from, side)
+			}
+			sideObj = map[string]interface{}{}
+			bs[side] = sideObj
+		}
+		if _, exists := sideObj[attr]; exists {
+			return common.ValidationErrorf("%s.%s conflicts with border_styles.%s.%s; keep one form", path, from, side, attr)
+		}
+		sideObj[attr] = v
+		return nil
+	}
+	setSide := func(side string, v interface{}, from string) error {
+		obj, ok := v.(map[string]interface{})
+		if !ok {
+			return common.ValidationErrorf("%s.%s must be an object like {\"style\":\"solid\",\"color\":\"#000000\"}", path, from)
+		}
+		for attr, av := range obj {
+			if !attrs[attr] {
+				return common.ValidationErrorf("%s.%s.%s is not a border attribute (want style/weight/color)", path, from, attr)
+			}
+			if err := setSideAttr(side, attr, av, from); err != nil {
+				return err
+			}
+		}
+		return nil
+	}
+	// border_style with a weight-vocabulary value means "thin solid line".
+	setAllScalar := func(attr string, v interface{}, from string) error {
+		if attr == "style" {
+			if s, ok := v.(string); ok && borderWeights[strings.ToLower(s)] {
+				if err := setSideAttr("all", "weight", strings.ToLower(s), from); err != nil {
+					return err
+				}
+				return setSideAttr("all", "style", "solid", from)
+			}
+		}
+		return setSideAttr("all", attr, v, from)
+	}
+
+	for _, key := range []string{"borders", "border"} {
+		v, has := in[key]
+		if !has {
+			continue
+		}
+		obj, ok := v.(map[string]interface{})
+		if !ok {
+			return common.ValidationErrorf("%s.%s must be an object — either side-keyed ({\"top\":{…},\"bottom\":{…}} / {\"all\":{…}}) or attribute-keyed ({\"style\":\"solid\",\"color\":\"#000\"} = all four sides)", path, key)
+		}
+		sideKeyed := false
+		for k := range obj {
+			if sides[k] {
+				sideKeyed = true
+				break
+			}
+		}
+		if sideKeyed {
+			for side, sv := range obj {
+				if !sides[side] {
+					return common.ValidationErrorf("%s.%s.%s is not a valid side (want top/bottom/left/right/all)", path, key, side)
+				}
+				if err := setSide(side, sv, key); err != nil {
+					return err
+				}
+			}
+		} else if err := setSide("all", v, key); err != nil {
+			return err
+		}
+		delete(in, key)
+	}
+	for _, side := range []string{"top", "bottom", "left", "right"} {
+		// Both word orders appear in the wild: border_bottom (07-20 eval) and
+		// bottom_border (07-21), same for the flattened attribute triples.
+		for _, key := range []string{"border_" + side, side + "_border"} {
+			if v, has := in[key]; has {
+				if err := setSide(side, v, key); err != nil {
+					return err
+				}
+				delete(in, key)
+			}
+		}
+		for attr := range attrs {
+			for _, key := range []string{"border_" + side + "_" + attr, side + "_border_" + attr} {
+				if v, has := in[key]; has {
+					if err := setSideAttr(side, attr, v, key); err != nil {
+						return err
+					}
+					delete(in, key)
+				}
+			}
+		}
+	}
+	for attr := range attrs {
+		key := "border_" + attr
+		if v, has := in[key]; has {
+			if err := setAllScalar(attr, v, key); err != nil {
+				return err
+			}
+			delete(in, key)
+		}
+	}
+	return nil
+}

shortcuts/sheets/styles_acceptance_test.go

@@ -0,0 +1,410 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package sheets
+
+import (
+	"fmt"
+	"strings"
+	"testing"
+)
+
+// ─── styles acceptance contract ───────────────────────────────────────
+//
+// Two closure properties that turn the --styles acceptance surface from
+// "endless patching" into a locked contract (07-20 rerun lesson: the
+// redesign moved traffic onto the payload path while the flag path's
+// forgiveness layers stayed behind):
+//
+//  1. Vocabulary parity — every style the flag path (+cells-set-style)
+//     can express must be accepted verbatim by the payload path.
+//  2. Prior corpus — every model spelling observed in eval traces must
+//     either normalize to the canonical form or produce a targeted
+//     prescription. Silent ignoring and bare rejection are both bugs.
+//     New eval finding → add a corpus row → fix → locked forever.
+
+// acceptStyleItem runs one cell_styles item through the styles-put pipeline
+// and returns the emitted cell prototype (cell_styles/border_styles) or the
+// error.
+func acceptStyleItem(t *testing.T, fields map[string]interface{}) (map[string]interface{}, error) {
+	t.Helper()
+	item := map[string]interface{}{"range": "A1:B2"}
+	for k, v := range fields {
+		item[k] = v
+	}
+	ops, err := stylesPutOperations(stylesPutView(map[string]interface{}{
+		"styles": []interface{}{map[string]interface{}{
+			"name":        "S1",
+			"cell_styles": []interface{}{item},
+		}},
+	}), testToken)
+	if err != nil {
+		return nil, err
+	}
+	input := ops[0].(map[string]interface{})["input"].(map[string]interface{})
+	cells := input["cells"].([][]interface{})
+	return cells[0][0].(map[string]interface{}), nil
+}
+
+// TestStylesAcceptance_VocabularyParity locks property 1: iterate the
+// +cells-set-style flag vocabulary from flag-defs and assert the payload
+// path accepts each field with a valid value and emits it.
+func TestStylesAcceptance_VocabularyParity(t *testing.T) {
+	t.Parallel()
+	defs, err := loadFlagDefs()
+	if err != nil {
+		t.Fatalf("loadFlagDefs: %v", err)
+	}
+	spec, ok := defs["+cells-set-style"]
+	if !ok {
+		t.Fatal("no +cells-set-style flag defs")
+	}
+	sample := func(df flagDef) interface{} {
+		if len(df.Enum) > 0 {
+			return df.Enum[0]
+		}
+		switch df.Type {
+		case "float64", "int":
+			return float64(12)
+		}
+		switch df.Name {
+		case "font-family":
+			return "Arial"
+		case "number-format":
+			return "0.00"
+		default: // colors and any future string field
+			return "#112233"
+		}
+	}
+	for _, df := range spec.Flags {
+		if df.Kind != "own" || df.Name == "range" {
+			continue
+		}
+		df := df
+		t.Run(df.Name, func(t *testing.T) {
+			t.Parallel()
+			field := strings.ReplaceAll(df.Name, "-", "_")
+			var value interface{}
+			if df.Name == "border-styles" {
+				value = map[string]interface{}{"all": map[string]interface{}{"style": "solid"}}
+			} else {
+				value = sample(df)
+			}
+			proto, err := acceptStyleItem(t, map[string]interface{}{field: value})
+			if err != nil {
+				t.Fatalf("payload path rejects flag-path field %s: %v", field, err)
+			}
+			if df.Name == "border-styles" {
+				if _, ok := proto["border_styles"].(map[string]interface{}); !ok {
+					t.Fatalf("border_styles not emitted: %v", proto)
+				}
+				return
+			}
+			cs, _ := proto["cell_styles"].(map[string]interface{})
+			if cs == nil || cs[field] == nil {
+				t.Fatalf("field %s silently dropped: %v", field, proto)
+			}
+		})
+	}
+}
+
+// stylesPriorCorpus is the observed-model-spelling corpus (source: eval
+// batches 2026-07-08 → 07-20). Every row must either normalize (checked via
+// wantCell) or produce a targeted prescription (wantErr). Add a row for every
+// new spelling an eval surfaces — never let one be silently ignored.
+var stylesPriorCorpus = []struct {
+	name    string
+	fields  map[string]interface{}
+	wantErr string                                    // "" = must be accepted
+	check   func(proto map[string]interface{}) string // "" = ok, else failure detail
+}{
+	// border family (07-20: largest cluster)
+	{name: "borders attr-keyed means all sides",
+		fields: map[string]interface{}{"borders": map[string]interface{}{"style": "solid", "color": "#DDDDDD"}},
+		check:  wantBorder("top", "style", "solid")},
+	{name: "border side-keyed",
+		fields: map[string]interface{}{"border": map[string]interface{}{"top": map[string]interface{}{"style": "solid"}}},
+		check:  wantBorder("top", "style", "solid")},
+	{name: "border_bottom object",
+		fields: map[string]interface{}{"border_bottom": map[string]interface{}{"style": "solid"}},
+		check:  wantBorder("bottom", "style", "solid")},
+	{name: "border_style weight-vocabulary means thin solid",
+		fields: map[string]interface{}{"border_style": "thin"},
+		check:  wantBorder("top", "weight", "thin")},
+	{name: "border_style style-vocabulary",
+		fields: map[string]interface{}{"border_style": "dashed"},
+		check:  wantBorder("top", "style", "dashed")},
+	{name: "border_color scalar",
+		fields: map[string]interface{}{"border_color": "#FF0000"},
+		check:  wantBorder("top", "color", "#FF0000")},
+	{name: "border_top_color flattened",
+		fields: map[string]interface{}{"border_top_color": "#FF0000"},
+		check:  wantBorder("top", "color", "#FF0000")},
+	{name: "border_left_weight flattened",
+		fields: map[string]interface{}{"border_left_weight": "thin"},
+		check:  wantBorder("left", "weight", "thin")},
+	{name: "border_styles invalid side prescribed",
+		fields:  map[string]interface{}{"border_styles": map[string]interface{}{"outer": map[string]interface{}{"style": "solid"}}},
+		wantErr: "not a valid side"},
+	// wrap family
+	{name: "wrap_text boolean", fields: map[string]interface{}{"wrap_text": true}, check: wantStyle("word_wrap", "auto-wrap")},
+	{name: "text_wrap string", fields: map[string]interface{}{"text_wrap": "auto-wrap"}, check: wantStyle("word_wrap", "auto-wrap")},
+	{name: "word_wrap false", fields: map[string]interface{}{"word_wrap": false}, check: wantStyle("word_wrap", "overflow")},
+	// alignment family
+	{name: "horizontal_align shorthand", fields: map[string]interface{}{"horizontal_align": "center"}, check: wantStyle("horizontal_alignment", "center")},
+	{name: "valign shorthand", fields: map[string]interface{}{"valign": "top"}, check: wantStyle("vertical_alignment", "top")},
+	{name: "CSS center for vertical", fields: map[string]interface{}{"vertical_alignment": "center"}, check: wantStyle("vertical_alignment", "middle")},
+	{name: "casing normalized", fields: map[string]interface{}{"font_weight": "BOLD"}, check: wantStyle("font_weight", "bold")},
+	// weight vocabulary in the FULL nested form's style slot (07-21 rerun:
+	// the dominant residual — 8 tasks wrote border_styles.<side>.style:"thin")
+	{name: "full-form thin in style slot",
+		fields: map[string]interface{}{"border_styles": map[string]interface{}{"top": map[string]interface{}{"style": "thin"}}},
+		check:  wantBorder("top", "weight", "thin")},
+	{name: "full-form all-shorthand medium in style slot",
+		fields: map[string]interface{}{"border_styles": map[string]interface{}{"all": map[string]interface{}{"style": "medium"}}},
+		check:  wantBorder("bottom", "weight", "medium")},
+	// side-first word order + Google Sheets wrap word (07-21 evening batch)
+	{name: "side-first bottom_border object",
+		fields: map[string]interface{}{"bottom_border": map[string]interface{}{"style": "solid"}},
+		check:  wantBorder("bottom", "style", "solid")},
+	{name: "side-first bottom_border_style scalar",
+		fields: map[string]interface{}{"bottom_border_style": "solid"},
+		check:  wantBorder("bottom", "style", "solid")},
+	{name: "wrap_strategy aliases to word_wrap",
+		fields: map[string]interface{}{"wrap_strategy": "auto-wrap"},
+		check:  wantStyle("word_wrap", "auto-wrap")},
+	// prescriptions (ambiguous / unsupported / typo)
+	{name: "fore_color prescribed", fields: map[string]interface{}{"fore_color": "#F00"}, wantErr: "ambiguous"},
+	{name: "indent rejected not ignored", fields: map[string]interface{}{"indent": float64(2)}, wantErr: "not a supported style field"},
+	{name: "unknown field carries did-you-mean and the field list",
+		fields: map[string]interface{}{"fontcolor": "#000000"}, wantErr: `did you mean "font_color"`},
+	{name: "enum typo gets did-you-mean", fields: map[string]interface{}{"vertical_alignment": "botom"}, wantErr: "did you mean"},
+}
+
+func wantStyle(field, want string) func(map[string]interface{}) string {
+	return func(proto map[string]interface{}) string {
+		cs, _ := proto["cell_styles"].(map[string]interface{})
+		if cs == nil || cs[field] != want {
+			return fmt.Sprintf("cell_styles.%s = %v, want %q", field, cs[field], want)
+		}
+		return ""
+	}
+}
+
+func wantBorder(side, attr, want string) func(map[string]interface{}) string {
+	return func(proto map[string]interface{}) string {
+		bs, _ := proto["border_styles"].(map[string]interface{})
+		sideObj, _ := bs[side].(map[string]interface{})
+		if sideObj == nil || sideObj[attr] != want {
+			return fmt.Sprintf("border_styles.%s.%s = %v, want %q", side, attr, sideObj[attr], want)
+		}
+		return ""
+	}
+}
+
+func TestStylesAcceptance_PriorCorpus(t *testing.T) {
+	t.Parallel()
+	for _, tc := range stylesPriorCorpus {
+		tc := tc
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			proto, err := acceptStyleItem(t, tc.fields)
+			if tc.wantErr != "" {
+				if err == nil || !strings.Contains(err.Error(), tc.wantErr) {
+					t.Fatalf("want prescription containing %q, got err=%v", tc.wantErr, err)
+				}
+				return
+			}
+			if err != nil {
+				t.Fatalf("corpus spelling rejected: %v", err)
+			}
+			if detail := tc.check(proto); detail != "" {
+				t.Fatal(detail)
+			}
+		})
+	}
+}
+
+// TestStylesPut_CoalescesSameStyleRanges pins the declarative-spec
+// optimization: per-row entries with the identical style fuse into one
+// rectangle, so row-by-row specs (07-21 rerun: 184/203/861-op expansions
+// against the 100-op cap) no longer hit the cap.
+func TestStylesPut_CoalescesSameStyleRanges(t *testing.T) {
+	t.Parallel()
+
+	t.Run("150 same-style rows fuse into one stamp", func(t *testing.T) {
+		t.Parallel()
+		entries := make([]interface{}, 0, 150)
+		for r := 1; r <= 150; r++ {
+			entries = append(entries, map[string]interface{}{
+				"range": fmt.Sprintf("A%d:F%d", r, r), "font_weight": "bold",
+			})
+		}
+		ops, err := stylesPutOperations(stylesPutView(map[string]interface{}{
+			"styles": []interface{}{map[string]interface{}{"name": "S1", "cell_styles": entries}},
+		}), testToken)
+		if err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+		if len(ops) != 1 {
+			t.Fatalf("got %d ops, want 1 fused stamp", len(ops))
+		}
+		input := ops[0].(map[string]interface{})["input"].(map[string]interface{})
+		if input["range"] != "A1:F150" {
+			t.Fatalf("range = %v, want A1:F150", input["range"])
+		}
+	})
+
+	t.Run("different styles stay separate", func(t *testing.T) {
+		t.Parallel()
+		ops, err := stylesPutOperations(stylesPutView(map[string]interface{}{
+			"styles": []interface{}{map[string]interface{}{"name": "S1", "cell_styles": []interface{}{
+				map[string]interface{}{"range": "A1:F1", "font_weight": "bold"},
+				map[string]interface{}{"range": "A2:F2", "background_color": "#EEEEEE"},
+			}}},
+		}), testToken)
+		if err != nil || len(ops) != 2 {
+			t.Fatalf("ops=%d err=%v, want 2", len(ops), err)
+		}
+	})
+
+	t.Run("horizontal fuse with same rows", func(t *testing.T) {
+		t.Parallel()
+		ops, err := stylesPutOperations(stylesPutView(map[string]interface{}{
+			"styles": []interface{}{map[string]interface{}{"name": "S1", "cell_styles": []interface{}{
+				map[string]interface{}{"range": "A1:C5", "font_weight": "bold"},
+				map[string]interface{}{"range": "D1:F5", "font_weight": "bold"},
+			}}},
+		}), testToken)
+		if err != nil || len(ops) != 1 {
+			t.Fatalf("ops=%d err=%v, want 1", len(ops), err)
+		}
+		input := ops[0].(map[string]interface{})["input"].(map[string]interface{})
+		if input["range"] != "A1:F5" {
+			t.Fatalf("range = %v, want A1:F5", input["range"])
+		}
+	})
+}
+
+// TestTypedCellsHabitualKeys pins the typed --cells cell-object fixes
+// (recurring server-side 900015206 across 07-20/07-21 reruns).
+func TestTypedCellsHabitualKeys(t *testing.T) {
+	t.Parallel()
+
+	t.Run("style object rewrites to cell_styles through batch", func(t *testing.T) {
+		t.Parallel()
+		translated, err := translateBatchOp(subOp("+cells-set", map[string]interface{}{
+			"sheet_name": "S1", "range": "A1",
+			"cells": []interface{}{[]interface{}{map[string]interface{}{
+				"value": "x", "style": map[string]interface{}{"font_weight": "bold"},
+			}}},
+		}), testToken, 0)
+		if err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+		input := translated["input"].(map[string]interface{})
+		cell := input["cells"].([]interface{})[0].([]interface{})[0].(map[string]interface{})
+		cs, _ := cell["cell_styles"].(map[string]interface{})
+		if cs == nil || cs["font_weight"] != "bold" {
+			t.Fatalf("cell = %v, want cell_styles.font_weight bold", cell)
+		}
+		if _, has := cell["style"]; has {
+			t.Fatalf("style key must be renamed, got %v", cell)
+		}
+	})
+
+	t.Run("type key gets a prescription", func(t *testing.T) {
+		t.Parallel()
+		_, err := translateBatchOp(subOp("+cells-set", map[string]interface{}{
+			"sheet_name": "S1", "range": "A1",
+			"cells": []interface{}{[]interface{}{map[string]interface{}{
+				"value": "x", "type": "text",
+			}}},
+		}), testToken, 0)
+		requireValidation(t, err, "not a cell field")
+	})
+}
+
+// TestStylesAcceptance_ResizeAndMergeCorpus extends the corpus to the
+// row/col_sizes and cell_merges sections.
+func TestStylesAcceptance_ResizeAndMergeCorpus(t *testing.T) {
+	t.Parallel()
+
+	runSection := func(section string, entry interface{}) ([]interface{}, error) {
+		return stylesPutOperations(stylesPutView(map[string]interface{}{
+			"styles": []interface{}{map[string]interface{}{
+				"name":  "S1",
+				section: []interface{}{entry},
+			}},
+		}), testToken)
+	}
+	pixelValue := func(t *testing.T, ops []interface{}, key string) interface{} {
+		t.Helper()
+		input := ops[0].(map[string]interface{})["input"].(map[string]interface{})
+		block, _ := input[key].(map[string]interface{})
+		if block == nil || block["type"] != "pixel" {
+			t.Fatalf("%s = %v, want pixel block", key, input[key])
+		}
+		return block["value"]
+	}
+
+	t.Run("size alone implies pixel", func(t *testing.T) {
+		t.Parallel()
+		ops, err := runSection("row_sizes", map[string]interface{}{"range": "1:1", "size": float64(36)})
+		if err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+		if v := pixelValue(t, ops, "resize_height"); v != 36 {
+			t.Fatalf("value = %v, want 36", v)
+		}
+	})
+
+	t.Run("width alone implies pixel on col_sizes", func(t *testing.T) {
+		t.Parallel()
+		ops, err := runSection("col_sizes", map[string]interface{}{"range": "A:C", "width": float64(120)})
+		if err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+		if v := pixelValue(t, ops, "resize_width"); v != 120 {
+			t.Fatalf("value = %v, want 120", v)
+		}
+	})
+
+	t.Run("type auto still works on rows", func(t *testing.T) {
+		t.Parallel()
+		if _, err := runSection("row_sizes", map[string]interface{}{"range": "1:1", "type": "auto"}); err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+	})
+
+	t.Run("neither size nor type prescribed", func(t *testing.T) {
+		t.Parallel()
+		_, err := runSection("row_sizes", map[string]interface{}{"range": "1:1"})
+		requireValidation(t, err, "needs size (px) or type")
+	})
+
+	t.Run("wrong-dimension word prescribed", func(t *testing.T) {
+		t.Parallel()
+		_, err := runSection("col_sizes", map[string]interface{}{"range": "A:C", "height": float64(36)})
+		requireValidation(t, err, "does not apply")
+	})
+
+	t.Run("raw OpenAPI merge_type accepted", func(t *testing.T) {
+		t.Parallel()
+		ops, err := runSection("cell_merges", map[string]interface{}{"range": "A1:B2", "merge_type": "MERGE_ALL"})
+		if err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+		input := ops[0].(map[string]interface{})["input"].(map[string]interface{})
+		if input["merge_type"] != "all" {
+			t.Fatalf("merge_type = %v, want all", input["merge_type"])
+		}
+	})
+
+	t.Run("bare string merge accepted", func(t *testing.T) {
+		t.Parallel()
+		if _, err := runSection("cell_merges", "A1:B2"); err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+	})
+}

shortcuts/sheets/styles_prescription_test.go

@@ -0,0 +1,250 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package sheets
+
+import (
+	"strings"
+	"testing"
+)
+
+// TestTablePut_StylesErrorsAggregate pins the one-retry contract for
+// --styles: every issue across sections and ops is reported in a single
+// error (eval V2U032 burned three round trips fixing a border side, then
+// row_sizes.type, then size — each surfaced only after the previous fix).
+func TestTablePut_StylesErrorsAggregate(t *testing.T) {
+	t.Parallel()
+	sc := shortcutFromRegistry(t, "+table-put")
+	_, _, err := runShortcutCapturingErr(t, sc, []string{
+		"--url", testURL,
+		"--sheets", `{"sheets":[{"name":"s","columns":["a"],"data":[["x"]]}]}`,
+		"--styles", `{"styles":[{"name":"s",
+			"cell_styles":[{"range":"A1:A1","border_styles":{"horizontal":{"style":"solid"}}}],
+			"row_sizes":[{"range":"1:1","type":"custom"}],
+			"col_sizes":[{"range":"A:A","type":"pixel"}]}]}`,
+		"--dry-run",
+	})
+	ve := requireValidation(t, err, "--styles has 3 issues")
+	for _, want := range []string{
+		"border_styles.horizontal is not a valid side",
+		`row_sizes[0].type "custom" is invalid`,
+		"col_sizes[0].type pixel requires size",
+	} {
+		if !strings.Contains(ve.Message, want) {
+			t.Errorf("aggregated message should contain %q, got %q", want, ve.Message)
+		}
+	}
+	// D2: each type/size error inlines a complete valid op.
+	if !strings.Contains(ve.Message, `{"range":"2:10","type":"pixel","size":32}`) {
+		t.Errorf("row_sizes error should inline a full valid example, got %q", ve.Message)
+	}
+	if !strings.Contains(ve.Message, `{"range":"A:C","type":"pixel","size":120}`) {
+		t.Errorf("col_sizes error should inline a full valid example, got %q", ve.Message)
+	}
+}
+
+// TestTablePut_StylesBorderAllExpands verifies the "all" shorthand is
+// rewritten to four explicit sides instead of being rejected (or worse,
+// passed through for the server to reject, as happened on the typed-cells
+// path in eval V2U013/V2U021).
+func TestTablePut_StylesBorderAllExpands(t *testing.T) {
+	t.Parallel()
+	sc := shortcutFromRegistry(t, "+table-put")
+	stdout, _, err := runShortcutCapturingErr(t, sc, []string{
+		"--url", testURL,
+		"--sheets", `{"sheets":[{"name":"s","columns":["a"],"data":[["x"]]}]}`,
+		"--styles", `{"styles":[{"name":"s","cell_styles":[{"range":"A1:A1","border_styles":{"all":{"style":"solid","weight":"thin"}}}]}]}`,
+		"--dry-run",
+	})
+	if err != nil {
+		t.Fatalf("border all should expand to four sides and pass, got: %v", err)
+	}
+	// table-put's dry-run body carries the tool input as an escaped JSON
+	// string, so match the escaped key form.
+	for _, side := range []string{`\"top\"`, `\"bottom\"`, `\"left\"`, `\"right\"`} {
+		if !strings.Contains(stdout, side) {
+			t.Errorf("dry-run body should carry expanded side %s, got %q", side, stdout)
+		}
+	}
+	if strings.Contains(stdout, `\"all\"`) {
+		t.Errorf("dry-run body must not carry the raw all shorthand, got %q", stdout)
+	}
+}
+
+// TestCellsSet_BorderAllAndMisNestedBorder covers the typed --cells path:
+// the "all" shorthand expands CLI-side, and border_styles mis-nested inside
+// cell_styles is intercepted with a move-it prescription instead of a
+// server-side 900015206.
+func TestCellsSet_BorderAllAndMisNestedBorder(t *testing.T) {
+	t.Parallel()
+
+	t.Run("border all expands", func(t *testing.T) {
+		t.Parallel()
+		sc := shortcutFromRegistry(t, "+cells-set")
+		stdout, _, err := runShortcutCapturingErr(t, sc, []string{
+			"--url", testURL,
+			"--sheet-name", "s",
+			"--range", "A1",
+			"--cells", `[[{"value":"x","border_styles":{"all":{"style":"solid"}}}]]`,
+			"--dry-run",
+		})
+		if err != nil {
+			t.Fatalf("border all should expand and pass, got: %v", err)
+		}
+		if strings.Contains(stdout, `"all"`) || !strings.Contains(stdout, `"top"`) {
+			t.Errorf("dry-run body should carry expanded sides, got %q", stdout)
+		}
+	})
+
+	t.Run("mis-nested border_styles intercepted", func(t *testing.T) {
+		t.Parallel()
+		sc := shortcutFromRegistry(t, "+cells-set")
+		_, _, err := runShortcutCapturingErr(t, sc, []string{
+			"--url", testURL,
+			"--sheet-name", "s",
+			"--range", "A1",
+			"--cells", `[[{"value":"x","cell_styles":{"font_weight":"bold","border_styles":{"top":{"style":"solid"}}}}]]`,
+			"--dry-run",
+		})
+		ve := requireValidation(t, err, "cell_styles.border_styles is not valid")
+		if !strings.Contains(ve.Message, "sibling of cell_styles") {
+			t.Errorf("message should prescribe moving it up one level, got %q", ve.Message)
+		}
+	})
+}
+
+// TestCellsSetStyle_BorderAllExpands covers the --border-styles flag path
+// (+cells-set-style / +cells-batch-set-style go through borderStylesFromFlag,
+// not the typed --cells or --styles walkers): the "all" shorthand must expand
+// CLI-side here too, or the backend rejects {"all":…}.
+func TestCellsSetStyle_BorderAllExpands(t *testing.T) {
+	t.Parallel()
+	sc := shortcutFromRegistry(t, "+cells-set-style")
+	stdout, _, err := runShortcutCapturingErr(t, sc, []string{
+		"--url", testURL,
+		"--sheet-name", "s",
+		"--range", "A1:A1",
+		"--border-styles", `{"all":{"style":"solid","weight":"thin"}}`,
+		"--dry-run",
+	})
+	if err != nil {
+		t.Fatalf("border all should expand to four sides and pass, got: %v", err)
+	}
+	for _, side := range []string{`"top"`, `"bottom"`, `"left"`, `"right"`} {
+		if !strings.Contains(stdout, side) {
+			t.Errorf("dry-run body should carry expanded side %s, got %q", side, stdout)
+		}
+	}
+	if strings.Contains(stdout, `"all"`) {
+		t.Errorf("dry-run body must not carry the raw all shorthand, got %q", stdout)
+	}
+}
+
+// TestCellsMerge_RawAPIVocabularyNormalizes pins MERGE_ALL → all (the raw
+// OpenAPI enum agents copy from Lark API docs) via the enum alias table.
+func TestCellsMerge_RawAPIVocabularyNormalizes(t *testing.T) {
+	t.Parallel()
+	sc := shortcutFromRegistry(t, "+cells-merge")
+	stdout, _, err := runShortcutCapturingErr(t, sc, []string{
+		"--url", testURL,
+		"--sheet-name", "s",
+		"--range", "A1:B2",
+		"--merge-type", "MERGE_ALL",
+		"--dry-run",
+	})
+	if err != nil {
+		t.Fatalf("MERGE_ALL should normalize to all and pass, got: %v", err)
+	}
+	if !strings.Contains(stdout, `"all"`) {
+		t.Errorf("dry-run body should carry the normalized merge type, got %q", stdout)
+	}
+}
+
+// TestCellsSetStyle_WordWrapBooleanNormalizes pins --word-wrap true →
+// auto-wrap (eval V2U029).
+func TestCellsSetStyle_WordWrapBooleanNormalizes(t *testing.T) {
+	t.Parallel()
+	sc := shortcutFromRegistry(t, "+cells-set-style")
+	stdout, _, err := runShortcutCapturingErr(t, sc, []string{
+		"--url", testURL,
+		"--sheet-name", "s",
+		"--range", "A1:A1",
+		"--word-wrap", "true",
+		"--dry-run",
+	})
+	if err != nil {
+		t.Fatalf("--word-wrap true should normalize to auto-wrap, got: %v", err)
+	}
+	if !strings.Contains(stdout, "auto-wrap") {
+		t.Errorf("dry-run body should carry auto-wrap, got %q", stdout)
+	}
+}
+
+// TestUnderscoreFlagFormsParse pins the wire-vocabulary underscore rewrite:
+// --sheet_name / --border_styles parse as their hyphen forms (agents copy
+// field names out of JSON payloads where underscores are canonical).
+func TestUnderscoreFlagFormsParse(t *testing.T) {
+	t.Parallel()
+	sc := shortcutFromRegistry(t, "+cells-set-style")
+	stdout, _, err := runShortcutCapturingErr(t, sc, []string{
+		"--url", testURL,
+		"--sheet_name", "s",
+		"--range", "A1:A1",
+		"--font_weight", "bold",
+		"--dry-run",
+	})
+	if err != nil {
+		t.Fatalf("underscore flag forms should parse as hyphen forms, got: %v", err)
+	}
+	if !strings.Contains(stdout, "bold") {
+		t.Errorf("dry-run body should carry the style, got %q", stdout)
+	}
+}
+
+// TestPrintFlagSchema_UnderscoreFlagName pins --flag-name border_styles
+// resolving the border-styles schema (eval V2U013 burned a retry on this).
+func TestPrintFlagSchema_UnderscoreFlagName(t *testing.T) {
+	t.Parallel()
+	print := printFlagSchemaFor("+cells-set-style")
+	out, err := print("border_styles")
+	if err != nil {
+		t.Fatalf("underscore flag-name should resolve the hyphen schema, got: %v", err)
+	}
+	if len(out) == 0 {
+		t.Fatal("expected schema output")
+	}
+}
+
+// TestPrintFlagSchema_DottedPathSlices pins the schema sub-path slicing
+// contract on the real embedded chart schema: a dotted --flag-name returns
+// just that subtree, and a path miss lists the keys actually available.
+func TestPrintFlagSchema_DottedPathSlices(t *testing.T) {
+	t.Parallel()
+	print := printFlagSchemaFor("+chart-create")
+
+	t.Run("slices a nested subtree", func(t *testing.T) {
+		t.Parallel()
+		out, err := print("properties.snapshot.plotArea.axes")
+		if err != nil {
+			t.Fatalf("dotted path should slice, got: %v", err)
+		}
+		full, err2 := print("properties")
+		if err2 != nil {
+			t.Fatalf("full dump: %v", err2)
+		}
+		if len(out) == 0 || len(out) >= len(full) {
+			t.Errorf("slice should be non-empty and smaller than the full schema (%d vs %d bytes)", len(out), len(full))
+		}
+	})
+
+	t.Run("path miss lists available keys", func(t *testing.T) {
+		t.Parallel()
+		_, err := print("properties.snapshot.nosuchkey")
+		if err == nil {
+			t.Fatal("expected error for unknown path segment")
+		}
+		if !strings.Contains(err.Error(), "available keys:") {
+			t.Errorf("error should list available keys, got %v", err)
+		}
+	})
+}

shortcuts/vc/vc_detail.go

@@ -79,27 +79,42 @@ func fetchMeetingDetail(ctx context.Context, runtime *common.RuntimeContext, mee
 		result.NoteID = v
 	}
 
-	// Step 2: query minute_token via recording API
-	minuteToken, minuteHint, minuteErr := fetchMeetingMinuteToken(runtime, meetingID)
-	if minuteErr != nil {
-		// Recording API failed — surface the error but keep data from step 1
-		result.Error = fmt.Sprintf("failed to query minutes: %v", minuteErr)
-		minuteHint = ""
-	}
-	if minuteToken != "" {
-		result.MinuteToken = minuteToken
+	// Step 2: query minute_token via recording API — only meaningful once the
+	// meeting has ended. While it is still in progress the note/minute are not
+	// generated yet, so skip the recording call and surface an informational
+	// hint instead of letting an unclassified recording error fail the command.
+	inProgress := meetingInProgress(meeting)
+	var minuteHint string
+	if inProgress {
+		minuteHint = "meeting is still in progress; note and minute are not generated yet"
+	} else {
+		minuteToken, hint, minuteErr := fetchMeetingMinuteToken(runtime, meetingID)
+		minuteHint = hint
+		if minuteErr != nil {
+			// Recording lookup is a best-effort supplement; step 1 already
+			// succeeded, so degrade the failure to a hint rather than failing
+			// the whole command.
+			minuteHint = fmt.Sprintf("failed to query minutes: %v", minuteErr)
+		}
+		if minuteToken != "" {
+			result.MinuteToken = minuteToken
+		}
 	}
 
-	// Add hints for empty resources (not errors, just informational)
-	var emptyFields []string
-	if result.NoteID == "" {
-		emptyFields = append(emptyFields, "note_id")
-	}
-	if result.MinuteToken == "" && minuteErr == nil && minuteHint == "" {
-		emptyFields = append(emptyFields, "minute_token")
-	}
-	if len(emptyFields) > 0 {
-		result.Hint = fmt.Sprintf("%s not found for this meeting", strings.Join(emptyFields, ", "))
+	// Add hints for empty resources (not errors, just informational). For an
+	// in-progress meeting the "not found" wording is noise, so we only emit the
+	// single in-progress hint below.
+	if !inProgress {
+		var emptyFields []string
+		if result.NoteID == "" {
+			emptyFields = append(emptyFields, "note_id")
+		}
+		if result.MinuteToken == "" && minuteHint == "" {
+			emptyFields = append(emptyFields, "minute_token")
+		}
+		if len(emptyFields) > 0 {
+			result.Hint = fmt.Sprintf("%s not found for this meeting", strings.Join(emptyFields, ", "))
+		}
 	}
 	if minuteHint != "" {
 		if result.Hint != "" {
@@ -112,6 +127,36 @@ func fetchMeetingDetail(ctx context.Context, runtime *common.RuntimeContext, mee
 	return result
 }
 
+// meetingTimeField reads a meeting time field as a string regardless of whether
+// the API returned it as a JSON string or number. VC serializes int64
+// timestamps as strings, but coercing via %v keeps parsing robust either way;
+// float64(0) renders as "0", which parseFlexibleTime treats as "absent".
+func meetingTimeField(meeting map[string]any, key string) string {
+	v, ok := meeting[key]
+	if !ok || v == nil {
+		return ""
+	}
+	return strings.TrimSpace(fmt.Sprintf("%v", v))
+}
+
+// meetingInProgress reports whether a meeting is still ongoing, using the same
+// start/end heuristic as +meeting-events (meetingEventsMeetingFromPayload): a
+// meeting is ongoing when it has a start time but no end time, or its end time
+// is not after its start time. It reads the RAW timestamp fields, not the
+// FormatTime-rendered result strings, because parseFlexibleTime only accepts
+// Unix timestamps or RFC3339. Empty or "0" values are treated as absent.
+func meetingInProgress(meeting map[string]any) bool {
+	start, hasStart := parseFlexibleTime(meetingTimeField(meeting, "start_time"))
+	end, hasEnd := parseFlexibleTime(meetingTimeField(meeting, "end_time"))
+	if !hasStart {
+		return false
+	}
+	if !hasEnd {
+		return true
+	}
+	return !end.After(start)
+}
+
 // VCDetail gets meeting details including note_id and minute_token.
 var VCDetail = common.Shortcut{
 	Service:     "vc",

shortcuts/vc/vc_detail_test.go

@@ -269,11 +269,58 @@ func TestFetchMeetingDetail_RecordingAPIErrorButNoteOK(t *testing.T) {
 		if result.MinuteToken != "" {
 			t.Errorf("minute_token = %q, want empty", result.MinuteToken)
 		}
-		if !strings.Contains(result.Error, "failed to query minutes") || !strings.Contains(result.Error, "weird API error") {
-			t.Errorf("error = %q, want contains 'failed to query minutes' and 'weird API error'", result.Error)
+		if result.Error != "" {
+			t.Errorf("error = %q, want empty: a recording lookup failure must not fail the command", result.Error)
 		}
-		if strings.Contains(result.Hint, "minute_token") {
-			t.Errorf("hint = %q, should not mention minute_token when there is an error", result.Hint)
+		if !strings.Contains(result.Hint, "failed to query minutes") || !strings.Contains(result.Hint, "weird API error") {
+			t.Errorf("hint = %q, want contains 'failed to query minutes' and 'weird API error'", result.Hint)
+		}
+		return nil
+	}); err != nil {
+		t.Fatalf("unexpected error: %v", err)
+	}
+}
+
+// TestFetchMeetingDetail_MeetingInProgress pins the in-progress behavior: when a
+// meeting is still ongoing (end_time not after start_time), +detail must not
+// call the recording API at all — it returns meeting metadata with an
+// informational hint and no error. Deliberately register NO recording stub so
+// that any recording call would fail on an unmatched request.
+func TestFetchMeetingDetail_MeetingInProgress(t *testing.T) {
+	t.Setenv("LARKSUITE_CLI_CONFIG_DIR", t.TempDir())
+	f, _, _, reg := cmdutil.TestFactory(t, defaultConfig())
+	reg.Register(&httpmock.Stub{
+		Method: "GET",
+		URL:    "/open-apis/vc/v1/meetings/m_live",
+		Body: map[string]interface{}{
+			"code": 0, "msg": "ok",
+			"data": map[string]interface{}{"meeting": map[string]interface{}{
+				"id":         "m_live",
+				"topic":      "Live Meeting",
+				"meeting_no": "912052453",
+				// end_time == start_time signals an ongoing meeting.
+				"start_time": "1752000000",
+				"end_time":   "1752000000",
+			}},
+		},
+	})
+
+	if err := botExec(t, "detail-live", f, func(_ context.Context, rctx *common.RuntimeContext) error {
+		result := fetchMeetingDetail(context.Background(), rctx, "m_live")
+		if result.Topic != "Live Meeting" {
+			t.Errorf("topic = %q, want 'Live Meeting'", result.Topic)
+		}
+		if result.Error != "" {
+			t.Errorf("error = %q, want empty for an in-progress meeting", result.Error)
+		}
+		if result.MinuteToken != "" {
+			t.Errorf("minute_token = %q, want empty for an in-progress meeting", result.MinuteToken)
+		}
+		if !strings.Contains(result.Hint, "in progress") {
+			t.Errorf("hint = %q, want to mention the meeting is in progress", result.Hint)
+		}
+		if strings.Contains(result.Hint, "not found for this meeting") {
+			t.Errorf("hint = %q, should not emit not-found noise for an in-progress meeting", result.Hint)
 		}
 		return nil
 	}); err != nil {

skills/lark-approval/references/lark-approval-initiate.md

@@ -69,19 +69,17 @@ lark-cli approval approvals get \
 |---|---|---|
 | `--data '{...}'` | 是 | 请求体，使用 JSON 传入 |
 | `approval_code` | 是 | 审批定义 Code；必须先通过 `approvals search` / `approvals get` 确认 |
-| `form` | 是 | 表单值，**JSON 数组字符串**，不是普通对象 |
+| `form` | 否 | 表单值，**JSON 数组字符串**，不是普通对象；API 层非必填，但审批定义存在必填控件或用户需要提交表单值时必须传 |
 | `node_approver_list` | 否 | 节点审批人列表；仅在定义要求补充审批人时传 |
 | `node_cc_list` | 否 | 节点抄送人列表；仅在用户明确需要补充节点抄送人时传 |
 | `uuid` | 否 | 幂等标识；重复重试同一请求时建议显式传入 |
-| `--params '{...}'` | 否 | 查询参数，使用 JSON 传入 |
-| `user_id_type` | 否 | 用户 ID 类型：`user_id`、`union_id`、`open_id`；涉及人员类 ID 时建议显式传 `open_id` |
 | `--as user` | 否 | 建议显式指定用户身份；审批发起通常应使用用户身份 |
 | `--yes` | 是 | 写操作确认；真实执行时必须显式传入 |
 | `--dry-run` | 否 | 预览 API 调用，不执行 |
 
 ### 4. 组装 `form`
 
-`instances create --data.form` 是一个 JSON 数组字符串。组装原则：
+`instances create --data.form` 是可选字段；传入时必须是一个 JSON 数组字符串。无表单或无需填写表单值的审批可省略 `form`，但只要审批定义包含需要提交的控件，就必须按控件结构组装后传入。组装原则：
 
 - 先用 `approvals.get.form` 识别有哪些控件、每个控件的 `id` / `type` / 可选值范围，再按本文中的创建参数规则与 [`lark-approval-instance-form-control-parameters.md`](./lark-approval-instance-form-control-parameters.md) 重新组装创建 payload。
 - 提交时必须至少保证每个控件的 `id`、`type` 与 `value` 符合当前接口要求；不要假设定义快照里出现的其他字段都能直接照搬。
@@ -173,7 +171,6 @@ lark-cli approval instances create \
       }
     ]
   }' \
-  --params '{"user_id_type":"open_id"}' \
   --as user \
   --yes
 ```

skills/lark-approval/references/lark-approval-instances-initiated.md

@@ -14,6 +14,9 @@ lark-cli approval instances initiated --params '{"page_size":20}' --as user
 # 只看某个审批定义下我发起的实例
 lark-cli approval instances initiated --params '{"definition_code":"<DEFINITION_CODE>","page_size":20}' --as user
 
+# 按发起时间范围筛选（秒级时间戳）
+lark-cli approval instances initiated --params '{"start_timestamp":"<START_SECONDS>","end_timestamp":"<END_SECONDS>","page_size":20}' --as user
+
 # 使用 page_token 翻页
 lark-cli approval instances initiated --params '{"page_size":20,"page_token":"example_page_token"}' --as user
 
@@ -30,6 +33,8 @@ lark-cli approval instances initiated --params '{"page_size":20}' --as user --dr
 |------|------|------|
 | `--params '{...}'` | 否 | 查询参数，使用 JSON 传入；不传时使用默认分页与筛选 |
 | `definition_code` | 否 | 审批定义 Code，用于只查看某个审批定义下我发起的实例 |
+| `start_timestamp` | 否 | 按发起时间筛选，时间范围开始值，秒级时间戳 |
+| `end_timestamp` | 否 | 按发起时间筛选，时间范围结束值，秒级时间戳 |
 | `locale` | 否 | 返回语言：`zh-CN`、`en-US`、`ja-JP` |
 | `page_size` | 否 | 分页大小 |
 | `page_token` | 否 | 翻页标记；首次请求不填，后续使用上一次返回的 `page_token` |
@@ -101,6 +106,7 @@ lark-cli approval instances initiated \
 
 - **这是定位“我发起的审批实例”的首选命令**：如果你的目标是撤回、抄送、查看某个已发起审批，优先从这里拿 `instance_code`。
 - **优先用 `definition_code` 缩小范围**：当你已知审批定义时，先筛掉无关实例，可显著提升可读性。
+- **按时间排查时使用 `start_timestamp` / `end_timestamp`**：这两个值都是秒级时间戳，用于按发起时间缩小结果范围。
 - **结果很多时优先 `--format table`**：适合人工快速浏览。
 - **`count` 只在第一页返回**：做分页处理时不要假设后续页还会带总数。
 - **`instance_status` 可直接判断下一步**：例如状态为 `1` 时通常可继续查看详情或考虑撤回，状态为 `4` 表示已经撤销，无需重复撤回。

skills/lark-approval/references/lark-approval-tasks-query.md

@@ -14,6 +14,9 @@ lark-cli approval tasks query --params '{"topic":"1"}' --as user
 # 查询已办审批
 lark-cli approval tasks query --params '{"topic":"2"}' --as user
 
+# 按任务时间范围筛选（秒级时间戳）
+lark-cli approval tasks query --params '{"topic":"1","start_timestamp":"<START_SECONDS>","end_timestamp":"<END_SECONDS>"}' --as user
+
 # 使用 page_token 翻页
 lark-cli approval tasks query --params '{"topic":"1","page_token":"example_page_token"}' --as user
 
@@ -28,6 +31,8 @@ lark-cli approval tasks query --params '{"topic":"1"}' --format table --as user
 | `--params '{"topic":"..."}'` | 是 | 查询参数，使用 JSON 传入 |
 | `topic` | 是 | 任务分组主题，见下方“topic 枚举” |
 | `definition_code` | 否 | 审批定义 Code，用于仅查询某个审批定义下的任务 |
+| `start_timestamp` | 否 | 按任务时间筛选，时间范围开始值，秒级时间戳 |
+| `end_timestamp` | 否 | 按任务时间筛选，时间范围结束值，秒级时间戳 |
 | `locale` | 否 | 返回语言：`zh-CN`、`en-US`、`ja-JP` |
 | `page_size` | 否 | 分页大小 |
 | `page_token` | 否 | 翻页标记；首次请求不填，后续使用上一次返回的 `page_token` |
@@ -67,10 +72,14 @@ lark-cli approval tasks query --params '{"topic":"1"}' --format table --as user
 | `tasks[].summaries` | 表单摘要字段列表 |
 | `tasks[].support_api_operate` | 是否支持通过 API 同意或拒绝该任务 |
 | `tasks[].user_id` | 任务所属用户 ID |
+| `tasks[].instance_external_id` | 三方审批实例 ID，仅第三方审批实例存在 |
+| `tasks[].task_external_id` | 三方审批任务 ID，仅第三方审批任务存在 |
+| `tasks[].link` | 三方审批跳转链接 |
 
 ## 使用建议
 
 - 常见处理链：先用 `tasks query` 拿到 `task_id` 和 `instance_code`，若用户需要查看详情、当前节点、表单内容、流程进度等内容，则调用 `instances get` 查看详情，最后执行 `tasks approve` / `tasks reject` / `tasks transfer` / `tasks add_sign` / `tasks rollback`。
 - 如果你只想看“已发起的审批实例”，使用 `instances initiated`；`tasks query` 更适合围绕“任务分组”来拉取列表。
+- 按时间排查任务时使用 `start_timestamp` / `end_timestamp` 缩小范围；这两个值都是秒级时间戳。
 - 需要继续翻页时，直接把上一次返回的 `page_token` 放回 `--params`。
 - 当结果量较大时，优先使用 `--format table` 提升可读性。

skills/lark-approval/references/lark-approval-tasks-rollback.md

@@ -23,6 +23,12 @@ lark-cli approval tasks rollback \
   --as user \
   --yes
 
+# 退回到发起节点（发起节点 ID 为 START）
+lark-cli approval tasks rollback \
+  --data '{"instance_code":"<INSTANCE_CODE>","task_id":"<TASK_ID>","node_ids":["START"],"comment":"退回发起人补充材料"}' \
+  --as user \
+  --yes
+
 # 传多个候选节点 ID（以实际审批定义支持情况为准）
 lark-cli approval tasks rollback \
   --data '{"instance_code":"<INSTANCE_CODE>","task_id":"<TASK_ID>","node_ids":["<NODE_ID_1>","<NODE_ID_2>"],"comment":"退回上一处理节点"}' \
@@ -43,7 +49,7 @@ lark-cli approval tasks rollback \
 | `--data '{...}'` | 是 | 请求体 JSON，使用 JSON 传入 |
 | `instance_code` | 是 | 审批实例 Code；通常先通过 `tasks query` 或 `instances initiated` / `instances get` 获取 |
 | `task_id` | 是 | 审批任务 ID；通常先通过 `tasks query` 获取 |
-| `node_ids` | 是 | 退回目标节点 ID 数组；执行前应先确认这些节点确实可作为退回目标 |
+| `node_ids` | 是 | 退回目标节点 ID 数组；发起节点 ID 为 `START`；执行前应先确认这些节点确实可作为退回目标 |
 | `comment` | 否 | 审批意见或退回说明，例如 `请补充附件后重新提交`、`预算说明不完整，请补充` |
 | `--as user` | 否 | 建议显式指定用户身份；审批退回通常必须以用户身份执行 |
 | `--yes` | 否 | 确认执行高风险写操作；未带时可能返回 `confirmation_required` / exit 10 |
@@ -75,7 +81,7 @@ lark-cli approval instances get --params '{"instance_code":"<INSTANCE_CODE>"}' -
 ## 使用建议
 
 - **`instance_code` 和 `task_id` 要成对使用**：仅有实例 ID 或仅有任务 ID 都不足以准确执行退回操作。
-- **`node_ids` 是必填项**：退回并不是“自动退回上一步”，而是要明确给出目标节点 ID 数组。
+- **`node_ids` 是必填项**：退回并不是“自动退回上一步”，而是要明确给出目标节点 ID 数组；退回发起节点时传 `START`。
 - **先确认节点是否可退回**：不同审批定义支持的退回目标可能不同；在不确定时，先通过 `instances get` 或业务侧流程信息核实。
 - **优先从 `tasks query` 的待办列表拿任务参数**：尤其是 `topic=1` 的待办审批，最适合作为 rollback 的输入来源。
 - **先检查是否支持 API 操作**：如果 `tasks[].support_api_operate` 为 `false`，说明该任务可能不支持通过 API 执行处理动作，退回前应谨慎验证。

skills/lark-base/references/lark-base-cell-value.md

@@ -16,15 +16,20 @@
 
 ## 2. 各类型 CellValue
 
-### 2.1 text / phone / url
+### 2.1 text
 
-用字符串。URL 字段也传 URL 字符串；普通文本里可以保留 Markdown 风格链接文本，平台会按字段类型处理。
+text 字段的 `style.type` 影响单元格检查逻辑：
+`type=plain` 传 Markdown 格式的字符串。
+`type=url` 传一个带 title 的 Markdown 格式链接，或单独传一个链接。
+`type=phone` 传合法电话号码。
+`type=email` 传合法邮箱字符串。
 
 ```json
 {
-    "标题": "Hello",
+    "标题": "Hello, [lark-cli](https://github.com/larksuite/cli)",
+    "官网": "[官网](https://example.com)",
     "联系电话": "1380000000000",
-    "官网": "https://example.com"
+    "邮箱": "owner@example.com"
 }
 ```
 

skills/lark-base/references/lark-base-field-create.md

@@ -23,12 +23,12 @@ lark-cli base +field-create \
 lark-cli base +field-create \
   --base-token <base_token> \
   --table-id <table_id> \
-  --json '{"name":"状态","type":"select","multiple":false,"options":[{"name":"Todo","hue":"Blue","lightness":"Lighter"},{"name":"Done","hue":"Green","lightness":"Light"}]}'
+  --json '{"name":"状态","type":"select","multiple":false,"default_value":["Todo"],"options":[{"name":"Todo","hue":"Blue","lightness":"Lighter"},{"name":"Done","hue":"Green","lightness":"Light"}]}'
 
 lark-cli base +field-create \
   --base-token <base_token> \
   --table-id <table_id> \
-  --json '{"name":"负责人","type":"user","multiple":false,"description":"用于标记记录的直接负责人；协作约定可参考[团队字段约定](https://example.com/field-spec)"}'
+  --json '{"name":"负责人","type":"user","multiple":false,"default_value":[{"$slot":"current_user"}],"description":"用于标记记录的直接负责人；协作约定可参考[团队字段约定](https://example.com/field-spec)"}'
 ```
 
 ## 参数
@@ -51,6 +51,7 @@ POST /open-apis/base/v3/bases/:base_token/tables/:table_id/fields
 - `--json` 必须是 **JSON 对象**，顶层直接传字段定义，不要再套一层。
 - 顶层最少包含：`name`、`type`。
 - 所有字段类型都支持可选 `description`；支持纯文本，也支持 Markdown 链接，如 `协作约定可参考[团队字段约定](https://example.com/field-spec)`。
+- 需要字段默认值时传 `default_value`，直接使用字段对应 CellValue；`datetime` / `user` 的动态填充用 `$slot`。完整规则见 [lark-base-field-json.md](lark-base-field-json.md)。
 - `type` 不同，必填子字段不同：
   - `select`：`multiple` 控制是否多选，`options` 定义静态选项，`dynamic_options_source` 定义动态选项来源。静态与动态选项配置二选一，不能同时传。
   - `link`：必须有 `link_table`，可选 `bidirectional`、`bidirectional_link_field_name`。
@@ -64,6 +65,7 @@ POST /open-apis/base/v3/bases/:base_token/tables/:table_id/fields
   "name": "状态",
   "type": "select",
   "multiple": false,
+  "default_value": ["Todo"],
   "options": [
     { "name": "Todo", "hue": "Blue", "lightness": "Lighter" },
     { "name": "Done", "hue": "Green", "lightness": "Light" }

skills/lark-base/references/lark-base-field-json.md

@@ -9,6 +9,7 @@
 - `--json` 必须是 JSON 对象。
 - 顶层统一使用：`type` + `name` + 类型特有字段。
 - 所有字段类型都支持可选 `description`；支持纯文本，也支持 Markdown 链接。
+- 字段默认值使用 `default_value`，直接传对应 CellValue；支持范围只有 `text`、`number`、静态 `select`、`datetime`、`user`。清空默认值传 `null`；省略表示创建时不设置、更新时不修改。
 - 不要使用旧结构：`field_name`、`property`、`ui_type`、数字枚举 `type`。
 - `+field-update` 使用同样的字段 JSON 结构，但语义是 `PUT`；这是高风险写入操作，建议先 `+field-get` 再按目标状态全量提交，并带 `--yes`。
 - `type=formula` 或 `type=lookup` 创建/更新前，必须先读对应 guide。
@@ -27,12 +28,12 @@
 
 | 类型 | 最小必填字段 | 常见补充字段 |
 |------|--------------|-------------|
-| `text` | `type` `name` | `style.type` |
-| `number` | `type` `name` | `style` |
-| `select` | `type` `name` | `multiple` + `options`，或 `multiple` + `dynamic_options_source` |
-| `datetime` | `type` `name` | `style.format` |
+| `text` | `type` `name` | `style.type` `default_value` |
+| `number` | `type` `name` | `style` `default_value` |
+| `select` | `type` `name` | `multiple` + `options` + 静态 `default_value`，或 `multiple` + `dynamic_options_source` |
+| `datetime` | `type` `name` | `style.format` `default_value` |
 | `created_at` / `updated_at` | `type` `name` | `style.format` |
-| `user` / `group_chat` | `type` `name` | `multiple` |
+| `user` / `group_chat` | `type` `name` | `multiple`；仅 `user` 支持 `default_value` |
 | `created_by` / `updated_by` | `type` `name` | 无 |
 | `link` | `type` `name` `link_table` | `bidirectional` `bidirectional_link_field_name` |
 | `formula` | `type` `name` `expression` | 无 |
@@ -47,31 +48,37 @@
 ### 3.1 text
 
 文本字段；电话、超链接、邮箱、条码也都属于 `text`，通过 `style.type` 区分。
+支持 `default_value`：静态 Markdown 文本字符串；`phone` style 必须是合法电话号码；`url` style 传一个 Markdown 链接或裸 URL；`email` style 必须是合法邮箱字符串，不要传 Markdown 链接或 `mailto:`。
 
 最小写法（默认 `style.type` 为 `plain`）：
 
 ```json
 {
   "type": "text",
-  "name": "标题"
+  "name": "标题",
+  "default_value": "默认标题"
 }
 ```
 
 常用写法：
 
+默认值可以是 Markdown 文本
 ```json
 {
   "type": "text",
   "name": "标题",
-  "description": "主标题字段"
+  "description": "主标题字段",
+  "default_value": "未命名"
 }
 ```
 
+`style.type=phone` 时默认值是合法电话号码字符串。
 ```json
 {
   "type": "text",
   "name": "联系电话",
-  "style": { "type": "phone" }
+  "style": { "type": "phone" },
+  "default_value": "+8613800000000"
 }
 ```
 
@@ -79,7 +86,17 @@
 {
   "type": "text",
   "name": "官网",
-  "style": { "type": "url" }
+  "style": { "type": "url" },
+  "default_value": "[官网](https://example.com)"
+}
+```
+
+```json
+{
+  "type": "text",
+  "name": "邮箱",
+  "style": { "type": "email" },
+  "default_value": "owner@example.com"
 }
 ```
 
@@ -88,13 +105,15 @@
 ### 3.2 number
 
 数字字段；货币、进度、评分都属于 `number`，通过 `style.type` 区分。
+支持 `default_value`：静态 JSON number；所有 number style 都按这个规则写。
 
 最小写法（默认 `style.type` 为 `plain`）：
 
 ```json
 {
   "type": "number",
-  "name": "工时"
+  "name": "工时",
+  "default_value": 8
 }
 ```
 
@@ -118,7 +137,8 @@
     "precision": 2,
     "percentage": false,
     "thousands_separator": true
-  }
+  },
+  "default_value": 8
 }
 ```
 
@@ -151,7 +171,8 @@
 {
   "type": "number",
   "name": "完成度",
-  "style": { "type": "progress", "percentage": true, "color": "Blue" }
+  "style": { "type": "progress", "percentage": true, "color": "Blue" },
+  "default_value": 0.65
 }
 ```
 
@@ -180,6 +201,7 @@
 #### 静态选项
 
 支持字段：`multiple`、`options`
+支持 `default_value`：静态选项名数组；即使 `multiple=false` 也写数组，如 `["Todo"]`。
 
 默认值 / 约束：
 - `multiple` 默认 `false`
@@ -189,12 +211,14 @@
 - `options[].hue` 可用：`Red`、`Orange`、`Yellow`、`Lime`、`Green`、`Turquoise`、`Wathet`、`Blue`、`Carmine`、`Purple`、`Gray` 缺省值为 `Blue`
 - `options[].lightness` 可用：`Lighter`、`Light`、`Standard`、`Dark`、`Darker` 缺省值为 `Lighter`
 - 选项里没有 `id`，只有 `name`。
+- 支持 `default_value` 配置：填选项名数组。
 
 ```json
 {
   "type": "select",
   "name": "状态",
   "multiple": false,
+  "default_value": ["Todo"],
   "options": [
     { "name": "Todo", "hue": "Blue", "lightness": "Lighter" },
     { "name": "Done", "hue": "Green", "lightness": "Light" }
@@ -205,6 +229,7 @@
 #### 动态选项
 
 支持字段：`multiple`、`dynamic_options_source`
+动态选项不支持 `default_value`。
 
 默认值 / 约束：
 - `multiple` 默认 `false`
@@ -213,6 +238,7 @@
 - `dynamic_options_source.field_id` 填来源字段 id 或字段名
 - `dynamic_options_source` 仅创建支持；更新已有字段时不要传
 - 引用选项条件 / 级联筛选条件：这个功能在 Base 前端支持，属于 UI-only 属性，OpenAPI 里不支持，CLI 不能读取、创建或更新；不要根据接口返回缺失判断未配置
+- 动态选项不支持配置 `default_value`。
 
 ```json
 {
@@ -229,13 +255,15 @@
 ### 3.4 datetime
 
 手动填写的日期/时间字段。系统时间用 `created_at` / `updated_at`。
+支持 `default_value`：静态时间字符串，或 `{ "$slot": "record_created_time" }`。`datetime + record_created_time` 是自动填充可编辑单元格；`created_at` 是只读创建时间元信息。
 
 最小写法：
 
 ```json
 {
   "type": "datetime",
-  "name": "截止时间"
+  "name": "截止时间",
+  "default_value": "2026-03-24 10:00:00"
 }
 ```
 
@@ -251,7 +279,8 @@
 {
   "type": "datetime",
   "name": "截止时间",
-  "style": { "format": "yyyy-MM-dd HH:mm" }
+  "style": { "format": "yyyy-MM-dd HH:mm" },
+  "default_value": { "$slot": "record_created_time" }
 }
 ```
 
@@ -276,12 +305,19 @@
 ### 3.6 user / group_chat
 
 人员字段和群字段都支持 `multiple`。
+`user` 支持 `default_value`：人员 CellValue 数组，元素可用 `{ "id": "ou_xxx" }` 或 `{ "$slot": "current_user" }`；不要猜用户 ID。`group_chat` 不支持默认值。
 
 默认值 / 约束：
 - `multiple` 默认 `true`
+- `user` 字段支持 `default_value` 配置，`group_chat` 字段不支持 `default_value` 配置。
 
 ```json
-{ "type": "user", "name": "负责人", "multiple": true }
+{
+  "type": "user",
+  "name": "负责人",
+  "multiple": true,
+  "default_value": [{ "$slot": "current_user" }, { "id": "ou_xxx" }]
+}
 ```
 
 ```json
@@ -488,3 +524,4 @@ Object（对象字段）、Button（按钮字段）、Stage（流程字段）暂
 - `number` 的精度、货币、进度、评分配置都放在 `style` 下，不要写顶层 `precision`。
 - `datetime` 是手动日期字段；系统时间请改用 `created_at` / `updated_at`。
 - `formula` / `lookup` 没读 guide 前不要直接写。
+- 只有 `text`、`number`、静态 `select`、`datetime`、`user` 支持 `default_value`；清空统一传 `"default_value": null`。其他字段类型不要配置默认值。

skills/lark-base/references/lark-base-field-update.md

@@ -11,14 +11,14 @@ lark-cli base +field-update \
   --base-token <base_token> \
   --table-id <table_id> \
   --field-id <field_id> \
-  --json '{"name":"状态","type":"select","multiple":false,"options":[{"name":"Todo","hue":"Blue","lightness":"Lighter"},{"name":"Doing","hue":"Orange","lightness":"Light"},{"name":"Done","hue":"Green","lightness":"Light"}]}' \
+  --json '{"name":"状态","type":"select","multiple":false,"default_value":["Doing"],"options":[{"name":"Todo","hue":"Blue","lightness":"Lighter"},{"name":"Doing","hue":"Orange","lightness":"Light"},{"name":"Done","hue":"Green","lightness":"Light"}]}' \
   --yes
 
 lark-cli base +field-update \
   --base-token <base_token> \
   --table-id <table_id> \
   --field-id <field_id> \
-  --json '{"name":"负责人","type":"user","multiple":false,"description":"用于标记记录的直接负责人"}' \
+  --json '{"name":"负责人","type":"user","multiple":false,"default_value":null,"description":"用于标记记录的直接负责人"}' \
   --yes
 ```
 
@@ -47,6 +47,7 @@ PUT /open-apis/base/v3/bases/:base_token/tables/:table_id/fields/:field_id
 - `--json` 必须是 **JSON 对象**，顶层直接传字段定义。
 - 更新语义是 `PUT`（全量字段配置更新），不要只传零散片段；至少显式包含 `name`、`type`，并补齐该类型所需关键配置。
 - 所有字段类型都支持可选 `description`；支持纯文本，也支持 Markdown 链接。
+- 需要字段默认值时传 `default_value`，直接使用字段对应 CellValue；传 `null` 清空，省略表示不修改现有默认值。完整规则见 [lark-base-field-json.md](lark-base-field-json.md)。
 - `select` 更新时：`options` 仍按对象数组传，避免混入无效字段。
 - `link` 更新限制：
   - 不能把非 `link` 字段改成 `link`，也不能把 `link` 改成非 `link`。
@@ -59,6 +60,7 @@ PUT /open-apis/base/v3/bases/:base_token/tables/:table_id/fields/:field_id
   "name": "状态",
   "type": "select",
   "multiple": false,
+  "default_value": ["Doing"],
   "options": [
     { "name": "Todo", "hue": "Blue", "lightness": "Lighter" },
     { "name": "Doing", "hue": "Orange", "lightness": "Light" },

skills/lark-event/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: lark-event
 version: 1.0.0
-description: "Lark/Feishu real-time event listening / subscribing / consuming: stream events as NDJSON via `lark-cli event consume <EventKey>` (covers IM messages/reactions/chat changes, Task updates, VC meeting started/joined/ended, Minutes generated, Whiteboard updated, etc.). Use for Lark bots, real-time message processing, long-running subscribers, streaming webhook/push handlers. Supports `--max-events` / `--timeout` bounded runs and a stderr ready-marker contract — designed for AI agents running as subprocesses."
+description: "Lark/Feishu real-time event listening / subscribing / consuming: stream events as NDJSON via `lark-cli event consume <EventKey>` (covers IM messages/reactions/chat changes, Approval status changes, Task updates, VC meeting started/joined/ended, Minutes generated, Whiteboard updated, etc.). Use for Lark bots, real-time message processing, long-running subscribers, streaming webhook/push handlers. Supports `--max-events` / `--timeout` bounded runs and a stderr ready-marker contract — designed for AI agents running as subprocesses."
 metadata:
   requires:
     bins: ["lark-cli"]
@@ -147,6 +147,7 @@ Lark-defined semantic tags (**not** JSON Schema's standard `format`). Common val
 
 | Topic      | Reference                                                                    | Coverage |
 |------------|------------------------------------------------------------------------------|---|
+| Approval   | [`references/lark-event-approval.md`](references/lark-event-approval.md)     | Catalog of 2 Approval EventKeys (`approval.instance.status_changed_v4`, `approval.task.status_changed_v4`) + optional/multi `subscription_type` pre-registration + user-auth subscription lifecycle + flat output field reference |
 | IM         | [`references/lark-event-im.md`](references/lark-event-im.md)                 | Catalog of 12 IM EventKeys + shape notes (flat vs V2 envelope) + `im.message.receive_v1` field gotchas (`sender_id` is open_id only; `.content` is plain text except for `interactive` cards) + common jq recipes (filter by chat_type / message_type / sender); for `card.action.trigger` see also [`../lark-im/references/lark-im-card-action-reply.md`](../lark-im/references/lark-im-card-action-reply.md) |
 | Task       | [`references/lark-event-task.md`](references/lark-event-task.md)             | Catalog of 1 Task EventKey (`task.task.update_user_access_v2`) + Native V2 envelope shape + task commit types + user/bot subscription notes |
 | VC         | [`references/lark-event-vc.md`](references/lark-event-vc.md)                 | Catalog of 4 VC EventKeys (`vc.meeting.participant_meeting_started_v1`, `vc.meeting.participant_meeting_joined_v1`, `vc.meeting.participant_meeting_ended_v1`, `vc.note.generated_v1`) + field reference + source type semantics (meeting only) |

skills/lark-event/references/lark-event-approval.md

@@ -0,0 +1,170 @@
+# Approval Events
+
+> **Prerequisite:** Read [`../SKILL.md`](../SKILL.md) first for the `event consume` essentials (commands, subprocess contract, jq usage).
+
+## Key catalog (2)
+
+| EventKey | Purpose |
+|---|---|
+| `approval.instance.status_changed_v4` | An approval instance status changed |
+| `approval.task.status_changed_v4` | An approval task status changed |
+
+Both keys use a **Custom schema**. The raw Lark schema 2.0 envelope is flattened: event metadata is exposed as `type`, `event_id`, and `timestamp`, while approval business fields are exposed at the top level.
+
+Both keys carry a **PreConsume hook** that subscribes the current authorized user through the Approval subscription APIs before listening. The consumer intentionally does **not** unsubscribe on exit; the server-side Approval subscription relation remains until it is canceled outside `event consume`. These keys require `--as user`.
+
+## Listener and subscription selection
+
+At the raw CLI level, each `event consume` process accepts exactly one EventKey. `approval.instance.status_changed_v4` and `approval.task.status_changed_v4` have different output shapes, so listening to both still means two processes.
+
+For Approval only, `subscription_type` is an optional setup param used by PreConsume to register server-side Approval subscription relations before the local listener starts. It is **not** an output field, a local event filter, or a local subscription identity. The pushed event does not say which subscription relation caused delivery, and one business event can match both relations; deduplicate with `event_id` when needed.
+
+`subscription_type` may be omitted, a single value, a comma-separated list, or a JSON string array:
+
+```bash
+# Omitted: register both INVOLVED_APPROVAL and MANAGED_APPROVAL for this EventKey
+lark-cli event consume approval.instance.status_changed_v4 --as user
+
+# Single relation
+lark-cli event consume approval.instance.status_changed_v4 \
+  -p subscription_type=INVOLVED_APPROVAL \
+  --as user
+
+# Explicit multi-relation registration for one local consumer
+lark-cli event consume approval.task.status_changed_v4 \
+  -p subscription_type=INVOLVED_APPROVAL,MANAGED_APPROVAL \
+  --as user
+
+# JSON array form; quote it for the shell
+lark-cli event consume approval.task.status_changed_v4 \
+  -p 'subscription_type=["INVOLVED_APPROVAL","MANAGED_APPROVAL"]' \
+  --as user
+```
+
+| Value | Meaning |
+|---|---|
+| `INVOLVED_APPROVAL` | Receive events where the current user is the approval requester or approver |
+| `MANAGED_APPROVAL` | Receive events under approval definitions managed by the current user |
+
+User-intent inference:
+
+| User intent | EventKey(s) | `subscription_type` |
+|---|---|---|
+| Mentions approval instances, approval forms, approval order/status, or "instance status" | `approval.instance.status_changed_v4` | infer from relation words below |
+| Mentions approval tasks, approval todo items, approver operations, or "task status" | `approval.task.status_changed_v4` | infer from relation words below |
+| Says "approval status changes/events" without saying task vs instance | both EventKeys | infer from relation words below |
+| Says "my approvals", "approvals involving me", "I requested/approved", "待我审批", "我发起/我参与" | requested EventKey(s) | `INVOLVED_APPROVAL` |
+| Says "approvals I manage", "managed definitions", "definitions managed by me", "我管理的审批定义" | requested EventKey(s) | `MANAGED_APPROVAL` |
+| Explicitly asks for both involved and managed, or says "all approval subscriptions" | requested EventKey(s), or both if EventKey is also ambiguous | omit `subscription_type`, or pass both values in one `-p` |
+| Relation is ambiguous and the user wants broad coverage | requested EventKey(s), or both if EventKey is also ambiguous | omit `subscription_type` so PreConsume registers both |
+
+If the user's wording omits the relation and broad listening is acceptable, omit `subscription_type`. Ask only when registering both relations would be materially harmful.
+
+## Scopes & auth
+
+| EventKey | Scope | Auth |
+|---|---|---|
+| `approval.instance.status_changed_v4` | `approval:instance:read` | user |
+| `approval.task.status_changed_v4` | `approval:task:read` | user |
+
+## Subscription behavior
+
+Startup calls the endpoint for the selected EventKey:
+
+```text
+POST /open-apis/approval/v4/instances/subscription
+POST /open-apis/approval/v4/tasks/subscription
+```
+
+For each resolved `subscription_type`, PreConsume sends one request body:
+
+```json
+{"subscription_type":"INVOLVED_APPROVAL"}
+```
+
+If `subscription_type` is omitted, PreConsume sends two registration requests for that EventKey: one with `INVOLVED_APPROVAL`, then one with `MANAGED_APPROVAL`. If listening to both instance and task events, run two consumers; each consumer may omit `subscription_type` to register both relations for its own EventKey.
+
+Do not start two consumers for the same Approval EventKey merely to split `INVOLVED_APPROVAL` and `MANAGED_APPROVAL`. The server push and flattened output are keyed by EventKey and cannot be distinguished by subscription relation.
+
+Shutdown behavior:
+
+`event consume` does not call the Approval unsubscribe APIs when it exits. This applies to graceful exit, Ctrl+C / SIGTERM, stdin EOF, `--timeout`, and `--max-events`.
+
+To stop future delivery for a user, cancel the Approval subscription relation outside this consumer. The unsubscribe APIs are separate operations and are not called by `event consume`.
+
+## Output fields
+
+Common fields:
+
+| Field | Type | Description |
+|---|---|---|
+| `type` | string | Event type |
+| `event_id` | string | Globally unique event ID; use for deduplication |
+| `timestamp` | string (timestamp_ms) | Event delivery time in milliseconds, taken from `header.create_time` |
+
+Instance event fields:
+
+| Field | Type | Description |
+|---|---|---|
+| `approval_code` | string | Approval definition code; not a subscription dimension |
+| `instance_code` | string | Approval instance code |
+| `external_id` | string | Third-party approval instance id, when present |
+| `status` | string enum | `PENDING`, `APPROVED`, `REJECTED`, `CANCELED`, `DELETED`, `REVERTED`, `OVERTIME_CLOSE`, `OVERTIME_RECOVER` |
+| `operate_time` | string (timestamp_ms) | Status change time |
+| `start_user` | object | Instance starter user IDs, omitted when unavailable |
+| `start_user.open_id` | string (open_id) | Instance starter open_id, when present |
+| `start_user.union_id` | string (union_id) | Instance starter union_id, when present |
+| `start_user.user_id` | string (user_id) | Instance starter tenant user_id, when present |
+
+Task event fields:
+
+| Field | Type | Description |
+|---|---|---|
+| `approval_code` | string | Approval definition code; not a subscription dimension |
+| `instance_code` | string | Approval instance code |
+| `task_id` | string | Approval task id |
+| `external_id` | string | Third-party approval external id, when present |
+| `task_external_id` | string | Third-party task external id, when emitted |
+| `assigned_user` | object | Task assignee or operator user IDs, omitted for automatic flows without an operator |
+| `assigned_user.open_id` | string (open_id) | Task assignee or operator open_id, when present |
+| `assigned_user.union_id` | string (union_id) | Task assignee or operator union_id, when present |
+| `assigned_user.user_id` | string (user_id) | Task assignee or operator tenant user_id, when present |
+| `status` | string enum | `REVERTED`, `PENDING`, `APPROVED`, `REJECTED`, `TRANSFERRED`, `ROLLBACK`, `DONE`, `OVERTIME_CLOSE`, `OVERTIME_RECOVER` |
+| `operate_time` | string (timestamp_ms) | Status change time |
+
+## Examples
+
+```bash
+# Stream approval instance updates broadly; registers both involved and managed relations
+lark-cli event consume approval.instance.status_changed_v4 \
+  --as user
+
+# Stream approval instance updates only for approvals involving the current user
+lark-cli event consume approval.instance.status_changed_v4 \
+  -p subscription_type=INVOLVED_APPROVAL \
+  --as user
+
+# Stream approval task updates for definitions managed by the current user
+lark-cli event consume approval.task.status_changed_v4 \
+  -p subscription_type=MANAGED_APPROVAL \
+  --as user
+
+# Broad approval status listening:
+# run both EventKeys as separate processes; omit subscription_type so each registers both relations.
+lark-cli event consume approval.instance.status_changed_v4 \
+  --as user > approval-instance.ndjson &
+lark-cli event consume approval.task.status_changed_v4 \
+  --as user > approval-task.ndjson &
+wait
+
+# Listen to both involved and managed task subscriptions with one local consumer.
+lark-cli event consume approval.task.status_changed_v4 \
+  -p subscription_type=INVOLVED_APPROVAL,MANAGED_APPROVAL \
+  --as user > approval-task.ndjson
+
+# Project a compact approval-task record
+lark-cli event consume approval.task.status_changed_v4 \
+  -p subscription_type=INVOLVED_APPROVAL \
+  --as user \
+  --jq '{event_id, task_id, status, at: .operate_time}'
+```

skills/lark-sheets/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: lark-sheets
-version: 3.0.2
+version: 3.1.0
 description: "飞书电子表格：创建和操作电子表格。支持创建表格、管理工作表与行列结构（增删/合并/调整尺寸/隐藏/冻结）、读写单元格（值/公式/样式/批注/单元格图片）、查找替换、多操作原子批量更新，以及图表、透视表、条件格式、筛选器、迷你图、浮动图片等对象的创建与维护。当用户需要创建电子表格、管理工作表、批量读写或编辑数据、统计汇总与可视化、表格美化、公式计算（含 Excel 公式迁移）、金融/财务建模（DCF、三张表、预算、Sensitivity 等）等任务时使用。若用户是想按名称或关键词搜索云空间（云盘/云存储）里的表格文件，请改用 lark-drive 的 drive +search 先定位资源。当用户给出 doubao.com 的 /sheets/ URL/token 时，也应直接使用本 skill，不要因为域名不是飞书而回退到 WebFetch；路由依据是 URL 路径模式和 token，而不是域名。"
 metadata:
   requires:
@@ -15,13 +15,7 @@ metadata:
 
 ## 术语约定
 
-下列词在本 skill 各文档中可能交替出现，但**指同一对象**；解析用户口语时按此映射，不要当成不同概念：
-
-| 标准用语 | 同义 / 口语（均指同一对象） | 说明 |
-| --- | --- | --- |
-| 工作表（sheet） | 子表、tab、标签页 | spreadsheet 内的单张表；`sheet_id` 是其稳定标识 |
-| 电子表格（spreadsheet） | 工作簿、表格 | 顶层容器；由 `--url` 或 `--spreadsheet-token` 定位 |
-| reference_id | id | **表内对象**的稳定标识，即各对象主键 flag 接受的值（见下表）。⚠️ 与 `lark-sheets-float-image` 的 `--image-uri`（图片上传句柄）不是一回事，后者不属于 reference_id |
+同一对象的交替说法，按此映射解析用户口语：**工作表（sheet）**= 子表 / tab / 标签页（`sheet_id` 是稳定标识）；**电子表格（spreadsheet）**= 工作簿 / 表格（顶层容器，由 `--url` 或 `--spreadsheet-token` 定位）；**reference_id** = 表内对象的稳定标识，即各对象主键 flag 接受的值（与 `--image-uri` 图片上传句柄不是一回事）。
 
 每类对象用各自的主键 flag 定位（命名不统一，按此表对照，不要凭直觉拼）：
 
@@ -34,32 +28,33 @@ metadata:
 
 ## 飞书表格编辑准则（动手前必守，所有编辑类任务一律生效）
 
-下列准则横切所有飞书表格任务，**动手前先过一遍**——即使你是被索引直接路由进某个工具参考也一律生效。每条只给一句话纲要，展开与边界见括注的 reference。
+下列准则横切所有任务，**动手前先过一遍**——被索引直接路由进某个工具参考也一律生效；展开与边界见括注的 reference。
 
 1. **最小改动**：除任务要改的单元格 / 列外，原表其它单元格、行列结构、Sheet 名、合并区、格式 1:1 保持；中间结果放原数据右侧或新建空白 Sheet，**禁止删 / 改名 / 隐藏 / 移动已存在 Sheet**；改写类任务精确圈定行列，不该转的原值 1:1 保留。
 2. **真实写回 + 回读校验**：交付必须是对在线表格的真实写入，写完用 `+csv-get` / `+cells-get` / `+<对象>-list` 回读确认实际生效——**写操作返回 `ok` 只代表请求被接受、不代表结果符合预期**；写公式后查错误码、筛选 / 排序后核对前几行、删除 / 清空后确认已空。禁止只在文本里声称"已完成"。
 3. **读全再写**：批量填充 / 补齐 / 修正类任务先确认真实数据末行再写，只探前 N 行会漏写表尾（确定末行流程见 `lark-sheets-read-data`）。
-4. **公式优先于硬编码**：能用公式表达的计算（总计 / 占比 / 增长率 / 提取 / 查找）一律写公式而非静态值；**凡可由表内其它单元格推导的派生值默认就用公式，即使用户没说"联动 / 自动更新"**；写任何飞书公式前先读 `lark-sheets-formula-translation`，而且**只要公式真实写入表格，收尾默认就要继续跑 `lark-sheets-formula-verify` 的 `+formula-verify`，直到 `status='success'`**。
+4. **公式优先于硬编码**：能用公式表达的计算（总计 / 占比 / 提取 / 查找）一律写公式而非静态值——**凡可由表内其它单元格推导的派生值默认用公式，即使用户没说"联动"**；写公式前先读 `lark-sheets-formula-translation`，**公式落表后收尾必跑 `+formula-verify` 直到 `status='success'`**。
 5. **续写 / 扩展继承样式**：续写、补齐、复制区块、新增行列时禁止只读值只写值，必须连带 `cell_styles` + `border_styles` + 合并 + 行高一起继承（清单见 `lark-sheets-write-cells`，四边框最易漏）。
-6. **多步写入合并 `+batch-update`**：多个连续写入、或同一工具对多区域重复调用，合并为单次原子 `+batch-update`（语义见 `lark-sheets-batch-update`）。
+6. **多步写入分流**：美化收尾（样式 / 合并 / 行高列宽 / 冻结的任意组合）→ 一次 `+styles-put` 声明式规格交付（见 `lark-sheets-styles-put`）；**同一个写操作**打多个区域 → 用该命令自身的复数形态（`--ranges` / map 入参）；只有**跨类型的原子操作链**（如插列 → 写表头 → 回填数据）才用 `+batch-update`（high-risk-write，**调用必带 `--yes`**，fail-fast 不回滚；语义见 `lark-sheets-batch-update`）。
 7. **分组汇总用透视表**："按 X 统计 Y / 分组汇总 / 各类数量金额"用 `+pivot-{create|update|delete}`，禁止用 SUMIF / 本地脚本拼一张假透视表。
 8. **拆成可验证 checklist**：落地前把指令拆成所有"独立可验证子要点"，逐点 `assert` 全过才交付（多维排序每维一点、多目标每目标一点、范围类核起 / 末 / 边界）；只做第一个要点属违规。
 9. **全量处理前置断言条数**：翻译 / 打标 / 批量公式落地等逐条任务，先把预期条数硬编码再 `assert actual == expected`，禁止输出"已完成前 N 条，剩余继续"的半成品。
+10. **缺失值不编造**：补齐 / 扩展 / 按原表格式续填时，查不到或无法确定的值一律留空 + 备注注明（"暂未发布 / 未知 / 待核实"），禁止用推算值 / 估算值 / 凭空数据充数；原表若已示范缺失值写法（空值 + 备注），照抄该约定。宁可留空标注，不填不可靠的数。
 
-> 上述准则的实操展开——读取路径、原生工具优先级、脚本配合、易漏陷阱——见下方「执行要点」节；端到端工作流为：了解结构（`+workbook-info`）→ 读数据 → 理解语义 → 原生工具优先 → 写入 → 回读验证。
+> 端到端工作流：了解结构（`+workbook-info`）→ 读数据 → 理解语义 → 原生工具优先 → 写入 → 回读验证；实操展开见下方「执行要点」。
 
 ## 场景 → 命令速查（拿不准命令名先查这里，别按直觉拼）
 
-把高频意图映射到**真实存在**的 shortcut / flag。agent 常从 Excel / Google Sheets / 飞书 OpenAPI 误迁移命令名或 flag，先对照本表，避免一次必然失败的试错。完整 shortcut 见各工具参考。**选定命令后别急着写——先读「动手前读」列指向的 reference 再动手**：命令名对得上不代表用法对，写入 / 清除 / 透视类尤其容易漏掉 reference 里的防错、类型与样式继承规则。
+把高频意图映射到**真实存在**的 shortcut / flag（agent 常从 Excel / Google Sheets / OpenAPI 误迁移命令名）。**选定命令后先读「动手前读」列指向的 reference 再动手**——命令名对得上不代表用法对。
 
 | 你要做的事 | ✅ 正确写法 | 动手前读 | ❌ 不存在（会被 cobra 拒） |
 | --- | --- | --- | --- |
-| 读数据（纯值 / CSV） | `+csv-get`（范围用 `--range`） | `lark-sheets-read-data` | `+get-range`、`+range-get`、`+cells-read` |
+| 读数据（纯值 / CSV） | `+csv-get`（`--range` 可省略 = 读整个子表，无需先探行列；限定范围才传） | `lark-sheets-read-data` | `+get-range`、`+range-get`、`+cells-read` |
 | 读值 + 公式 / 样式 / 批注 | `+cells-get --include value,formula,style,comment,data_validation` | `lark-sheets-read-data` | `+get-cell`、`+cell-get`、`--with-styles`、`--with-merges`、`--include-merged-cells` |
-| 写纯文本值（整块 CSV 平铺；列里**没有**需字面保真的数值 / 日期标签 / 编号——点分日期 `12.10`、编号 `001` 会被 csv-put 数值化，不算纯文本） | `+csv-put`（定位用 `--start-cell`，单个左上角锚点格；也接受 `--range` 别名，区间自动取左上角） | `lark-sheets-write-cells` | 把含点分日期(`12.10`)/编号(`001`)的列裸灌 `+csv-put`——会被数值化（`12.10`→`12.1`、`001`→`1`，尾零/前导零丢失），改用 `+table-put` 声明 `dtypes:object` |
-| 写带类型的数据到**已有**表（列里有数字 / 金额 / 百分比 / 日期 / 计数等**本质是量值**的数据——不看当下要不要排序 / 求和，量值一律走这里） | `+table-put --sheets` 完整 payload `{"sheets":[{...}]}`（列名走 `columns`、二维数据走 `data`、列 pandas dtype 走 `dtypes`、列展示格式走 `formats`；来源不限 DataFrame——Counter / dict / list 同理；要同时美化加 `--styles` 一步带样式（区域底色 / 边框 / 列宽 / 行高 / 合并），不必事后再刷；payload 里不存在的 sheet 名会自动建子表，详见 write-cells） | `lark-sheets-write-cells` | 在本地把数字拼成 `"$1,234"` / `"30.5%"` 字符串再 `+csv-put`（会落成文本、丢失计算能力；常见借口见下方 ⚠️） |
+| 写纯文本值（整块 CSV 平铺；列里没有需字面保真的编号 / 点分日期） | `+csv-put`（定位用 `--start-cell` 左上角锚点格，也接受 `--range` 别名） | `lark-sheets-write-cells` | 把含点分日期(`12.10`)/编号(`001`)的列裸灌 `+csv-put`——会被数值化（`12.10`→`12.1`、`001`→`1`），改用 `+table-put` 声明 `dtypes:object` |
+| 写带类型的数据到**已有**表（列里有数字 / 金额 / 百分比 / 日期等**量值**——不看当下要不要排序求和，量值一律走这里） | `+table-put --sheets '{"sheets":[{"name":…,"columns":[…],"dtypes":{…},"formats":{…},"data":[[…]]}]}'`（不存在的 sheet 名自动建子表；同时美化加 `--styles` 一步带样式，详见 write-cells） | `lark-sheets-write-cells` | 在本地把数字拼成 `"$1,234"` / `"30.5%"` 字符串再 `+csv-put`（落成文本、丢计算能力，见下方 ⚠️） |
 | **新建**电子表格并写带类型的数据（类型保真需求同上，但目标表还不存在） | `+workbook-create --sheets`（协议与 `+table-put` 同构、一步建表 + typed 写入，无需先建空表再 `+table-put`；date / number 不丢；`--styles` 同样可在建表同一步带全套样式，详见 workbook） | `lark-sheets-workbook` | 用 `--values` 灌日期 / 数字（会落成文本、丢类型） |
-| 写公式 / 富写入（样式 · 批注 · 图片 · 富文本），或需精确矩形定位的值 | `+cells-set`（定位用 `--range`；批注 / 图片 / 富文本只能用它，公式也可；**公式落表后继续 `+formula-verify` 收尾**） | `lark-sheets-write-cells` | — |
+| 写公式 / 富写入（样式 · 批注 · 图片 · 富文本），或需精确矩形定位的值 | `+cells-set`（单区域 `--range`+`--cells`；**散布多处 / 跨表用 `--writes` 一次原子交付**，每项自带 sheet_name；公式落表后继续 `+formula-verify` 收尾） | `lark-sheets-write-cells` | — |
 | 插图：图片**绑定到某条记录**、随行走（凭证 / 证件照 / 商品图 / 头像 / 二维码 / 每行配图） | `+cells-set-image`（单格 `--range`，嵌入单元格内） | `lark-sheets-write-cells` | — |
 | 插图：**自由摆放、不绑数据**的装饰 / 标识（logo / 水印 / 封面大图 / banner） | `+float-image-create`（浮动图片，自由定位 + 尺寸 + 层级） | `lark-sheets-float-image` | — |
 | 查找 / 替换文本 | `+cells-search`（找，关键字用 `--find`）、`+cells-replace`（替换） | `lark-sheets-search-replace` | `+cells-find`、`+find`、`--query` |
@@ -68,37 +63,53 @@ metadata:
 | 复核某次（AI）编辑改了什么 / 取两个版本间的变更 | `+changeset-get --start-revision <编辑前版本>`（省略 `--end-revision` 取到最新；版本差 ≤ 20） | `lark-sheets-changeset` | — |
 | 取当前文档 revision（版本号） | `+revision-get` | `lark-sheets-workbook` | — |
 | 导出 xlsx / 单表 csv | `+workbook-export` | `lark-sheets-workbook` | — |
-| 导入本地 xlsx/xls/csv 文件为飞书电子表格 | `+workbook-import --file ./x.xlsx`（本地表格文件 → 飞书电子表格的正解；仅要导成多维表格 bitable 时才用 `drive +import --type bitable`） | `lark-sheets-workbook` | `drive +import`（导电子表格时绕了 drive 通道、还要多给 `--type`，应直接用 `+workbook-import`）、把 .xlsx 在本地读成数据再 `+workbook-create` 重灌（多此一举，应直接 `+workbook-import`）、要把文件并入某个**已有在线工作簿**（给它加子表）却用它——import 只会新建独立表，加子表应走 `+sheet-copy` / `+sheet-create` |
-| 参考某个**已有在线表**、把多个本地文件 / 数据各作为一张子表**追加**进去（不另起独立表） | 先 `+workbook-info` 拿模板子表 `sheet_id` → `+sheet-copy` 逐张复制模板子表（公式 / 合并 / 分组底色 / 列宽 / 条件格式全继承）再用 `+cells-*` 只改数据；无模板可继承时 `+sheet-create` 建空子表 + `+table-put --sheets/--styles` 写入 | `lark-sheets-workbook` | 把文件 `+workbook-import` / `+workbook-create` 另起一张**独立新表**（目标是并入已有工作簿时就跑偏了；这两条只产新表、不接受已有表定位） |
-| 清除内容 / 格式 | `+cells-clear`（范围维度用 `--scope`，取值 content / formats / all） | `lark-sheets-range-operations` | `--type` |
-| 批量清除多区域 | `+cells-batch-clear`（`--scope`） | `lark-sheets-batch-update` | `--target` |
-| 调整列宽 / 行高 | `+cols-resize` / `+rows-resize`（行、列是两个独立命令） | `lark-sheets-range-operations` | `--dimension`（无此 flag） |
+| 导入本地 xlsx/xls/csv 文件为飞书电子表格 | `+workbook-import --file ./x.xlsx`（仅要导成多维表格 bitable 时才用 `drive +import --type bitable`） | `lark-sheets-workbook` | `drive +import`（绕路）、本地读 .xlsx 再 `+workbook-create` 重灌（多此一举）、想并入**已有工作簿**却用它（import 只会另起新表，加子表走 `+sheet-copy` / `+sheet-create`） |
+| 参考某个**已有在线表**、把多份数据各作为一张子表**追加**进去 | 先 `+workbook-info` → `+sheet-copy` 复制模板子表（公式 / 合并 / 底色 / 列宽全继承）再 `+cells-*` 只改数据；无模板可继承时 `+sheet-create` + `+table-put --sheets/--styles` | `lark-sheets-workbook` | `+workbook-import` / `+workbook-create` 另起独立新表（这两条只产新表、不接受已有表定位） |
+| **已有**表美化收尾（样式 / 边框 / 合并 / 行高列宽 / 冻结的任意组合，单表或多表） | `+styles-put --styles '{"styles":[{"name":…,"cell_styles":[…],"cell_merges":[…],"row_sizes":[…],"col_sizes":[…],"freeze":{…}}]}'`（一份规格一次原子交付，词汇同 `+table-put --styles`） | `lark-sheets-styles-put` | 拼 `+batch-update` 的 `--operations` 子操作数组做美化、逐区域多次 `+cells-set-style` |
+| 清除内容 / 格式 | `+cells-clear --yes`（需确认；范围维度用 `--scope`，取值 content / formats / all） | `lark-sheets-range-operations` | `--type` |
+| 批量清除多区域 | `+cells-batch-clear --yes`（需确认；`--scope`） | `lark-sheets-batch-update` | `--target` |
+| 调整列宽 / 行高 | `+cols-resize` / `+rows-resize`（行、列是两个独立命令；连同样式一起调时并入 `+styles-put` 的 `row_sizes` / `col_sizes`） | `lark-sheets-range-operations` | `--dimension`（无此 flag） |
 | 分组汇总 / 透视 | `+pivot-create`（默认不传落点 flag → 自动新建子表，零覆盖） | `lark-sheets-pivot-table` | 用 SUMIF / 本地脚本拼一张假透视表 |
-| 画图表 / 可视化（柱 / 折线 / 饼 / 条 / 散点 / 组合…） | `+chart-create` | `lark-sheets-chart` | matplotlib / 本地画图再贴图（原生图表可交互、随数据更新） |
+| 画图表 / 可视化（柱 / 折线 / 饼 / 条 / 散点 / 组合…） | `+chart-create`（先 `+chart-create --print-example <column\|bar\|line\|pie\|combo…>` 本地拿最小可用 `--properties` 模板，改 refs / index 即可用） | `lark-sheets-chart` | matplotlib / 本地画图再贴图（原生图表可交互、随数据更新） |
 | 条件高亮 / 数据条 / 色阶 / 重复值标记 | `+cond-format-create` | `lark-sheets-conditional-format` | `+highlight`、`+conditional-format`、逐格 `+cells-set-style` 硬凑 |
 | 筛选 / 只看符合条件的行 | `+filter-create` | `lark-sheets-filter` | pandas filter 后覆盖写回（会毁原数据；要保存多份筛选状态用 `+filter-view-create`） |
 
-> ⚠️ **动手前的触发式必读（按动作判定，不看主场景）**：本次操作只要**涉及样式 / 美化**（底色 / 边框 / 字号 / 对齐 / 数字格式 / 汇总行 / 配色 / 列宽行高），动手前先读 `lark-sheets-visual-standards`；只要**要写飞书公式**，动手前先读 `lark-sheets-formula-translation`（飞书函数与 Excel 有差异，凭直觉迁移易错），**写完后再读 `lark-sheets-formula-verify` 并执行 `+formula-verify` 收尾**。哪怕主任务是"建表 / 展开数据 / 录入"，只要动作里含美化或写公式就适用——别因"这不算专门的美化 / 公式任务"而跳过。
-> ⚠️ **两种图片别选错**：图若**绑定某条记录、要随行排序 / 筛选 / 增删**（凭证 / 证件照 / 每行配图，话里带「对应 / 每行 / 这列」等绑定词）→ 单元格图片 `+cells-set-image`；只是自由摆放的装饰（logo / 水印 / 封面）→ 浮动图片 `+float-image-create`。别因「浮动图更好控制 / 更熟」默认选浮动图。
-> ⚠️ **纯文本还是数值语义（看数据本质，不看当下用途）**：金额 / 百分比 / 比率 / 计数 / 日期等**本质是量值**的数据 → 一律数值写入，常规二维表用 `+table-put`（`dtypes` 声明类型 + `formats` 设展示格式），版式装不下（多级 / 合并表头的宽表 leaderboard 等）改用 `+cells-set` 传数字（百分比传小数 `0.4`）+ `number_format`，照样显示 `40%` 且数值无损。只有编号 / 身份证 / 单据号这类**本质是标识符**、要字面保真的才用 `+csv-put` 平铺。**几个常见借口都不成立**——"只是 leaderboard / 报表展示不用算""版式复杂""样式以后再刷、先铺文本"都不是把百分比写成 `"40%"` 字符串灌 `+csv-put` 的理由（展示不改变它是数值；类型不能后补，落成文本就回不来）。判据与操作展开见 `lark-sheets-write-cells`「数字还是文本」。
-> ⚠️ **要新建子表 / 整表美化 → 别默认「`+csv-put` 写值再事后刷样式」**：`+table-put` / `+workbook-create` 的 `--styles` 能在写数据的**同一步**带全套样式（区域底色 / 边框 / 列宽 / 行高 / 合并），且 `+table-put` 的 payload 里若 sheet 名不在工作簿中会自动新建子表——**纯文本表要新建子表 + 美化时同样走这里**（`--styles` 与列是否 typed 无关），比「`+csv-put` 写值 + 多次 `+cells-batch-set-style` / `+*-resize` 刷样式」少好几次调用（冻结行列等 sheet 级属性仍需 `+dim-freeze` 单独一步）。
-> ⚠️ **定位 flag**：`+cells-get` / `+cells-set` / `+csv-get` 用 `--range`；`+csv-put` 规范用 `--start-cell`（单个左上角锚点格），也接受 `--range` 别名（区间自动取左上角），二者择一即可。
-> ⚠️ **读取附加信息**一律走 `+cells-get --include …`，**没有** `--with-styles` 这类 flag；**看合并单元格**用 `+sheet-info` 的 `merged_cells`，不要在 `+cells-get` 里找 merge flag。
+> ⚠️ **动手前的触发式必读（按动作判定，不看主场景）**：动作里**含样式 / 美化**（底色 / 边框 / 字号 / 对齐 / 数字格式 / 配色 / 列宽行高）→ 先读 `lark-sheets-visual-standards`；**要写飞书公式** → 先读 `lark-sheets-formula-translation`，写完跑 `+formula-verify` 收尾（见 `lark-sheets-formula-verify`）。主任务是建表 / 录入也一样适用。
+> ⚠️ **两种图片别选错**：图**绑定某条记录、随行走**（凭证 / 证件照 / 每行配图）→ `+cells-set-image`；自由摆放的装饰（logo / 水印 / 封面）→ `+float-image-create`。别因「浮动图更熟」默认选浮动图。
+> ⚠️ **纯文本还是数值语义（看数据本质，不看当下用途）**：金额 / 百分比 / 日期 / 计数等**量值**一律数值写入——常规二维表用 `+table-put`（`dtypes` + `formats`），宽表 / 合并表头版式用 `+cells-set` 传数字（百分比传小数 `0.4`）+ `number_format`。只有编号 / 身份证等**标识符**才 `+csv-put` 平铺。"只是展示不用算 / 样式以后再刷"不构成把量值写成字符串的理由——类型不能后补。判据见 `lark-sheets-write-cells`「数字还是文本」。
+> ⚠️ **要新建子表 / 整表美化 → 别「`+csv-put` 写值再事后刷样式」**：`+table-put` / `+workbook-create` 的 `--styles` 在写数据**同一步**带全套样式（底色 / 边框 / 列宽行高 / 合并 / 冻结），payload 里不存在的 sheet 名自动建子表，纯文本表同样适用；比事后多次刷样式少好几次调用。存量表事后美化则一次 `+styles-put` 交付（同一份 `--styles` 词汇）。
+> ⚠️ **定位 flag**：`+cells-get` / `+cells-set` / `+csv-get` 用 `--range`；`+csv-put` 用 `--start-cell`（也接受 `--range` 别名，区间取左上角）。
+> ⚠️ **读取附加信息**一律走 `+cells-get --include …`（无 `--with-styles` 这类 flag）；**看合并单元格**用 `+sheet-info` 的 `merged_cells`。
+
+💡 **高频写命令签名（照抄改参即可；各命令 `--help` 的 Tips 段有同款示例）**：
+
+```bash
+lark-cli sheets +cells-set --url <U> --sheet-name S1 --range A1:B1 --cells '[[{"value":"名称"},{"formula":"=SUM(B2:B9)"}]]'  # --cells 恒为二维数组 [[…]]，单格也是 [[{…}]]
+lark-cli sheets +cells-set-style --url <U> --sheet-name S1 --range A1:D1 --font-weight bold --background-color "#F0F0F0" --horizontal-alignment center
+lark-cli sheets +styles-put --url <U> --styles - <<'JSON'
+{"styles":[{"name":"S1","cell_styles":[{"range":"A1:D1","font_weight":"bold","background_color":"#F0F0F0"}],"col_sizes":[{"range":"A:D","type":"pixel","size":120}],"freeze":{"rows":1}}]}
+JSON
+lark-cli sheets +batch-update --url <U> --yes --operations - <<'JSON'
+[{"shortcut":"+cells-set","input":{"sheet_name":"S1","range":"A1","cells":[[{"value":"x"}]]}}]
+JSON
+lark-cli sheets +dim-freeze --url <U> --sheet-name S1 --dimension row --count 2
+lark-cli sheets +dim-insert --url <U> --sheet-name S1 --position 3 --count 2 --inherit-style before  # 行/列由 --position 决定：数字=行、字母=列，无 --dimension
+lark-cli sheets +cols-resize --url <U> --sheet-name S1 --range A:C --width 120  # 像素；分列不同宽用 --widths '{"A":80,"C:E":120}'
+lark-cli sheets +sheet-copy --url <U> --sheet-name 源表名 --title 副本名  # --sheet-name=源表、--title=新表名
+```
 
 ## 执行要点（读取 / 原生工具 / 陷阱）
 
-准则的实操展开。端到端工作流：了解结构 → 读数据 → 理解语义 → 原生工具优先 → 写入 → 回读验证。
-
 ### 读取：按需求选路径（细则见 `lark-sheets-read-data`）
 
 | 用户需求 | 读取路径 |
 |---|---|
-| "完善 / 补齐 / 填空 / 修正所有 XX"、分析 / 清洗 / 大数据 | 原生优先（公式 / `+pivot` / `+filter`）；表达不了再分批 `+csv-get` 导出 + 脚本处理 + 分批回写（默认覆盖所有对应数据行，不以用户选区为准） |
-| "查一下 / 看看 / 统计 / 汇总"等只读 | `+csv-get` 读到上下文 |
+| "完善 / 补齐 / 修正所有 XX"、分析 / 清洗 / 大数据 | 原生优先（公式 / `+pivot` / `+filter`）；表达不了再分批 `+csv-get` 导出 + 脚本处理 + 分批回写（默认覆盖所有对应数据行） |
+| "查一下 / 统计 / 汇总"等只读 | `+csv-get` 读到上下文 |
 | 需要公式 / 样式 / 批注 | `+cells-get` |
 | 续写 / 扩展已有内容 | `+csv-get` 看结构 + `+cells-get` 读源区样式 + `+sheet-info --include row_heights,merges`（见准则 5） |
 
-> "补齐 / 填空"类用只读路径探 10 行就写会漏写表尾——写入前先按 `lark-sheets-read-data` 确认真实数据末行（准则 3）。
+> "补齐 / 填空"类只探前 10 行就写会漏写表尾——先按 `lark-sheets-read-data` 确认真实数据末行（准则 3）。
 
 ### 计算：原生工具优先，代码兜底（强化准则 7）
 
@@ -122,16 +133,16 @@ metadata:
 
 ### 易漏陷阱
 
-- **`+dim-insert` 不继承行高**：只继承值 / 公式 / 边框，新行回落默认高度截断长文本；插行填长文本前读相邻行 `row_height`，用 `+batch-update` 合 `+rows-resize` 补齐。
-- **公式容错**：日期 / 查找 / 数值转换公式用 `IFERROR` 包裹；写完读结果列首末各 5 行查 `#VALUE!` / `#REF!` / `#DIV/0!`，然后继续跑 `+formula-verify` 直到 `status='success'`；同一方案试错上限 3 次。
+- **`+dim-insert` 不继承行高**：只继承值 / 公式 / 边框；插行填长文本前读相邻行 `row_height`，用 `+batch-update` 合 `+rows-resize` 补齐。
+- **公式容错**：日期 / 查找 / 转换公式用 `IFERROR` 包裹；写完查首末各 5 行错误码，再跑 `+formula-verify` 到 `status='success'`；同一方案试错上限 3 次。
 - **循环引用**：聚合公式引用范围不能含目标 cell 自身或其传递依赖。
-- **隐藏行列**：`+csv-get` 默认含隐藏行列；设 `--skip-hidden=true` 只看可见，但返回行序号与实际行号不再对应。
-- **跨 sheet 对象**：图表 / 条件格式 / 透视表 / 浮动图片可能分布在多个子表，操作前先 `+workbook-info` 掌握全局。
-- **NLP 任务分批**：语义理解 / 翻译 / 改写 / 分类等用 NLP 处理（代码只做分批 / 行号映射 / 写回）；数据量大必须分批（通常 30 行 / 批），每批处理完即时写回，单批生成通常 ≤ 300 行，多批用 `+batch-update`。
+- **隐藏行列**：`+csv-get` 默认含隐藏行列；`--skip-hidden=true` 只看可见，但返回行序号与实际行号不再对应。
+- **跨 sheet 对象**：图表 / 条件格式 / 透视表 / 浮动图片可能分布在多个子表，先 `+workbook-info` 掌握全局。
+- **NLP 任务分批**：语义理解 / 翻译 / 打标用 NLP 处理（代码只做分批 / 行号映射 / 写回）；大数据量分批（约 30 行 / 批）即时写回，多批用 `+batch-update`。
 
 ## References
 
-本 skill 的 reference 分两组：先读**通用方法与规范**（横切所有任务的样式、公式规则，不含具体 shortcut），它们规定了"怎么做对"；再按操作对象进入**工具参考**查具体 shortcut 与调用细节。编辑类任务务必先过一遍通用方法与规范，连同上方「飞书表格编辑准则」对所有工具参考一律生效。
+reference 分两组：先读**通用方法与规范**（横切所有任务的样式 / 公式规则），再按操作对象进入**工具参考**查具体 shortcut。编辑类任务务必先过通用方法与规范，连同上方「飞书表格编辑准则」对所有工具参考一律生效。
 
 ### 通用方法与规范（先读，横切所有任务，不含具体 shortcut）
 
@@ -151,6 +162,7 @@ metadata:
 | [Lark Sheet Search & Replace](references/lark-sheets-search-replace.md) | 在飞书表格中搜索和替换文本，支持限定范围、大小写匹配、精确匹配、正则表达式。当用户需要"查找"、"搜索"、"定位"某个值，或"替换"、"批量修改文本"、"把 A 改成 B"时使用。不要用于理解表格结构（应读取数据）、不要用于数据分析（应读取数据后计算）、不要把用户操作动作中的关键词（如"汇总金额""统计数量"）当作搜索词。 |
 | [Lark Sheet Write Cells](references/lark-sheets-write-cells.md) | 向飞书表格的指定区域批量写入值、公式、样式、批注或单元格图片。适用场景：填写数据、设置公式、修改格式、添加批注、嵌入单元格图片（如需操作浮动图片，请使用 lark-sheets-float-image）；若只需把一块 CSV 批量铺到表格上（值或公式，不带样式/批注），直接使用 `+csv-put` 更短更快。追加数据需先通过 lark-sheets-sheet-structure 插入行列。只要这次写入真实落了公式，收尾默认继续执行 `lark-sheets-formula-verify`。 |
 | [Lark Sheet Range Operations](references/lark-sheets-range-operations.md) | 对飞书表格中指定区域执行结构性操作（不涉及写入单元格数据值）。适用场景：清除内容或格式（"清空"、"删除内容"、"去掉格式"）、合并/取消合并单元格、调整行高列宽（"加宽列"、"自适应列宽"）、移动/复制/填充/排序数据（"移动数据"、"复制到"、"自动填充"、"按某列排序"）。写入单元格数据请使用 lark-sheets-write-cells。 |
+| [Lark Sheet Styles Put](references/lark-sheets-styles-put.md) | 把一份声明式视觉规格（样式/边框/合并/行高列宽/冻结）一次性应用到已有飞书表格的多个子表，整份规格原子提交。当任务是对存量表做美化收尾、批量刷样式、统一版式时使用。样式取值标准见 lark-sheets-visual-standards；建新表带样式走 lark-sheets-workbook（+workbook-create --styles）、写数据同步带样式走 lark-sheets-write-cells（+table-put --styles），三者共用同一份 --styles 词汇。仅针对飞书表格。 |
 | [Lark Sheet Batch Update](references/lark-sheets-batch-update.md) | 将多个飞书表格写入操作合并为一次批量执行，按顺序依次完成。适合需要连续执行多个写入操作的场景（如先修改结构再写入数据）。 |
 | [Lark Sheet Chart](references/lark-sheets-chart.md) | 管理飞书表格中的图表（柱形图、折线图、饼图、条形图、面积图、散点图、组合图、雷达图等）。当用户需要创建图表、修改图表样式或数据源、查看已有图表配置、删除图表时使用。也适用于用户提到"数据可视化"、"画个图"、"趋势分析"、"对比图"、"占比分析"、"做个图表"等数据可视化相关场景。 |
 | [Lark Sheet Pivot Table](references/lark-sheets-pivot-table.md) | 管理飞书表格中的数据透视表。当用户需要创建透视表、修改透视表的行列字段/聚合方式/筛选条件、查看已有透视表配置、删除透视表时使用。也适用于用户提到"分组汇总"、"交叉分析"、"按XXX统计"、"按字段分组"、"再分下组"、"多维分析"、"数据透视"等场景。 |
@@ -164,42 +176,22 @@ metadata:
 
 ## 公共 flag 速查
 
-各 reference 的每个 shortcut 标题下用一行徽章标注该 shortcut 支持的公共 / 系统 flag，例如：
-
-- `_公共四件套 · 系统：--dry-run_` — URL/token + sheet 定位（两组各**必给一个**，详见下方「公共 flag」），加 `--dry-run`
-- `_公共：URL/token（无 sheet 定位） · 系统：--yes、--dry-run_` — 只接 URL/token，常见于 `+batch-update` 等不强制 sheet 定位的 shortcut
-
-徽章里只列名字。type / 必填 / 描述都在本段统一声明：
+各 reference 的 shortcut 标题下用一行徽章标注支持的公共 / 系统 flag（如 `_公共四件套 · 系统：--dry-run_`；`_公共：URL/token（无 sheet 定位）…_` 表示只接 URL/token）。type / 必填 / 描述在本段统一声明：
 
 ### 公共 flag（定位资源）
 
 **公共四件套** = `--url` / `--spreadsheet-token` / `--sheet-id` / `--sheet-name`，分成两组 XOR，**每组都必须给且只能给一个**（XOR = 二选一必填，不是"可选"）：
 
-1. **spreadsheet 定位（必填）**：`--url` 与 `--spreadsheet-token` 二选一，**必须给其中之一**。两个都不给 → 校验报错 `specify at least one of --url or --spreadsheet-token`；两个都给 → 互斥冲突。
-   - **`--url` 解析 `/sheets/`、`/spreadsheets/` 与 `/wiki/` 三种链接**（从路径里抽出 token；也可以直接把裸 token 传给 `--spreadsheet-token`）。其它形态的链接不会被解析成表格 token。
-   - **`/wiki/` 知识库链接可直接传 `--url`**：会自动定位到链接背后的电子表格；若该链接背后不是电子表格（而是文档 / 多维表格等），则报错。
-   - **例外**：`+workbook-create`（新建表 + 可选写入数据）与 `+workbook-import`（把本地文件导入为新表）都产出一张**还不存在**的表格，**不接受任何 spreadsheet / sheet 定位 flag**——`+workbook-create` 只有 `--title` / `--folder-token` / `--values` / `--styles` / `--sheets`，`+workbook-import` 只有 `--file`（必填）/ `--folder-token` / `--name`。
-2. **sheet 定位（公共四件套 shortcut 必填）**：`--sheet-id` 与 `--sheet-name` 二选一，**必须给其中之一**。两个都不给 → 校验报错 `specify at least one of --sheet-id or --sheet-name`。
-   - ⚠️ **不确定 sheet 名时禁止直接猜 `Sheet1`**：除非用户对话明确说出 sheet 名 / id，或上下文（之前的工具调用 / URL 锚点 `?sheet=xxx`）已经出现过具体值，否则**第一步先调 `+workbook-info --url "..."`**（或 `--spreadsheet-token`）拿 `sheets[].sheet_id` / `sheets[].title` 列表再选。中文环境下子表常叫"数据" / "Sheet"（无数字）/ "工作表 1" / 业务名，猜 `Sheet1` 大概率撞 `sheet not found`，比先查多耗一次失败调用 + 重试。
-   - ⚠️ **`--range` 里的 `Sheet1!` 前缀不能替代 sheet 定位**：即使写了 `--range 'Sheet1!A1:B2'`，仍**必须**额外传 `--sheet-id` 或 `--sheet-name`，否则照样报上面的错。
-   - ⚠️ **A1 reference 含 `!`**（`--source` / `--range` / `--ranges`）**：整段用单引号包裹**，如 `--range 'Sheet1!A1:B2'`——单引号能挡住 bash 的 history expansion（`!` 被拦成 `event not found`；双引号挡不住；别改用 `set +H`，原因见下方「复合 JSON / 大入参」）。sheet 名含特殊字符（`-` / 空格 / 非 ASCII）需在内部按 A1 标准再包一层单引号时，用 `'\''` 转义保持外层单引号，如 `--source ''\''Sales-2025'\''!A1:D100'`。
-   - **例外**：徽章标为 `_公共：URL/token（无 sheet 定位）…_` 的 shortcut（如 `+workbook-info` / `+workbook-export` / `+batch-update` / `+dropdown-update|delete` / `+cells-batch-set-style` / `+cells-batch-clear` / `+sheet-create`）**不接受也不需要** sheet 定位，只给一组 spreadsheet 定位即可。`+pivot-create` 用 `--target-sheet-id` / `--target-sheet-name`（XOR，可都不传，落点细节见 `lark-sheets-pivot-table`）。
-
-| Flag | Type | 必填 | 说明 |
-| --- | --- | --- | --- |
-| `--url` | string | 二选一必填（与 `--spreadsheet-token`） | spreadsheet 或 wiki URL |
-| `--spreadsheet-token` | string | 二选一必填（与 `--url`） | spreadsheet token |
-| `--sheet-id` | string | 二选一必填（与 `--sheet-name`；仅公共四件套 shortcut） | 工作表 reference_id |
-| `--sheet-name` | string | 二选一必填（与 `--sheet-id`；仅公共四件套 shortcut） | 工作表名称 |
-
-**统一调用范式**（公共四件套 shortcut 的所有示例都遵循此形状，两组定位缺一不可）：
+1. **spreadsheet 定位（必填）**：`--url`（解析 `/sheets/`、`/spreadsheets/`、`/wiki/` 三种链接；wiki 链接自动定位背后的电子表格）与 `--spreadsheet-token`（裸 token）二选一。**例外**：`+workbook-create` / `+workbook-import` 产出**还不存在**的表，不接受任何定位 flag。
+2. **sheet 定位（公共四件套 shortcut 必填）**：`--sheet-id` 与 `--sheet-name` 二选一。
+   - ⚠️ **不确定 sheet 名时禁止猜 `Sheet1`**：除非对话或上下文已出现具体值，第一步先 `+workbook-info` 拿 `sheets[].sheet_id/title` 再选——中文表的子表常叫"数据"/"工作表 1"/业务名，猜名大概率撞 `sheet not found`。
+   - ⚠️ **`--range` 里的 `Sheet1!` 前缀不能替代 sheet 定位**：仍必须传 `--sheet-id` / `--sheet-name`。
+   - ⚠️ **A1 引用含 `!` 时整段用单引号包裹**（`--range 'Sheet1!A1:B2'`，挡 bash history expansion；别用 `set +H`，sh/dash 下非法）。sheet 名含 `-`/空格需内层再包单引号时用 `'\''` 转义：`--source ''\''Sales-2025'\''!A1:D100'`。
+   - **例外**：徽章标 `_公共：URL/token（无 sheet 定位）…_` 的 shortcut（`+workbook-info` / `+workbook-export` / `+batch-update` / `+styles-put` / `+dropdown-update|delete` / `+cells-batch-clear` / `+sheet-create`）不接受 sheet 定位。`+pivot-create` 用 `--target-sheet-id/name`（XOR，可都不传）。
 
 ```bash
-lark-cli sheets <shortcut> <workbook 定位> <sheet 定位> <其它 flag>
-#   workbook 定位：--url "..."        或 --spreadsheet-token "..."           （二选一，必给）
-#   sheet 定位：    --sheet-id "$SID"  或 --sheet-name "<真实表名>"            （二选一，必给；占位符不要原样填）
-# 例：lark-cli sheets +csv-get --url "https://.../sheets/shtXXX" --sheet-name "<真实表名>" --range "A1:F30"
-# 注意：真实表名不要直接填 "Sheet1"——大多数表的子表不叫这个；先 +workbook-info 拿 sheets[].title 再代入。
+# 统一调用范式：两组定位缺一不可（占位符别原样填；表名先 +workbook-info 查）
+lark-cli sheets +csv-get --url "https://.../sheets/shtXXX" --sheet-name "<真实表名>" --range "A1:F30"
 ```
 
 ### 系统 flag
@@ -208,27 +200,27 @@ lark-cli sheets <shortcut> <workbook 定位> <sheet 定位> <其它 flag>
 | --- | --- | --- | --- |
 | `--dry-run` | bool | 否 | 零副作用：仅打印请求路径与参数模板，不发起调用；多步操作会输出每个子操作的请求模板 |
 | `--yes` | bool | 是（仅 `high-risk-write`） | 二次确认；不带时退出码 10。详见 [`../lark-shared/SKILL.md`](../lark-shared/SKILL.md) 高风险审批协议 |
-| `--print-schema` | bool | 否 | 本地打印复合 JSON flag 的 JSON Schema 并退出，不发起任何调用、不需要其它 required flag。与 `--flag-name <name>` 搭配指定要查哪个 flag；省略 `--flag-name` 时列出该 shortcut 所有可查询的 flag。**仅在 shortcut 含复合 JSON flag 时有效**——判断方法：该 shortcut 的 Flags 表里出现类型标注为「复合 JSON」的 flag（如 `--cells` / `--properties` / `--operations` / `--border-styles` / `--sort-keys` / `--options`）即支持；纯标量 flag 的 shortcut 不支持。 |
-| `--flag-name` | string | 否 | 配合 `--print-schema` 使用，指定要打印 JSON Schema 的 flag 名（不带 `--` 前缀，如 `cells` / `properties` / `operations`）。 |
+| `--print-schema` | bool | 否 | 本地打印复合 JSON flag 的 JSON Schema 并退出，不发起调用、不需要其它 required flag。搭配 `--flag-name` 指定查哪个 flag；省略时列出该 shortcut 可查询的 flag。仅对含复合 JSON flag 的 shortcut 有效。 |
+| `--flag-name` | string | 否 | 配合 `--print-schema`：flag 名不带 `--` 前缀（`cells` / `properties`）。**支持点分路径切片**：`--flag-name properties.snapshot.plotArea.axes` 只打印该子树，大 schema（chart 的 properties 约 1700 行）按需取，别整篇翻页。 |
 
-**Agent 使用提示**：写复合 JSON flag（`--cells` / `--properties` / `--operations` / `--border-styles` / `--sort-keys` / `--options` 等）时，如果对结构不确定，先跑 `lark-cli sheets <shortcut> --print-schema --flag-name <name>` 把完整 JSON Schema 读出来再构造 payload，比靠 reference 的速查表更精确，也避免因为字段拼写或缺失被服务端拒绝。reference 的 `## Schemas` 段只给一层结构，深层只能靠 `--print-schema` 或 `## Examples` 的真实示例。
+> ⚠️ **high-risk-write 命令清单（首次调用就带 `--yes`，别等 exit 10 再补；或先 `--dry-run` 预览）**：`+batch-update`、`+cells-clear`、`+cells-batch-clear`、`+sheet-delete`、`+dim-delete`、`+dropdown-delete`，以及各对象删除 `+chart-delete` / `+pivot-delete` / `+cond-format-delete` / `+filter-delete` / `+filter-view-delete` / `+sparkline-delete` / `+float-image-delete`。
+
+**Agent 使用提示**：写复合 JSON flag 前对结构不确定时，先 `--print-schema --flag-name <name>`（深层字段用点分路径切片）再构造 payload；图表直接 `+chart-create --print-example <type>` 拿最小可用模板改参。reference 的 `## Schemas` 段只给一层结构。
 
 ### flag 内容类型与输出约定（术语速记）
 
-- flag 表里 JSON 类入参标三类：**复合 JSON** = 深层嵌套对象（用 `--print-schema` 取完整结构）；**简单 JSON** = 一维 / 二维标量数组（如 `["sheet1!A1:B2",...]` / `[["alice",95]]`，结构简单无需 print-schema）；**非 JSON 文本** = 原样文本（如 CSV）。`--print-schema` 只对**复合 JSON** flag 有效（同一 shortcut 的简单 JSON flag 如 `--colors` 不在此列）。
-- **envelope**：所有 shortcut 返回统一外层结构 `{ok, identity, data, ...}`。正文里 `envelope.data` 指业务数据层（如 `+csv-get` 的 `annotated_csv`）；写操作不会自动回读，如需校验请自行调用对应的 `+*-list` / `+*-get` / `+cells-get`。
+- JSON 类入参分三类：**复合 JSON** = 深层嵌套对象（`--print-schema` 可查）；**简单 JSON** = 一二维标量数组；**非 JSON 文本** = 原样文本（如 CSV）。`--print-schema` 只对复合 JSON flag 有效。
+- **envelope**：所有 shortcut 返回统一外层 `{ok, identity, data, ...}`；写操作不会自动回读，校验自行调用 `+*-list` / `+*-get` / `+cells-get`。
 
 ## 复合 JSON / 大入参：优先 stdin
 
-flag 帮助里标注支持 **Stdin** 的入参，当 payload 较大、含换行 / 引号等特殊字符，或已经落在某个文件里时，优先用 stdin（`-`）传入，避免命令行超长与 shell 转义问题。
-
-推荐写法：payload 写到用户项目目录之外的临时文件（放系统临时目录，避免污染项目），再用 stdin 喂进去：
+大 payload（`--operations` / `--cells` / `--sheets` / `--styles` / `--properties`…）、或含换行 / 引号 / `!` 等特殊字符时，优先 heredoc stdin（`-`）传入，避免命令行超长与 shell 转义问题：
 
 ```bash
-# TMPFILE 指向系统临时目录下的 payload 文件（脚本里用 tempfile.gettempdir() / os.tmpdir() 等取临时目录）
-lark-cli sheets +cells-set --url "..." --sheet-name "Sheet1" --range "A1:B2" --cells - < "$TMPFILE"
+lark-cli sheets +batch-update --url "..." --yes --operations - <<'JSON'
+[{"shortcut":"+cells-set","input":{...}}]
+JSON
 ```
 
-**参数含特殊字符（`!` / 引号 / 空格 / 非 ASCII）时，用单引号包裹该参数即可，不要起手 `set +H` 之类的 shell 开关来防转义。** `set +H`（关 bash history expansion）在 `sh` / `dash` 下是非法选项（`set: Illegal option -H`）、会让整条命令直接失败；而单引号挡得住 `!` 的 history expansion（否则报 `event not found`），对 bash 与 `sh` / `dash` 一致安全。参数本身含单引号、或 payload 较大时，按上文走 stdin。
-
-**`@file` 接绝对路径会被拒，且被拒后不要照报错提示做。** `@file` 出于安全只接受 cwd 下的相对路径，传 cwd 之外的绝对路径会被拒。此时报错会建议"先 cd 到目标目录，或改用相对路径"——**两条都不要照做**：cd 过去、或把临时文件写进用户项目目录，都会污染工作目录。正解是改用 stdin（`--<flag> - < 文件`）。
+- **stdin 每次调用只能给一个 flag**：`+table-put` 同时传 `--sheets` 与 `--styles` 两个大 JSON 时，一个走 `-`、另一个走 `@./styles.json`（`@file` 只接受 cwd 下相对路径，**绝对路径会被拒**；正解是 stdin，别 cd、别把临时文件写进用户项目目录）。
+- **参数含特殊字符时用单引号包裹即可，不要 `set +H`**（sh/dash 下非法直接报错）；参数本身含单引号或 payload 大时走 stdin。

skills/lark-sheets/references/lark-sheets-batch-update.md

@@ -8,22 +8,19 @@
 2. **批次完成后必须回读校验**：整个 `+batch-update` 执行成功后，用 `+csv-get` 或 `+cells-get` 抽样回读受影响区域，至少校验 3-5 个代表性单元格（首 / 中 / 末），与本地脚本预先计算的预期值对照。
 3. **预期条数前置断言**：涉及"批量填充 N 行"或"对 M 个区域分别写入"时，先把 N、M 硬编码进代码，回读后断言实际等于预期；不一致就再发一轮 `+batch-update` 补齐，禁止交付半成品。
 
-若本次 `+batch-update` 的任一子操作写入了公式、复制了公式模板、或导入了含公式的数据块，**回读校验之后还必须继续执行 `+formula-verify`**。`+batch-update` 的原子提交只保证“写入动作都执行了”，不保证整批公式运行结果 zero-error。
+若本次 `+batch-update` 的任一子操作写入了公式、复制了公式模板、或导入了含公式的数据块，**回读校验之后还必须继续执行 `+formula-verify`**。`+batch-update` 只保证"写入动作按序执行了"，不保证整批公式运行结果 zero-error。
 
 ## 使用场景
 
-写入。批量执行多个写入工具操作。将多个工具调用合并为一次请求，按顺序依次执行。适合需要连续执行多个写入操作的场景（如先修改结构再写入数据）。注意：不支持嵌套 `+batch-update`。
+写入。把**跨类型、有顺序依赖**的多个写入操作合并为一次请求按序执行（如插列 → 写表头 → 回填数据）。注意：不支持嵌套 `+batch-update`。
 
-**不可放进 `--operations` 的写 shortcut**（`shortcut` 枚举不含它们，强行写入会被校验拒）：`+cells-set-image`（需本地上传图片）、`+dropdown-update` / `+dropdown-delete` / `+cells-batch-set-style` / `+cells-batch-clear`（自身已是批量入口，不可再嵌套）、`+dim-move`。这些操作需在 `+batch-update` 之外单独调用。
+**先分流再动手（按操作组合选入口）**：美化收尾（样式 / 合并 / 行高列宽 / 冻结的任意组合）→ 一次 `+styles-put`（声明式规格，见 `lark-sheets-styles-put`），不要拼 `--operations` 子操作数组；**同一个写操作**打多个区域 → 用该命令自身的复数形态（`+cells-set --writes` / `+cells-batch-clear` / `+dim-delete --ranges` / resize 的 map 形态等）；只有跨类型的原子操作链才用本命令。
 
-**⚠️ 何时必须使用 `+batch-update`（硬性要求）**：
-- 需要对**多个**不同区域执行 `+cells-{merge|unmerge}` 时（如按分组合并多列相同内容）
-- 需要先插入行列再写入数据时（`+dim-{insert|delete|hide|unhide|freeze|group|ungroup}` + `+cells-set`）
-- 需要对多个区域执行不同写入操作时（多次 `+cells-set` + `+cells-clear` 等组合）
+**不可放进 `--operations` 的写 shortcut**（`shortcut` 枚举不含它们，强行写入会被校验拒）：`+cells-set-image`（需本地上传图片）、`+styles-put` / `+dropdown-update` / `+dropdown-delete` / `+cells-batch-clear`（自身已是批量入口，不可再嵌套）、`+dim-move`。这些操作需在 `+batch-update` 之外单独调用。
 
-**行高列宽批量不走这里**：多行 / 多列不同尺寸直接用 `+rows-resize --heights` / `+cols-resize --widths` 的 map 形态（如 `--widths '{"A":100,"C:E":120}'`，见 `lark-sheets-range-operations`），一次调用原子完成；map 形态不可作为 `--operations` 子操作嵌入（子操作里仍可用单区间形态 `range` + `height`/`width`）。
+**行高列宽批量不走这里**：多行 / 多列不同尺寸用 `+styles-put` 的 `row_sizes` / `col_sizes`（可与样式同批），或 `+rows-resize --heights` / `+cols-resize --widths` 的 map 形态（见 `lark-sheets-range-operations`）；map 形态不可作为 `--operations` 子操作嵌入（子操作里仍可用单区间形态 `range` + `height`/`width`）。
 
-当同一工具需要对多个区域重复调用时，**必须**改用 `+batch-update` 合并为单次请求——`+batch-update` 是原子提交（要么全成功要么整批回滚）；逐个调用非原子，中途失败会留下半成品。
+**执行语义（fail-fast，不回滚）**：默认首个失败的子操作即中断剩余操作，但**已执行成功的子操作不回滚**——服务端报 "N succeeded, M failed" 时前 N 个已实际生效。修复失败项后**只重发失败起的剩余子集**，整批重发会把已成功的操作（如插行）重复应用。传 `--continue-on-error` 则遇失败仍继续执行剩余操作。正因如此，含结构变更（插删行列 / 移动）的批次失败后要先回读确认现状再续发。
 
 **公式相关批处理的默认闭环**：
 - 写前：先读 `lark-sheets-formula-translation`，把公式改写成飞书可执行语义。
@@ -37,7 +34,6 @@
 | Shortcut | Risk | 分组 |
 | --- | --- | --- |
 | `+batch-update` | high-risk-write | 批量 |
-| `+cells-batch-set-style` | write | 批量 |
 | `+dropdown-update` | write | 对象 |
 | `+dropdown-delete` | high-risk-write | 对象 |
 | `+cells-batch-clear` | high-risk-write | 批量 |
@@ -50,29 +46,9 @@ _公共：URL/token（无 sheet 定位） · 系统：`--yes`、`--dry-run`_
 
 | Flag | Type | 必填 | 说明 |
 | --- | --- | --- | --- |
-| `--operations` | string + File + Stdin（复合 JSON） | required | JSON 数组：[{"shortcut":"+xxx-yyy","input":{...}}, ...]。shortcut 用 CLI 名；input 是该 shortcut 的入参集——含子表定位 sheet_id（或 sheet_name），但不含 spreadsheet token/url（后者只在顶层 --url/--spreadsheet-token 给一次；+batch-update 顶层没有 --sheet-id）；input 的键是该 shortcut 的 flag 展平成 JSON（如 "range":"A11:B12"），不是再套一层嵌套。基础 flag 查 --help，复合 JSON flag 查 --print-schema --flag-name <flag>；不要手填 operation 字段（由 CLI 按 shortcut 自动注入）。默认严格事务（首个失败即整批中断），传 --continue-on-error 切换为软批量（遇失败仍继续）；不支持嵌套；按数组顺序串行执行 |
+| `--operations` | string + File + Stdin（复合 JSON） | required | JSON 数组：[{"shortcut":"+xxx-yyy","input":{...}}, ...]。shortcut 用 CLI 名；input 是该 shortcut 的入参集——含子表定位 sheet_id（或 sheet_name），但不含 spreadsheet token/url（后者只在顶层 --url/--spreadsheet-token 给一次；+batch-update 顶层没有 --sheet-id）；input 的键是该 shortcut 的 flag 展平成 JSON（如 "range":"A11:B12"），不是再套一层嵌套。基础 flag 查 --help，复合 JSON flag 查 --print-schema --flag-name <flag>；不要手填 operation 字段（由 CLI 按 shortcut 自动注入）。默认 fail-fast：首个失败即中断剩余操作，**已执行的子操作不回滚**（服务端报 "N succeeded, M failed" 时 N 个已生效，修复后只重发失败起的剩余子集，不要整批重发）；传 --continue-on-error 遇失败仍继续；不支持嵌套；按数组顺序串行执行 |
 | `--continue-on-error` | bool | optional | 遇子操作失败时继续执行剩余操作；默认 false（首个失败即整批中断） |
 
-### `+cells-batch-set-style`
-
-_公共：URL/token（无 sheet 定位） · 系统：`--dry-run`_
-
-| Flag | Type | 必填 | 说明 |
-| --- | --- | --- | --- |
-| `--ranges` | string + File + Stdin（简单 JSON） | required | 目标范围 JSON 数组（最多 100 个），每项必须带 sheet 前缀（如 `["Sheet1!A1:B2","Sheet2!D1:D10"]`，前缀裸写不加引号）；前缀必须与 sheet 真实显示名完全一致（含大小写），不接受 sheet reference_id；支持跨 sheet；所有 range 应用同一组 style |
-| `--background-color` | string | optional | 背景颜色（十六进制，如 `#ffffff`） |
-| `--font-color` | string | optional | 字体颜色（十六进制，如 `#000000`） |
-| `--font-family` | string | optional | 字体名称（如 `Arial`、`微软雅黑`） |
-| `--font-size` | float64 | optional | 字体大小（px，例：10、12、14） |
-| `--font-style` | string | optional | 字体样式（可选值：`normal` / `italic`） |
-| `--font-weight` | string | optional | 字重（可选值：`normal` / `bold`） |
-| `--font-line` | string | optional | 字体线条样式（可选值：`none` / `underline` / `line-through`） |
-| `--horizontal-alignment` | string | optional | 水平对齐（可选值：`left` / `center` / `right`） |
-| `--vertical-alignment` | string | optional | 垂直对齐（可选值：`top` / `middle` / `bottom`） |
-| `--word-wrap` | string | optional | 换行策略（可选值：`overflow` / `auto-wrap` / `word-clip`） |
-| `--number-format` | string | optional | 数字格式（例：文本 `@`、数字 `0.00`、货币 `$#,##0.00`、日期 `mm/dd/yyyy`） |
-| `--border-styles` | string + File + Stdin（复合 JSON） | optional | 边框配置 JSON（结构同 +cells-set-style） |
-
 ### `+dropdown-update`
 
 _公共：URL/token（无 sheet 定位） · 系统：`--dry-run`_
@@ -115,16 +91,6 @@ _要批量执行的 CLI shortcut 操作列表，按声明顺序串行执行；
 - `shortcut` (enum) — CLI shortcut 名（不是底层 MCP tool 名） [+cells-set / +cells-set-style / +cells-clear / +cells-merge / +cells-unmerge / +cells-replace / +csv-put / +dropdown-set / +dim-insert / +dim-delete / +dim-hide / +dim-unhide / +dim-freeze / +dim-group / +dim-ungroup / +rows-resize / +cols-resize / +range-move / +range-copy / +range-fill / +range-sort / +sheet-create / +sheet-delete / +sheet-rename / +sheet-move / +sheet-copy / +sheet-hide / +sheet-unhide / +sheet-set-tab-color / +sheet-show-gridline / +sheet-hide-gridline / +chart-create / +chart-update / +chart-delete / +pivot-create / +pivot-update / +pivot-delete / +cond-format-create / +cond-format-update / +cond-format-delete / +filter-create / +filter-update / +filter-delete / +filter-view-create / +filter-view-update / +filter-view-delete / +sparkline-create / +sparkline-update / +sparkline-delete / +float-image-create / +float-image-update / +float-image-delete]
 - `input` (object) — 该 shortcut 的入参集——含子表定位 sheet_id（或 sheet_name），但不含 spreadsheet token/url（后者只在顶层 …
 
-### `+cells-batch-set-style` `--border-styles`
-
-_单元格边框配置，含 top/bottom/left/right 四个方向，每个方向的结构相同（见 top）_
-
-**顶层字段**：
-- `top` (object?) { style?: enum, weight?: enum, color?: string }
-- `bottom` (object?) { style?: enum, weight?: enum, color?: string }
-- `left` (object?) { style?: enum, weight?: enum, color?: string }
-- `right` (object?) { style?: enum, weight?: enum, color?: string }
-
 ### `+dropdown-update` `--options`
 
 _列表选项_
@@ -169,17 +135,6 @@ lark-cli sheets +batch-update --url "https://example.feishu.cn/sheets/shtXXX" --
 > ]
 > ```
 
-### `+cells-batch-set-style`
-
-多 range 应用同一组 style（服务端走 `+batch-update` 原子事务）：
-
-```bash
-# 表头行 + 汇总行同时刷成蓝底白字
-lark-cli sheets +cells-batch-set-style --url "..." \
-  --ranges '["sheet1!A1:F1","sheet1!A30:F30"]' \
-  --background-color "#1E5BC6" --font-color "#FFFFFF" --font-weight bold
-```
-
 ### `+cells-batch-clear`
 
 多 range 一次性清除（服务端走 `+batch-update` 原子事务）；`--scope` 同 `+cells-clear`（`content` / `formats` / `all`，默认 `content`），`high-risk-write` 强制 `--yes`：
@@ -195,6 +150,6 @@ lark-cli sheets +cells-batch-clear --url "..." \
 
 ### Validate / DryRun / Execute 约束
 
-- `Validate`：`+batch-update` 的 `--operations` 必须合法 JSON，且为非空数组；逐个子操作 `shortcut` / `input` 字段必填校验；**禁止嵌套 `+batch-update`**。`+cells-batch-set-style` 的 `--ranges` 必须 JSON 数组、每项带 sheet 前缀；样式 flag 至少一个非空（或带 `--border-styles`）。`+cells-batch-clear` 的 `--ranges` 同样必须 JSON 数组、每项带 sheet 前缀，`high-risk-write` 强制 `--yes` 或 `--dry-run`（`--scope` 默认 `content`）。
-- `DryRun`：按顺序输出每个子操作的目标 API + 请求 body 模板；首个失败则整批 fail-fast（不实际执行任何后续）。
-- `Execute`：按声明顺序串行执行；任一子操作失败立即中断并回滚到该子操作前状态（具体回滚能力取决于子操作类型，沿用 `+batch-update` 的语义）。
+- `Validate`：`+batch-update` 的 `--operations` 必须合法 JSON，且为非空数组；逐个子操作 `shortcut` / `input` 字段必填校验，input 键必须在该 shortcut 的 flag 词汇表内（未知键报错并提示最近似键与完整键契约）；**校验错误聚合上报**——所有子操作的首错一次性返回，全部修完再重发一次即可；**禁止嵌套 `+batch-update`**。`+cells-batch-clear` 的 `--ranges` 必须 JSON 数组、每项带 sheet 前缀，`high-risk-write` 强制 `--yes` 或 `--dry-run`（`--scope` 默认 `content`）。
+- `DryRun`：按顺序输出每个子操作的目标 API + 请求 body 模板，不发起调用。
+- `Execute`：按声明顺序串行执行；默认 fail-fast——任一子操作失败即中断剩余操作，**已成功的子操作不回滚**，报错会注明已生效数量与「仅重发失败起的剩余子集」的续发方式。

skills/lark-sheets/references/lark-sheets-chart.md

@@ -27,7 +27,7 @@
 
 **多图表需求**：当用户同时提到多种分析（如"统计占比 + 对比数量"），必须创建多个图表，每个对应一种类型，不要只做一个。
 
-**`--properties` 结构锚点（构造前必读）**：`--properties` 顶层只有 `position` / `offset` / `size` / `snapshot` 四个字段，**没有**顶层 `data`，也没有再嵌一层 `properties`。图表数据配置全部挂在 `snapshot.data` 下——下文及示例里出现的 `refs` / `headerMode` / `dim1` / `dim2` / `nameRef` 一律指 `snapshot.data.refs` / `snapshot.data.headerMode` / `snapshot.data.dim1` / `snapshot.data.dim2`（及其下的 `serie.nameRef` / `series[].nameRef`）；样式 / 堆叠 / 数据标签等在 `snapshot.plotArea` 下。完整结构以 `lark-cli sheets +chart-create --print-schema --flag-name properties` 为准。
+**`--properties` 结构锚点（构造前必读）**：`--properties` 顶层只有 `position` / `offset` / `size` / `snapshot` 四个字段，**没有**顶层 `data`，也没有再嵌一层 `properties`。图表数据配置全部挂在 `snapshot.data` 下——下文及示例里出现的 `refs` / `headerMode` / `dim1` / `dim2` / `nameRef` 一律指 `snapshot.data.refs` / `snapshot.data.headerMode` / `snapshot.data.dim1` / `snapshot.data.dim2`（及其下的 `serie.nameRef` / `series[].nameRef`）；样式 / 堆叠 / 数据标签等在 `snapshot.plotArea` 下。**构造起点优先用 `lark-cli sheets +chart-create --print-example <column|bar|line|area|pie|scatter|radar|combo>` 拿最小可用模板改参**（本地即时返回）；查深层字段用点分路径切片 `--print-schema --flag-name properties.snapshot.plotArea.axes`，别整篇 dump 翻页。完整结构以 `--print-schema --flag-name properties` 为准。
 
 **常见配置错误（必须注意）**：
 - **图表类型选择错误**：用户说"堆积柱形图/百分比堆积"时，应在 `properties.snapshot.plotArea.plot.extra.stack` 中配置堆叠；百分比堆叠需在该 stack 下设置 `percentage: true`。用户说"占比/比例"时，优先考虑饼图或百分比堆积图。注意区分 `column`（柱形图，纵向）与 `bar`（条形图，横向）是两个不同的 type 取值，"对比/各 XX" 类纵向柱默认用 `column`

skills/lark-sheets/references/lark-sheets-read-data.md

@@ -43,7 +43,8 @@
 注意：
 
 - `+csv-get` 和 `+cells-get` 支持分页/截断，注意检查 `has_more` / `truncated` 标志；两者在处理返回数据之前都必须先读 `warning_message`（上游 schema 要求先读它再用其它字段，内含定位与截断续读提示），`+cells-get` 还要用每个 range 的 `actual_range` / `row_indices` / `col_indices` 判断真实位置
-- 隐藏行列默认包含在返回结果中（`--skip-hidden=false`），如需只看可见数据设为 `true`。读取原语本身不标注哪些行列被隐藏：若要识别隐藏区间（以决定是否过滤、或如何解读混入的隐藏数据），用 `+sheet-info --include hidden_rows,hidden_cols` 取隐藏行列集合，再结合 `+csv-get` / `+cells-get` 返回的 `row_indices` / `col_indices` 判断每行 / 每列是否隐藏
+- `+cells-get` 可独立控制隐藏与筛选：`--skip-hidden` 只控制隐藏或分组折叠的行列，`--skip-filter` 只控制被筛选掉的行。两者都为 `false` 时返回完整数据；仅 `--skip-filter=true` 时保留隐藏行列但跳过筛选行；仅 `--skip-hidden=true --skip-filter=false` 时跳过隐藏行列但保留筛选行；两者都为 `true` 时只返回同时可见且未被筛选掉的数据。为兼容旧调用，未显式传 `--skip-filter` 时会继承 `--skip-hidden`。读取原语本身不标注哪些行列被隐藏：若要识别隐藏区间，用 `+sheet-info --include hidden_rows,hidden_cols` 取隐藏行列集合，再结合 `+cells-get` 返回的 `row_indices` / `col_indices` 判断真实位置
+- 要判断单元格内容是否被行高列宽挤到显示不全（排版检查、调整行高列宽前），给 `+cells-get` 加 `--include truncation`：会按字号 / 自动换行 / 行高列宽估算并返回被截断单元格的 `isRowTruncated` / `isColTruncated`（未返回视为未截断）。有额外计算开销，仅需要时才开
 
 **常见配置错误（必须注意）**：
 - **全量读取导致上下文溢出**：不要对大表（数百行以上）直接用 `+csv-get` 或 `+cells-get` 读取全部数据到上下文。大表场景必须分批读取：用 `--range` 切行窗口逐块读（`+csv-get` / `+cells-get` 单次返回量由 `--max-chars` 自动兜底，截断时返回 `has_more`）；过大时考虑导出到本地文件后用脚本处理再分批回写
@@ -53,7 +54,7 @@
   - 数据量大或会进入上下文上限时，分批读 + 本地处理 + 分批回写，不要一口气拉全表到上下文。
 - **`+cells-get` 滥用**：当只需要数据值时，使用 `+csv-get`（token 开销约为 `+cells-get` 的 1/5）。只有确实需要公式、样式或批注时才用 `+cells-get`
 - **忽略分页标志**：读取返回 `has_more=true` 时，说明还有更多数据。如果任务需要完整数据，必须继续分页读取，不能只处理第一页就开始写入
-- **直接按 `+cells-get` 返回二维数组下标推导真实位置**：`ranges[n].cells[i][j]` 里的 `i/j` 只是返回数组下标，不等于真实表格行列。定位真实行号必须用 `ranges[n].row_indices[i]`，定位真实列字母必须用 `ranges[n].col_indices[j]`；若 `--skip-hidden=true`、请求范围越界被裁剪，或最后一行是部分返回，错误地自己数下标会立刻错位
+- **直接按 `+cells-get` 返回二维数组下标推导真实位置**：`ranges[n].cells[i][j]` 里的 `i/j` 只是返回数组下标，不等于真实表格行列。定位真实行号必须用 `ranges[n].row_indices[i]`，定位真实列字母必须用 `ranges[n].col_indices[j]`；若 `--skip-hidden=true`、`--skip-filter=true`、请求范围越界被裁剪，或最后一行是部分返回，错误地自己数下标会立刻错位
 - **CSV 行号计数错误**：`+csv-get` 返回的 CSV 遵循 RFC 4180 标准，被双引号 `"..."` 包裹的字段中的换行符属于**字段内容的一部分**（即单元格内换行），不代表新的一行。计算行号时必须按**逻辑记录**计数，而非按物理换行符 `\n` 计数
 - **手动数列确定列号**：禁止通过在 CSV 表头中手动数逗号/字段来确定目标列的列字母。当列数超过 10 时，手动计数极易产生 off-by-one 偏移（例如把 W 列误判为 X 列）。**必须使用 `col_indices`**：先在 CSV 表头中找到目标字段名是第 j 个字段（0-based），再用 `col_indices[j]` 获取该列的实际列字母
 - **用数据列的值推导行号（常被巧合掩盖）**：CSV 中常见"序号 / ID / 编号 / No."等形似行号的列，其值与实际表格行号**没有任何绑定关系**——序号可能跳号（1,2,3,5,6...）、可能从非 1 开始、可能有重复或被中途重置。此规则适用于**所有需要行号的下游操作**：合并单元格、区间写入/清空/格式化、插入/删除行、条件格式范围、筛选器范围、图表数据源、透视表范围、搜索替换范围等等——**凡是要把行号填进任何工具参数的场景，行号一律从 `annotated_csv` 中目标行开头的 `[row=N]` 前缀直接读取**，禁止用"序号=行号"、"表头占 1 行所以数据从第 2 行开始"、"第 N 个序号就在第 N+1 行"等心算，也禁止先心算再"事后核对"。**危险特征**：前几十行中序号恰好等于表格行号（典型成因：表头 +1 与一次跳号 -1 的偏移互相抵消形成巧合），模型一旦把这个巧合当作规律，会在后续所有行沿用；而中间再出现跳号时，从该行起整块区域全部错位，且错位不自查很难发现。**正确工作流**：①在 `annotated_csv` 里定位目标逻辑行（按字段内容匹配）；②直接读取该行开头的 `[row=N]` 前缀得到真实表格行号；③把这个行号填进下游工具参数。区间操作时，起始行用 start 行的 `[row=N]`、结束行用 end 行的 `[row=N]`。**自检**：动手前，在 `annotated_csv` 靠后位置再抽 1~2 行，核对 `[row=N]` 是否与首列"序号"一致——不一致（典型：`[row=57] 58,...`）即说明有跳号/隐藏行，更要严格从 `[row=N]` 取值，不要被序号列迷惑
@@ -99,9 +100,11 @@ _公共四件套 · 系统：`--dry-run`_
 | Flag | Type | 必填 | 说明 |
 | --- | --- | --- | --- |
 | `--range` | string | required | A1 范围，如 `A1:F10`（不带 sheet 前缀；用 `--sheet-id` / `--sheet-name` 指定 sheet） |
-| `--include` | string_slice | optional | 要返回的信息类别，逗号分隔多个（可选值：`value` / `formula` / `style` / `comment` / `data_validation`） |
-| `--max-chars` | int | optional | 单次返回字符上限，默认 500000（兜底防爆）。大数据通常宜重定向落盘做分析；仅当要让结果直接进上下文、又不触发文件转存时才调小（如 25000），以 has_more 分页 |
-| `--skip-hidden` | bool | optional | 跳过隐藏行列，默认 `false` |
+| `--include` | string_slice | optional | 要返回的信息类别，逗号分隔多个。`truncation` 会额外按行高列宽 / 字号 / 自动换行估算每个单元格是否被截断显示，返回 `isRowTruncated` / `isColTruncated`（有额外计算开销，仅排版检查 / 调整行高列宽前才开）（可选值：`value` / `formula` / `style` / `comment` / `data_validation` / `truncation`） |
+| `--max-chars` | int | optional | 单次返回字符上限，默认 500000（兜底防爆）。要整表无截断直接用 --output-path 落盘（自动放开为无限）；仅当要让结果直接进上下文、又不落盘时才调小（如 25000），按 has_more 分页。 |
+| `--output-path` | string | optional | 把完整读取结果写入本地路径（如 `./out.json`），文件内容为 data 载荷的 JSON；stdout 只回一个含 output_path/字节数的确认信息。**一旦设置，字符上限默认放开为无限**（覆盖 --max-chars 默认），适合大表整表落盘再分析，避免 stdout 被 max_chars 截断。省略时按常规把结果打到 stdout。 |
+| `--skip-hidden` | bool | optional | 跳过隐藏或分组折叠的行列。默认 `false`；未显式传 `--skip-filter` 时，筛选行也会随之跳过（兼容旧行为） |
+| `--skip-filter` | bool | optional | 跳过被筛选掉的行。未设置时继承 `--skip-hidden`；显式设为 `false` 可在跳过隐藏行列的同时保留筛选行 |
 
 ### `+dropdown-get`
 
@@ -117,8 +120,9 @@ _公共四件套 · 系统：`--dry-run`_
 
 | Flag | Type | 必填 | 说明 |
 | --- | --- | --- | --- |
-| `--range` | string | required | A1 范围，如 `A1:F30`（不带 sheet 前缀；用 `--sheet-id` / `--sheet-name` 指定 sheet） |
-| `--max-chars` | int | optional | 单次返回字符上限，默认 500000（兜底防爆）。大数据通常宜重定向落盘做分析；仅当要让结果直接进上下文、又不触发文件转存时才调小（如 25000），以 has_more 分页 |
+| `--range` | string | optional | A1 范围，如 `A1:F30`（不带 sheet 前缀；用 `--sheet-id` / `--sheet-name` 指定 sheet）。**可省略：缺省读取整个子表**（按表格实际边界裁剪，返回的 actual_range 标注实际读取范围）；大表配合 --max-chars / --output-path 控制体量 |
+| `--max-chars` | int | optional | 单次返回字符上限，默认 500000（兜底防爆）。要整表无截断直接用 --output-path 落盘（自动放开为无限）；仅当要让结果直接进上下文、又不落盘时才调小（如 25000），按 has_more 分页。 |
+| `--output-path` | string | optional | 把完整读取结果写入本地路径（如 `./out.json`），文件内容为 data 载荷的 JSON；stdout 只回一个含 output_path/字节数的确认信息。**一旦设置，字符上限默认放开为无限**（覆盖 --max-chars 默认），适合大表整表落盘再分析，避免 stdout 被 max_chars 截断。省略时按常规把结果打到 stdout。 |
 | `--include-row-prefix` | bool | optional | 是否在每行前加 `[row=N]` 前缀，默认 `true` |
 | `--skip-hidden` | bool | optional | 跳过隐藏行列，默认 `false` |
 
@@ -131,6 +135,8 @@ _公共：URL/token（无 sheet 定位） · 系统：`--dry-run`_
 | `--sheet-id` | string | optional | 只读该子表（按 id）；省略则读所有子表 |
 | `--sheet-name` | string | optional | 只读该子表（按名）；省略则读所有子表 |
 | `--range` | string | optional | 读取的 A1 范围；省略则读每个子表的完整 used range（会跨过表中部的整行空行 / 整列空列，不会被截断） |
+| `--max-chars` | int | optional | 单次返回字符上限，默认 500000（兜底防爆）。底层工具即使不传也有约 50000 的默认截断，故此处显式发送以放宽；要整表无截断请用 --output-path 落盘（自动放开为无限）。 |
+| `--output-path` | string | optional | 把完整读取结果写入本地路径（如 `./out.json`），文件内容为 data 载荷的 JSON；stdout 只回一个含 output_path/字节数的确认信息。**一旦设置，字符上限默认放开为无限**（覆盖 --max-chars 默认），适合大表整表落盘再分析，避免 stdout 被 max_chars 截断。省略时按常规把结果打到 stdout。 |
 | `--no-header` | bool | optional | 把第一行当数据而非表头（列名取 col1/col2 …） |
 
 ## Examples
@@ -147,6 +153,10 @@ lark-cli sheets +csv-get --url "https://example.feishu.cn/sheets/shtXXX" --sheet
 
 # 用 sheet-name 模糊定位（运行时框架会先解析到 sheet-id）
 lark-cli sheets +csv-get --spreadsheet-token shtXXX --sheet-name "销售明细" --range "A1:F30"
+
+# 全量读：省略 --range 即读整个子表（按实际边界裁剪，返回 actual_range 标注实读范围），
+# 无需先 +workbook-info 探行列再拼 range；大表配合 --max-chars / --output-path
+lark-cli sheets +csv-get --spreadsheet-token shtXXX --sheet-name "销售明细"
 ```
 
 输出契约（envelope.data）：

skills/lark-sheets/references/lark-sheets-sheet-structure.md

@@ -23,7 +23,7 @@
 
 - 当表格存在合并单元格时，应结合返回的 `merged_cells` 判断表头、分组标题和区域语义
 - 不要把合并区域中非左上角的空白单元格理解为"无内容"；通常应将左上角单元格的内容视为整个合并区域的语义内容
-- 插入用 `+dim-insert`：`--position`（插入位置；行用 1-based 行号如 `3`，列用字母如 `C`，新行/列插在此位置**之前**）+ `--count`（插入数量，>0）。新行/列样式继承用 `--inherit-style`（`before`/`after`/`none`）
+- 插入用 `+dim-insert`：`--position`（插入位置；行用 1-based 行号如 `3`，列用字母如 `C`，新行/列插在此位置**之前**）+ `--count`（插入数量，>0）。新行/列样式继承用 `--inherit-style`（`before` 继承前一行/列 / `after` 继承后一行/列）；它只决定继承哪一侧的样式，**插入位置始终在 `--position` 之前，不改变插入方向**。⚠️ 不传时默认继承**后一行/列**（同 `after`）；底层无法插入"无格式"行/列，要真正的纯空白行/列，插入后再用 `+cells-clear --scope formats` 清除新行/列的格式。
 - 例如"在第 20 行后新增 116 行"：`--position 21 --count 116`（"第 20 行后"即 1-based 行号 21）
 
 **区间表达统一为 A1 风格**：所有涉及"一段连续行/列"的 shortcut 都用同一套 A1 闭区间字符串语法，**不存在 inclusive / exclusive / 0-based / 1-based 跨命令差异**：
@@ -40,7 +40,7 @@
 - **插入列直接用字母**：`+dim-insert` 的 `--position` 在列场景直接传字母（如 `C`），不要把列字母换算成 0-based 索引
 - **插入后引用偏移**：插入行/列后，原有数据的行号 / 列字母会发生偏移。如果插入后还需要对原有区域执行写入操作，必须重新计算偏移后的位置
 - **删除行列前先确认范围**：删除操作不可逆，执行前应确认 `--range` 精确无误。可先用 `+csv-get` 读取目标区域验证内容（`+csv-get` / `+cells-get` 见 `lark-sheets-read-data`）
-- **"在 D 列左侧新增一列"的正确写法**：`--position D --count 1`（新列插在 D 列之前）；要继承左侧列样式加 `--inherit-style before`
+- **"在 D 列左侧新增一列"的正确写法**：`--position D --count 1`（新列插在 D 列之前）；要继承左侧列样式加 `--inherit-style before`。不要把 `--inherit-style after` 当成“插到 D 列右侧”，它不是插入方向参数。
 - **`+dim-move` 同维度约束**：`--source-range` 是行区间时 `--target` 必须是行号（数字），是列区间时 `--target` 必须是列字母——不可一行一列混用
 - **插入列后必须检查多行表头合并区域**：很多表格有 2-3 行的合并表头。插入列后，原有的合并区域不会自动扩展到新列。必须先用 `+sheet-info --include merges` 读取合并区域，插入后将跨越插入位置的合并区域重新设置（用 `+cells-{merge|unmerge}`），否则新列的表头会是空的、格式不连续
 - **公式写入范围跳过表头行**：写入公式时从数据行开始（不是第 1 行）。先确认表头占几行（可能 1-3 行），公式的起始行 = 表头行数 + 1
@@ -76,7 +76,7 @@ _公共四件套 · 系统：`--dry-run`_
 
 | Flag | Type | 必填 | 说明 |
 | --- | --- | --- | --- |
-| `--inherit-style` | string | optional | 新行/列样式继承策略 enum：`before`（继承前一行/列）/ `after`（继承后一行/列）/ `none`（默认）（可选值：`before` / `after` / `none`） |
+| `--inherit-style` | string | optional | 新行/列样式继承 enum：`before`（继承前一行/列）/ `after`（继承后一行/列）；不传时默认继承后一行/列（同 `after`），底层无法插入无格式行/列。只决定继承哪侧样式、不改变插入方向（始终插在 `--position` 之前）；要纯空白行/列请插入后用 `+cells-clear --scope formats`（可选值：`before` / `after`） |
 | `--position` | string | required | 插入位置（在此行/列**之前**插入）：行用 1-based 行号如 `3`；列用字母如 `C` |
 | `--count` | int | required | 插入数量（>0） |
 
@@ -86,7 +86,8 @@ _公共四件套 · 系统：`--yes`、`--dry-run`_
 
 | Flag | Type | 必填 | 说明 |
 | --- | --- | --- | --- |
-| `--range` | string | required | 要删除的行/列闭区间；行用 1-based 数字如 `3:7` 或单行 `5`，列用字母如 `C:F` 或单列 `C` |
+| `--range` | string | xor | 要删除的行/列闭区间；行用 1-based 数字如 `3:7` 或单行 `5`，列用字母如 `C:F` 或单列 `C`。与 `--ranges` 二选一 |
+| `--ranges` | string + File + Stdin（简单 JSON） | xor | 要删除的多个行/列区间 JSON 数组（最多 100 个，如 `["5:5","8:8","11:13"]` 或 `["C:C","F:G"]`），全行或全列不可混用，区间不可重叠；与 `--range` 二选一。CLI 按位置**从大到小逆序**合成一次原子批量删除——正序删除会因前面的行/列被删导致后续索引前移错位，逆序由 CLI 代劳，无需自行排序 |
 
 ### `+dim-hide`
 
@@ -168,6 +169,11 @@ lark-cli sheets +dim-delete --url "..." --sheet-id "$SID" --range "5:7" --yes
 
 # 删除 D-F 列
 lark-cli sheets +dim-delete --url "..." --sheet-id "$SID" --range "D:F" --yes
+
+# 删除多个散布区间（如按查重结果删行）：--ranges 一次原子交付。
+# CLI 自动按位置从大到小逆序执行——正序会因前面的行被删导致后续索引前移错位；
+# 无需自行排序，也不要为此拼 +batch-update 的子操作数组
+lark-cli sheets +dim-delete --url "..." --sheet-id "$SID" --ranges '["5:5","8:8","11:13"]' --yes
 ```
 
 ### `+dim-hide` / `+dim-unhide`
@@ -207,6 +213,6 @@ lark-cli sheets +dim-freeze --url "..." --sheet-id "$SID" --dimension row --coun
 
 ### Validate / DryRun / Execute 约束
 
-- `Validate`：XOR 公共四件套；`--range` / `--source-range` 必须是合法 A1 闭区间（行用数字、列用字母，不可混用）；`+dim-insert` 的 `--count` > 0；`+dim-move` 的 `--target` 必须与 `--source-range` 同维度（行 vs 列）；`+dim-delete` 强制 `--yes` 或 `--dry-run`；`+rows-resize` / `+cols-resize` 的统一形态（`--range` + `--height`/`--width` 或 `--type`）与 map 形态（`--heights`/`--widths`）二选一、不可混用；详见 `lark-sheets-range-operations.md`。
+- `Validate`：XOR 公共四件套；`--range` / `--source-range` 必须是合法 A1 闭区间（行用数字、列用字母，不可混用）；`+dim-insert` 的 `--count` > 0；`+dim-move` 的 `--target` 必须与 `--source-range` 同维度（行 vs 列）；`+dim-delete` 强制 `--yes` 或 `--dry-run`，`--range` 与 `--ranges` 二选一、`--ranges` 各区间同维度且不可重叠（≤100 个）；`+rows-resize` / `+cols-resize` 的统一形态（`--range` + `--height`/`--width` 或 `--type`）与 map 形态（`--heights`/`--widths`）二选一、不可混用；详见 `lark-sheets-range-operations.md`。
 - `DryRun`：写操作输出"将要 PATCH 的目标范围 + 目标参数"。
 - `Execute`：写后不自动回读；如需确认，自行调用 `+sheet-info --include row_heights,col_widths,hidden_rows,hidden_cols,groups,frozen` 查看受影响的范围。

skills/lark-sheets/references/lark-sheets-styles-put.md

@@ -0,0 +1,91 @@
+# Lark Sheet Styles Put（+styles-put）
+
+> **本文定位**：对**已有**表格做美化收尾的默认入口——样式 / 边框 / 合并 / 行高列宽 / 冻结写成一份声明式规格，一次调用原子交付。样式**取什么值**（配色 / 字号 / 对齐 / 数字格式标准）以 `lark-sheets-visual-standards` 为唯一权威，本文只讲**怎么落地**。
+>
+> **边界（三分流判定，按操作组合选入口）**：目标是**样式 / 合并 / 行高列宽 / 冻结**的任意组合 → 本命令；**同一个写操作**打多个区域（如多区域清除、批量下拉）→ 用该命令自身的复数形态（`--ranges` / map 入参）；操作链**跨类型且有顺序依赖**（如插列 → 写表头 → 回填数据）→ `+batch-update`。美化收尾不需要也不应该拼 `--operations` 子操作数组。
+
+## 使用场景
+
+写入。对存量表格的多个子表批量应用视觉规格：新表美化、加汇总行后统一版式、按分组合并同类单元格、调列宽行高、冻结表头。整份规格展开为一次原子批量提交，全部生效或全部不生效；纯样式盖章可安全重放（同一份规格重发无副作用）。
+
+**词汇三处同构**：`--styles` 的字段词汇与 `+workbook-create --styles`（建新表同步美化）、`+table-put --styles`（写数据同步美化）完全一致——`cell_styles` / `cell_merges` / `row_sizes` / `col_sizes` / `freeze` 学一次三处通用。区别只有两点：本命令作用于**已有**表格（顶层 `--url` / `--spreadsheet-token` 定位），且 `cell_styles` 的 range 不受「本次写入区域」限制、可指向表内任意区域。
+
+**规格要点**：
+
+- 顶层 `{styles:[...]}`，每项对应一个目标子表，`name` 必须是真实子表名（不确定先 `+workbook-info` 查，禁止猜 `Sheet1`）。
+- 每个子表项按固定顺序执行：`cell_merges` → `cell_styles` → `row_sizes` → `col_sizes` → `freeze`；样式盖章允许覆盖含合并区的区域（合并区限制只针对值写入，样式不受限）。
+- `row_sizes` / `col_sizes` 只需 `{range, size}`（px，即像素尺寸；`standard` / 行的 `auto` 才需显式 `type`）。尺寸键统一是 `size`。
+- 加边框用 `border` 简写：`{"style":"solid","color":"#DDDDDD"}` 应用到四边；只有分侧不同样式才用 `border_styles` 完整形态。
+- `freeze` 用 `{rows:N, cols:N}` 冻结前 N 行 / 列，0 或省略表示该维度不冻结。
+
+**回读校验**：整份规格执行成功后按编辑准则抽样回读受影响区域（`+cells-get --include style` 或 `+sheet-info` 看合并 / 行高列宽 / 冻结），确认关键样式实际生效。
+
+## Shortcuts
+
+| Shortcut | Risk | 分组 |
+| --- | --- | --- |
+| `+styles-put` | write | 批量 |
+
+## Flags
+
+### `+styles-put`
+
+_公共：URL/token（无 sheet 定位） · 系统：`--dry-run`_
+
+| Flag | Type | 必填 | 说明 |
+| --- | --- | --- | --- |
+| `--styles` | string + File + Stdin（复合 JSON） | required | 对**已有**表格应用的视觉规格 JSON：顶层 `{styles:[...]}`，每项对应一个目标子表（`name` 用真实子表名），并至少给 `cell_styles` / `cell_merges` / `row_sizes` / `col_sizes` / `freeze` 之一。字段词汇与 `+workbook-create` / `+table-put` 的 `--styles` 完全同构（cell_styles 用 A1 range + 扁平样式字段，边框用 `border` 简写 {style,weight,color} 四边同款、分侧才用 border_styles；row/col sizes 用行/列范围 + size（px 即像素，standard/auto 才需 type）；merges 用单元格 range；freeze 用 `{rows:N, cols:N}` 冻结前 N 行/列）。整份规格展开为一次原子批量提交；range 不受「本次写入区域」限制，可指向表内任意区域 |
+
+## Schemas
+
+> 复合 JSON flag 字段速查（只列顶层 + 一层嵌套）。深层结构看下方 `## Examples`，或用 `--print-schema` 读完整 JSON Schema（用法见 SKILL.md「公共 flag 速查」与「Agent 使用提示」）。
+
+### `+styles-put` `--styles`
+
+
+**数组项**（类型 object）：
+- `cell_merges` (array<object>?) — 单元格合并操作数组；range 使用 A1 单元格范围，merge_type 默认 all each: { merge_type?: enum, range: string }
+- `cell_styles` (array<object>?) — 单元格样式操作数组；每项用 A1 单元格 range 指定范围，字段名与 +cells-set-style 对齐 each: { background_color?: string, border?: object, border_styles?: object, font_color?: string, font_family?: string, …共 14 项 }
+- `col_sizes` (array<object>?) — 列宽操作数组；range 使用列范围如 A:C，给 size（px）即像素列宽（type 可省略）；type 为 standard 时不带 size each: { range: string, size?: number, type?: enum }
+- `freeze` (object?) — 冻结行列：rows = 冻结前 N 行，cols = 冻结前 N 列（0 或省略 = 该维度不冻结） { cols?: integer, rows?: integer }
+- `name` (string) — 子表名
+- `row_sizes` (array<object>?) — 行高操作数组；range 使用行范围如 1:3，给 size（px）即像素行高（type 可省略）；type 为 standard/auto 时不带 size each: { range: string, size?: number, type?: enum }
+
+## Examples
+
+### `+styles-put`
+
+表头美化 + 按组合并 + 列宽 + 冻结首行，一次交付：
+
+```bash
+lark-cli sheets +styles-put --url "https://example.feishu.cn/sheets/shtXXX" --styles - <<'JSON'
+{"styles":[{
+  "name": "Sheet1",
+  "cell_merges": [{"range":"A5:A8"},{"range":"A9:A12"}],
+  "cell_styles": [
+    {"range":"A1:F1","font_weight":"bold","background_color":"#1E5BC6","font_color":"#FFFFFF","horizontal_alignment":"center"},
+    {"range":"A2:F30","border":{"style":"solid","color":"#DDDDDD"}}
+  ],
+  "row_sizes":  [{"range":"1:1","size":36}],
+  "col_sizes":  [{"range":"A:C","size":120}],
+  "freeze":     {"rows":1}
+}]}
+JSON
+```
+
+多子表同一批交付（每个子表一个 styles 项）：
+
+```bash
+lark-cli sheets +styles-put --url "..." --styles - <<'JSON'
+{"styles":[
+  {"name":"明细","cell_styles":[{"range":"A1:H1","font_weight":"bold","background_color":"#F0F0F0"}],"freeze":{"rows":1}},
+  {"name":"汇总","cell_styles":[{"range":"A1:D1","font_weight":"bold"}],"col_sizes":[{"range":"A:D","type":"pixel","size":140}]}
+]}
+JSON
+```
+
+### Validate / DryRun / Execute 约束
+
+- `Validate`：`--styles` 必须是合法 JSON、`styles` 非空数组；每项 `name` 必填、至少给 `cell_merges` / `cell_styles` / `row_sizes` / `col_sizes` / `freeze` 之一；`cell_styles` 每项至少一个样式字段；展开后受子操作数（100）与总格数预算约束，超限报错给拆分建议。
+- `DryRun`：输出展开后每个子操作的请求模板，不发起调用。
+- `Execute`：整份规格合成一次批量请求按序执行；失败时报错会注明已生效的子操作区间与续发方式。

skills/lark-sheets/references/lark-sheets-workbook.md

@@ -197,10 +197,11 @@ _一个或多个子表的 typed 数据，每个数组元素写入一张子表；
 
 **数组项**（类型 object）：
 - `cell_merges` (array<object>?) — 单元格合并操作数组；range 使用 A1 单元格范围，merge_type 默认 all each: { merge_type?: enum, range: string }
-- `cell_styles` (array<object>?) — 单元格样式操作数组；每项用 A1 单元格 range 指定范围，字段名与 +cells-set-style 对齐 each: { background_color?: string, border_styles?: object, font_color?: string, font_family?: string, font_line?: enum, …共 13 项 }
-- `col_sizes` (array<object>?) — 列宽操作数组；range 使用列范围如 A:C，type 为 pixel/standard，pixel 需要 size each: { range: string, size?: number, type: enum }
+- `cell_styles` (array<object>?) — 单元格样式操作数组；每项用 A1 单元格 range 指定范围，字段名与 +cells-set-style 对齐 each: { background_color?: string, border?: object, border_styles?: object, font_color?: string, font_family?: string, …共 14 项 }
+- `col_sizes` (array<object>?) — 列宽操作数组；range 使用列范围如 A:C，给 size（px）即像素列宽（type 可省略）；type 为 standard 时不带 size each: { range: string, size?: number, type?: enum }
+- `freeze` (object?) — 冻结行列：rows = 冻结前 N 行，cols = 冻结前 N 列（0 或省略 = 该维度不冻结） { cols?: integer, rows?: integer }
 - `name` (string) — 子表名
-- `row_sizes` (array<object>?) — 行高操作数组；range 使用行范围如 1:3，type 为 pixel/standard/auto，pixel 需要 size each: { range: string, size?: number, type: enum }
+- `row_sizes` (array<object>?) — 行高操作数组；range 使用行范围如 1:3，给 size（px）即像素行高（type 可省略）；type 为 standard/auto 时不带 size each: { range: string, size?: number, type?: enum }
 
 ## Examples
 

skills/lark-sheets/references/lark-sheets-write-cells.md

@@ -73,7 +73,7 @@
 
 > 以下是用 `+cells-set`（及 `+cells-set-style`）做富写入时的常用模式与准则；选哪个 shortcut 见上方「使用场景」。
 
-`+cells-set` 为一块区域设置值 / 公式 / 批注 / 样式，也支持 `rich_text` 的 `type: "embed-image"` 嵌入单元格图片。**关键：`cells` 二维数组的行列维度必须与 `range`（闭区间）严格一致，否则触发 `InvalidCellRangeError`**——维度计算示例见文末 `## Schemas` 的 `--cells`。
+`+cells-set` 为一块区域设置值 / 公式 / 批注 / 样式，也支持 `rich_text` 的 `type: "embed-image"` 嵌入单元格图片。**关键：`--cells` 恒为二维数组（行 × 格），单格也是 `[[{"value":…}]]`；且行列维度必须与 `range`（闭区间）严格一致，否则触发 `InvalidCellRangeError`**——维度计算示例见文末 `## Schemas` 的 `--cells`。
 
 > **单元格图片 vs 浮动图片（最易选错）**：图若**属于某条记录、要随那行排序 / 筛选 / 增删**（凭证 / 证件照 / 每行配图，话里带「对应 / 每行 / 这列」等绑定词）→ **单元格图片**（本工具）：用 `+cells-set-image`（最短）或 `+cells-set` 的 `rich_text` + `type: "embed-image"`。只是自由摆放的装饰（logo / 水印 / 封面）→ 浮动图片，见 lark-sheets-float-image。别因「浮动图更好控制 / 更熟」默认选浮动图——它承载"对应某记录"的图会随增删行 / 排序错位。
 
@@ -89,6 +89,19 @@
 
 ⚠️ **逐行写入公式是常见低效写法**：对每一行单独调用 `+cells-set` 写公式（如 26 次）既慢又易错，且不会自动平移公式引用。正确做法是 1 次模板写入 + 1 次 `--copy-to-range`（公式引用自动平移）。
 
+💡 **多个不连续区域写入（批量修公式的正解）**：散布多处（可跨 sheet）的值 / 公式写入，用 `--writes` 一次原子交付——每项 `{sheet_name, range, cells}`（sheet 定位必须写在每项里），不要为此拼 `+batch-update` 的 `--operations`，也不要逐区域多次调用（非原子）：
+
+```bash
+lark-cli sheets +cells-set --url "..." --writes - <<'JSON'
+[
+  {"sheet_name":"明细","range":"D5","cells":[[{"formula":"=IFERROR(C5/B5,0)"}]]},
+  {"sheet_name":"汇总","range":"B3","cells":[[{"formula":"=SUM(明细!C:C)"}]]}
+]
+JSON
+```
+
+范围级统一样式不在 `--writes` 里做（cells 逐格 `cell_styles` 仅用于逐格差异化），写完接 `+styles-put`。
+
 💡 **写入公式前先按迁移规则改写**：如果公式来自 Excel 或包含数组场景，先读取并遵循 `lark-sheets-formula-translation` 的规则完成改写，再把最终公式写入 `formula` 字段。
 
 💡 **内容与样式分离写入（推荐）**：当需要同时写入内容和样式时，`cells` 中每个单元格都带上 `cell_styles` / `border_styles` 会导致入参非常冗长。由于同一区域的样式通常高度重复（如整列统一背景色、统一边框），推荐拆成两步：
@@ -102,7 +115,7 @@ Step 2: `+cells-set` — range="A2", cells 含 value + cell_styles + border_styl
 ```
 这比在 99 个单元格中都重复写样式 JSON 高效得多。
 
-💡 **样式更新是「部分合并」，不是整体覆盖**：`+cells-set-style` / `+cells-batch-set-style`（以及 `+cells-set` 的 `cell_styles` / `border_styles`）只改你**显式传入**的样式属性，未传的属性保留原值。两个实用推论：
+💡 **样式更新是「部分合并」，不是整体覆盖**：`+cells-set-style` / `+styles-put`（以及 `+cells-set` 的 `cell_styles` / `border_styles`）只改你**显式传入**的样式属性，未传的属性保留原值。两个实用推论：
 - **可分层叠加**：对同一区域先刷字体色、再单独刷背景色、再单独刷边框，后一步不会清掉前一步——美化已有区域时无需一次带齐所有字段，可拆成多次窄调用。
 - **`border_styles` 按边合并**：只传 `{"top":{...}}` 只更新上边框，`bottom` / `left` / `right` 保留原状；不必为了「只改一条边」而把四边全部重传。（例外见上方「新增行的边框/样式禁止用 `{}` 跳过」：**全新行**底子里没有边框，仍需把要显示的边都显式传出。）
 
@@ -236,7 +249,7 @@ lark-cli sheets +dropdown-set \
 
 > ⚠️ **`--source-range` 必须带 sheet 前缀**（即使跟 `--range` 同 sheet）。注意一个坑：回读这种 listFromRange 下拉单元格时，`data_validation.range` 看起来不带 sheet 前缀（形如 `$T$1:$T$3`），如果要把读出来的 range 反过来写回 `--source-range`，**必须自己重新补上 sheet 前缀**，否则会被拒。
 >
-> ⚠️ **`--ranges` 类批量 flag 的 sheet 前缀必须「裸写」**——`+cells-batch-set-style` / `+cells-batch-clear` / `+dropdown-update` / `+dropdown-delete` 的 `--ranges` 解析器不接受引号：表名含点或空格（如 `2025.9`、`一月份`）也直接写 `2025.9!A1`，写成 `'2025.9'!A1` 会被当成表名一部分、报 `sheet not found`。**但 `--source-range`、透视表 `--source`、`--range` 走 A1 标准**：sheet 名带单引号（如 `'Sheet1'!A1:B2`）是标准写法、裸写也接受，回读统一返回带引号形式——别把 `--ranges` 的裸写要求套到这些 flag 上。
+> ⚠️ **`--ranges` 类批量 flag 的 sheet 前缀必须「裸写」**——`+cells-batch-clear` / `+dropdown-update` / `+dropdown-delete` 的 `--ranges` 解析器不接受引号：表名含点或空格（如 `2025.9`、`一月份`）也直接写 `2025.9!A1`，写成 `'2025.9'!A1` 会被当成表名一部分、报 `sheet not found`。**但 `--source-range`、透视表 `--source`、`--range` 走 A1 标准**：sheet 名带单引号（如 `'Sheet1'!A1:B2`）是标准写法、裸写也接受，回读统一返回带引号形式——别把 `--ranges` 的裸写要求套到这些 flag 上。
 
 `+dropdown-update`（多 range 批量更新）的所有 flag 语义与 `+dropdown-set` 完全一致；只是目标 `--ranges` 由单值变成 JSON 数组（每项带 sheet 前缀），同一份选项 + 配色应用到所有 range。
 
@@ -259,8 +272,9 @@ _公共四件套 · 系统：`--dry-run`_
 
 | Flag | Type | 必填 | 说明 |
 | --- | --- | --- | --- |
-| `--range` | string | required | 写入区域（A1 格式） |
-| `--cells` | string + File + Stdin（复合 JSON） | required | JSON：2D 数组 `[[{cell},...],...]`，维度与 `--range` 完全一致；每个 cell 可含 `value` / `formula` / `cell_styles` / `note` / `rich_text`（含 `type="embed-image"` 单元格嵌图）等，完整字段跑 `--print-schema` |
+| `--range` | string | xor | 写入区域（A1 格式）。与 `--writes` 二选一（单区域用 --range+--cells，多区域用 --writes） |
+| `--cells` | string + File + Stdin（复合 JSON） | xor | JSON：2D 数组 `[[{cell},...],...]`，维度与 `--range` 完全一致；每个 cell 可含 `value` / `formula` / `cell_styles` / `note` / `rich_text`（含 `type="embed-image"` 单元格嵌图）等，完整字段跑 `--print-schema` |
+| `--writes` | string + File + Stdin（复合 JSON） | xor | 多区域写入 JSON 数组（最多 100 项），每项 `{sheet_name\|sheet_id, range, cells}`——**sheet 定位必须写在每项里**（与 +batch-update 子操作、+styles-put 项同惯例，不认顶层 --sheet-name），cells 结构同 `--cells`（二维数组，可逐格带 cell_styles/border_styles）。整批展开为**单次原子批量提交**，支持跨 sheet；典型场景：批量修复散布多处的公式、跨表同构写入——不要为此拼 +batch-update 的 --operations。与 `--range`+`--cells` 二选一；范围级统一样式不在此做，写完接 +styles-put |
 | `--allow-overwrite` | bool | optional | 允许覆盖非空 cell（默认 true）；设为 false 时遇非空 cell 报错 |
 | `--max-cells` | int | optional | 防爆，默认 50000（隐藏 flag：不在 `--help` 列出，但可正常传入） |
 | `--copy-to-range` | string | optional | 复制范围（A1 表示法）：把 --range 中 --cells 写入的内容（值/公式/样式，取决于实际传入字段）复制到该区域，公式引用自动平移（如 C2=B2 → C3=B3）。适合先写一行/一块模板再扩展填充整列/整区域（如 --range A1:G1 写模板、--copy-to-range A1:G100 填充 100 行）。支持整行 3:6、整列 C:E、到列尾 D3:D、到行尾 D3:3；支持英文逗号分隔多个目标区域，如 C1:D2,E5:F6 |
@@ -346,6 +360,16 @@ _【维度】行列数必须与 range 完全一致：'A1:C2'→[[_,_,_],[_,_,_]]
 - `multiple_values` (array<object>?) — 多值内容，用于支持多选的列表验证单元格 each: { value: oneOf, format?: string }
 - `data_validation` (object?) — 数据验证配置 { type: enum, items?: array<string>, range?: string, operator?: enum, values?: array<oneOf>, …共 9 项 }
 
+### `+cells-set` `--writes`
+
+_多区域写入项数组（最多 100 项），整批单次原子提交；支持跨 sheet_
+
+**数组项**（类型 object）：
+- `sheet_id` (string?) — 目标子表 reference_id；与 sheet_name 二选一，必须写在每一项里（不认顶层 sheet 定位）
+- `sheet_name` (string?) — 目标子表名；与 sheet_id 二选一，必须写在每一项里
+- `range` (string) — A1 矩形范围，行列维度必须与 cells 严格一致（同 --range）
+- `cells` (array) — 二维单元格数组，结构同 --cells（value / formula / cell_styles / border_styles 等，见 set_cell_…
+
 ### `+cells-set-style` `--border-styles`
 
 _单元格边框配置，含 top/bottom/left/right 四个方向，每个方向的结构相同（见 top）_
@@ -383,10 +407,11 @@ _一个或多个子表的 typed 数据，每个数组元素写入一张子表；
 
 **数组项**（类型 object）：
 - `cell_merges` (array<object>?) — 单元格合并操作数组；range 使用 A1 单元格范围，merge_type 默认 all each: { merge_type?: enum, range: string }
-- `cell_styles` (array<object>?) — 单元格样式操作数组；每项用 A1 单元格 range 指定范围，字段名与 +cells-set-style 对齐 each: { background_color?: string, border_styles?: object, font_color?: string, font_family?: string, font_line?: enum, …共 13 项 }
-- `col_sizes` (array<object>?) — 列宽操作数组；range 使用列范围如 A:C，type 为 pixel/standard，pixel 需要 size each: { range: string, size?: number, type: enum }
+- `cell_styles` (array<object>?) — 单元格样式操作数组；每项用 A1 单元格 range 指定范围，字段名与 +cells-set-style 对齐 each: { background_color?: string, border?: object, border_styles?: object, font_color?: string, font_family?: string, …共 14 项 }
+- `col_sizes` (array<object>?) — 列宽操作数组；range 使用列范围如 A:C，给 size（px）即像素列宽（type 可省略）；type 为 standard 时不带 size each: { range: string, size?: number, type?: enum }
+- `freeze` (object?) — 冻结行列：rows = 冻结前 N 行，cols = 冻结前 N 列（0 或省略 = 该维度不冻结） { cols?: integer, rows?: integer }
 - `name` (string) — 子表名
-- `row_sizes` (array<object>?) — 行高操作数组；range 使用行范围如 1:3，type 为 pixel/standard/auto，pixel 需要 size each: { range: string, size?: number, type: enum }
+- `row_sizes` (array<object>?) — 行高操作数组；range 使用行范围如 1:3，给 size（px）即像素行高（type 可省略）；type 为 standard/auto 时不带 size each: { range: string, size?: number, type?: enum }
 
 ## Examples
 
@@ -401,7 +426,7 @@ _一个或多个子表的 typed 数据，每个数组元素写入一张子表；
 | 只改**已有 cell 的样式**，不动 value/formula | `+cells-set-style` | `+cells-set`（会触发不必要的值写入） |
 | 把**单张图片嵌入**到某个 cell | `+cells-set-image` | `+cells-set`（参数更繁琐） |
 | **插行/列 + 写入** 这种多步组合，且要原子 | `+batch-update`（见 lark-sheets-batch-update） | 多次独立 `+cells-set`（非原子；插入会扰动后续 range） |
-| 在**多个不连续 range** 上应用同一组样式 | `+cells-batch-set-style`（见 lark-sheets-batch-update） | 多次 `+cells-set-style`（非原子） |
+| 在**多个不连续 range** 上应用同一组样式 | `+styles-put`（cell_styles 多项即多区域，见 lark-sheets-styles-put） | 多次 `+cells-set-style`（非原子） |
 
 ### `+cells-set`
 
@@ -511,6 +536,8 @@ lark-cli sheets +csv-put --spreadsheet-token shtXXX --sheet-id "$SID" \
 python export.py | lark-cli sheets +table-put --url "<表URL>" --sheets -
 # 某 sheet 带 "mode":"append" 追加到已有数据末尾、默认不重复表头
 lark-cli sheets +table-put --spreadsheet-token "<token>" --sheets @payload.json
+# --sheets 与 --styles 都是大 JSON 时：stdin 每次调用只能给一个 flag，一个走 -、另一个走 @cwd 相对路径
+lark-cli sheets +table-put --url "<表URL>" --sheets - --styles @styles.json < sheets.json
 ```
 
 每个 sheet 还可带 `"allow_overwrite": false`（遇非空拒写、保护原数据）、`"header": false`（只写数据不写表头）。完整字段跑 `+table-put --print-schema --flag-name sheets`。

skills/lark-slides/references/slides_xml_schema_definition.xml

@@ -1464,6 +1464,7 @@
         <xs:annotation>
             <xs:documentation>
                 表格元素, 用于展示结构化数据
+                宽高分配规则参考 HTML table 的整体值 / 子项值处理逻辑, 但在 SXSD 中做确定化约束, 以保证跨端实现一致。
                 边框规则：
                 - 后设置优先：相邻单元格线条只有一个颜色, 如果均设置, 则右下单元格的设置覆盖左上单元格
                 - 不设置边框属性时使用默认样式
@@ -1476,6 +1477,8 @@
                 - id: 表格唯一标识符(可选)
                 - topLeftX/topLeftY: 左上角坐标
                 - flipX/flipY: 水平/垂直翻转
+                - width: 表格目标总宽度(可选)。若设置, 优先用于为未填写列宽的列分配剩余宽度; 当所有列宽均为空时, 所有列均分该值; 当所有列均已填写且该值大于已填写列宽总和时, 多出空间继续分配给现有列, 默认按各列当前宽度作为权重分配; 当已填写列宽之和超过或无法容纳该值时, 保留已填写列宽, 并以最终列宽总和回写 table.width
+                - height: 表格目标总高度(可选)。若设置, 优先用于为未填写行高的行分配剩余高度; 当所有行高均为空时, 所有行均分该值; 当所有行均已填写且该值大于已填写行高总和时, 多出空间继续分配给现有行, 默认按各行当前高度作为权重分配; 当已填写行高之和超过或无法容纳该值时, 保留已填写行高, 并以最终行高总和回写 table.height
                 table 子元素:
                 - colgroup: 列组元素, 用于定义列的宽度
                 - tr: 行元素, 包含多个单元格
@@ -1484,10 +1487,10 @@
                 - col: 列元素
                 col 属性:
                 - span: 列跨数, 默认为1, 可选
-                - width: 列宽度, 默认值为110, 可选
+                - width: 列宽度输入值, 默认值为110, 可选。无 table.width 时, 已填写列保持原值, 空列使用默认值; 有 table.width 时, 已填写列保持原值, 空列优先均分剩余宽度; 若不存在空列且 table.width 大于已填写列宽总和, 则多出空间按各列当前宽度作为权重分配到所有列; 若剩余宽度不足则空列回退为默认值, 并以最终列宽总和作为 table.width
 
                 tr 属性:
-                - height: 行高, 默认为单元格高度
+                - height: 行高输入值, 默认值为37, 可选。无 table.height 时, 已填写行保持原值, 空行使用默认值; 有 table.height 时, 已填写行保持原值, 空行优先均分剩余高度; 若不存在空行且 table.height 大于已填写行高总和, 则多出空间按各行当前高度作为权重分配到所有行; 若剩余高度不足则空行回退为默认值, 并以最终行高总和作为 table.height。若行高低于内容高度, 需要手动修改行高
                 tr 子元素:
                 - td: 单元格元素, 用于显示数据
 
@@ -1543,6 +1546,8 @@
             <xs:attribute name="topLeftY" type="sml:YType" use="required"/>
             <xs:attribute name="flipX" type="xs:boolean" use="optional" default="false"/>
             <xs:attribute name="flipY" type="xs:boolean" use="optional" default="false"/>
+            <xs:attribute name="width" type="sml:PositiveSize" use="optional"/>
+            <xs:attribute name="height" type="sml:PositiveSize" use="optional"/>
         </xs:complexType>
     </xs:element>
 

skills/lark-slides/references/xml-format-guide.md

@@ -248,6 +248,21 @@
 - `<tr>` 内为 `<td>`
 - `<td>` 内可放 `<content>`
 
+`<table>` 可选设置 `width` 和 `height`，分别表示表格的目标总宽度和总高度：
+
+```xml
+<table topLeftX="80" topLeftY="120" width="800" height="300">
+  <colgroup>
+    <col width="240"/>
+    <col/>
+  </colgroup>
+  <tr height="80">
+    <td><content textType="body"><p>表头 1</p></content></td>
+    <td><content textType="body"><p>表头 2</p></content></td>
+  </tr>
+</table>
+```
+
 ### `<chart>`
 
 图表元素必须至少包含：

skills/lark-slides/scripts/xml_text_overlap_lint.py

@@ -38,6 +38,8 @@ SXSD_ATTR_ALIASES = {
     "fontColor": "color",
 }
 SERVER_FILLED_SXSD_ATTRS = {"id"}
+DEFAULT_TABLE_COLUMN_WIDTH = 110
+DEFAULT_TABLE_ROW_HEIGHT = 37
 _SXSD_TAG_ATTRIBUTES_CACHE: dict[str, set[str]] | None = None
 _ICONPARK_ICON_TYPES_CACHE: set[str] | None = None
 
@@ -88,6 +90,84 @@ def extract_numeric_attribute(tag_source: str, name: str) -> int | float | None:
     return int(value) if value.is_integer() else value
 
 
+def sum_sizes(sizes: list[int | float]) -> int | float:
+    return sum(sizes)
+
+
+def is_filled_size(size: int | float | None) -> bool:
+    return isinstance(size, (int, float)) and math.isfinite(size) and size > 0
+
+
+def fill_last_size_gap(sizes: list[int | float], target_size: int | float) -> list[int | float]:
+    if not sizes:
+        return sizes
+    final_sizes = [
+        size if index == len(sizes) - 1 else max(1, math.floor(size + 0.5))
+        for index, size in enumerate(sizes)
+    ]
+    remaining_size = target_size - sum_sizes(final_sizes[:-1])
+    if remaining_size >= 1:
+        final_sizes[-1] = remaining_size
+        return final_sizes
+
+    size_to_redistribute = 1 - remaining_size
+    for index in range(len(final_sizes) - 2, -1, -1):
+        reduction = min(final_sizes[index] - 1, size_to_redistribute)
+        final_sizes[index] -= reduction
+        size_to_redistribute -= reduction
+        if size_to_redistribute == 0:
+            final_sizes[-1] = 1
+            return final_sizes
+
+    final_sizes[-1] = 1
+    return final_sizes
+
+
+def solve_weighted_min_layout(
+    input_sizes: list[int | float | None], default_size: int | float, target_min_size: int | float | None
+) -> dict[str, Any]:
+    filled_indexes: list[int] = []
+    empty_indexes: list[int] = []
+    base_sizes: list[int | float] = []
+    for index, size in enumerate(input_sizes):
+        if is_filled_size(size):
+            filled_indexes.append(index)
+            base_sizes.append(size)
+        else:
+            empty_indexes.append(index)
+            base_sizes.append(0)
+    filled_sum = sum_sizes(base_sizes)
+
+    if target_min_size is None:
+        final_sizes = [default_size if index in empty_indexes else size for index, size in enumerate(base_sizes)]
+        return {"final_sizes": final_sizes, "actual_size": sum_sizes(final_sizes), "ratio": 1}
+
+    if not filled_indexes:
+        average_size = target_min_size / len(input_sizes)
+        final_sizes = fill_last_size_gap([average_size] * len(input_sizes), target_min_size)
+        return {"final_sizes": final_sizes, "actual_size": sum_sizes(final_sizes), "ratio": 1}
+
+    if empty_indexes:
+        remaining_size = target_min_size - filled_sum
+        final_sizes = [*base_sizes]
+        if remaining_size > 0:
+            average_size = remaining_size / len(empty_indexes)
+            empty_sizes = fill_last_size_gap([average_size] * len(empty_indexes), remaining_size)
+            for index, empty_size in zip(empty_indexes, empty_sizes):
+                final_sizes[index] = empty_size
+        else:
+            for index in empty_indexes:
+                final_sizes[index] = default_size
+        return {"final_sizes": final_sizes, "actual_size": sum_sizes(final_sizes), "ratio": 1}
+
+    ratio = max(1, target_min_size / filled_sum)
+    actual_size = max(target_min_size, filled_sum)
+    if ratio == 1:
+        return {"final_sizes": [*base_sizes], "actual_size": actual_size, "ratio": ratio}
+    final_sizes = fill_last_size_gap([size * ratio for size in base_sizes], actual_size)
+    return {"final_sizes": final_sizes, "actual_size": sum_sizes(final_sizes), "ratio": ratio}
+
+
 def strip_xml(value: str) -> str:
     stripped = re.sub(r"<!\[CDATA\[([\s\S]*?)\]\]>", r"\1", value)
     stripped = re.sub(r"<[^>]+>", " ", stripped)
@@ -515,8 +595,8 @@ def extract_elements(slide_xml: str) -> list[dict[str, Any]]:
     for match in re.finditer(r"<(shape|img|table|chart|whiteboard)\b([^>]*)>", slide_xml):
         kind, attrs = match.group(1), match.group(2)
         content = ""
-        if kind == "shape":
-            close_index = slide_xml.find("</shape>", match.end())
+        if kind in {"shape", "table"}:
+            close_index = slide_xml.find(f"</{kind}>", match.end())
             if close_index != -1:
                 content = slide_xml[match.end() : close_index]
 
@@ -525,6 +605,15 @@ def extract_elements(slide_xml: str) -> list[dict[str, Any]]:
         y = extract_numeric_attribute(attrs, "topLeftY")
         width = extract_numeric_attribute(attrs, "width")
         height = extract_numeric_attribute(attrs, "height")
+        rotation = extract_numeric_attribute(attrs, "rotation") or 0
+        table_layouts: dict[str, dict[str, Any] | None] = {}
+        if kind == "table":
+            width, table_layouts["width"] = resolve_table_dimension(
+                content, width, extract_table_column_sizes, DEFAULT_TABLE_COLUMN_WIDTH
+            )
+            height, table_layouts["height"] = resolve_table_dimension(
+                content, height, extract_table_row_sizes, DEFAULT_TABLE_ROW_HEIGHT
+            )
         if all(value is not None for value in [x, y, width, height]):
             element = {
                 "id": element_id,
@@ -534,8 +623,17 @@ def extract_elements(slide_xml: str) -> list[dict[str, Any]]:
                 "y": y,
                 "width": width,
                 "height": height,
+                "rotation": rotation,
                 "order": len(elements),
             }
+            if kind == "table":
+                element.update(
+                    {
+                        "declared_width": extract_numeric_attribute(attrs, "width"),
+                        "declared_height": extract_numeric_attribute(attrs, "height"),
+                        "table_layouts": table_layouts,
+                    }
+                )
             if kind == "shape":
                 element.update(
                     {
@@ -867,11 +965,158 @@ def detect_whiteboard_external_overlaps(
     return issues
 
 
+def element_canvas_bbox(element: dict[str, Any]) -> dict[str, int | float]:
+    bbox = {key: element[key] for key in ("x", "y", "width", "height")}
+    if element["kind"] != "chart" and not (element["kind"] == "shape" and element["type"] == "text"):
+        return bbox
+
+    rotation = element["rotation"]
+    if not isinstance(rotation, (int, float)) or not math.isfinite(rotation):
+        rotation = 0
+    rotation %= 360
+    if math.isclose(rotation, 0, abs_tol=1e-9):
+        return bbox
+    radians = math.radians(rotation)
+    sine = abs(math.sin(radians))
+    cosine = abs(math.cos(radians))
+    sine = 0 if math.isclose(sine, 0, abs_tol=1e-12) else sine
+    cosine = 0 if math.isclose(cosine, 0, abs_tol=1e-12) else cosine
+    rotated_width = element["width"] * cosine + element["height"] * sine
+    rotated_height = element["width"] * sine + element["height"] * cosine
+    return {
+        "x": element["x"] - (rotated_width - element["width"]) / 2,
+        "y": element["y"] - (rotated_height - element["height"]) / 2,
+        "width": rotated_width,
+        "height": rotated_height,
+    }
+
+
+def detect_elements_out_of_canvas(
+    elements: list[dict[str, Any]], slide_width: int | float, slide_height: int | float
+) -> list[dict[str, Any]]:
+    issues: list[dict[str, Any]] = []
+    for element in (
+        element
+        for element in elements
+        if element["kind"] in {"table", "chart"}
+        or (element["kind"] == "shape" and element["type"] == "text")
+    ):
+        bbox = element_canvas_bbox(element)
+        overflow = {
+            "left": max(-bbox["x"], 0),
+            "top": max(-bbox["y"], 0),
+            "right": max(bbox["x"] + bbox["width"] - slide_width, 0),
+            "bottom": max(bbox["y"] + bbox["height"] - slide_height, 0),
+        }
+        overflow_details = [
+            f"{side} by {amount:g}px" for side, amount in overflow.items() if amount > 0
+        ]
+        if not overflow_details:
+            continue
+        issues.append(
+            {
+                "level": "error",
+                "code": f'{element["kind"]}_out_of_canvas',
+                "elements": [element["id"]],
+                "canvas": {"width": slide_width, "height": slide_height},
+                "bbox": bbox,
+                "overflow": overflow,
+                "message": (
+                    f'{element["kind"]} {element["id"]} exceeds the {slide_width:g}x{slide_height:g} canvas '
+                    f'({", ".join(overflow_details)})'
+                ),
+                "hint": (
+                    "Move the table inside the canvas, reduce table.width/table.height, or split the table across "
+                    "slides."
+                    if element["kind"] == "table"
+                    else f'Move the {element["kind"]} inside the canvas or reduce its width/height.'
+                ),
+            }
+        )
+    return issues
+
+
+def extract_table_column_sizes(table_xml: str) -> list[int | float | None]:
+    sizes: list[int | float | None] = []
+    for match in re.finditer(r"<col\b([^>]*)/?>", table_xml):
+        attrs = match.group(1)
+        span = extract_numeric_attribute(attrs, "span") or 1
+        span_count = int(span) if math.isfinite(span) and span > 0 and float(span).is_integer() else 1
+        sizes.extend([extract_numeric_attribute(attrs, "width")] * span_count)
+    return sizes
+
+
+def extract_table_row_sizes(table_xml: str) -> list[int | float | None]:
+    return [extract_numeric_attribute(match.group(1), "height") for match in re.finditer(r"<tr\b([^>]*)>", table_xml)]
+
+
+def resolve_table_dimension(
+    table_xml: str,
+    declared_size: int | float | None,
+    extract_sizes: Any,
+    default_size: int | float,
+) -> tuple[int | float | None, dict[str, Any] | None]:
+    input_sizes = extract_sizes(table_xml)
+    if not input_sizes:
+        return declared_size, None
+    layout = solve_weighted_min_layout(
+        input_sizes, default_size, declared_size if is_filled_size(declared_size) else None
+    )
+    return layout["actual_size"], layout
+
+
+def format_size(size: int | float) -> str:
+    return f"{size:g}"
+
+
+def detect_table_layout_size_mismatches(elements: list[dict[str, Any]]) -> list[dict[str, Any]]:
+    issues: list[dict[str, Any]] = []
+    dimensions = {
+        "width": ("col", "column widths"),
+        "height": ("tr", "row heights"),
+    }
+    for table in (element for element in elements if element["kind"] == "table"):
+        for dimension, (child_tag, child_description) in dimensions.items():
+            target_size = table[f"declared_{dimension}"]
+            if not is_filled_size(target_size):
+                continue
+            layout = table["table_layouts"][dimension]
+            if layout is None:
+                continue
+            actual_size = layout["actual_size"]
+            if math.isclose(actual_size, target_size, rel_tol=1e-9, abs_tol=1e-9):
+                continue
+            issues.append(
+                {
+                    "level": "info",
+                    "code": "table_resolved_size_mismatch",
+                    "elements": [table["id"]],
+                    "dimension": dimension,
+                    "declared_size": target_size,
+                    "resolved_size": actual_size,
+                    "resolved_sizes": layout["final_sizes"],
+                    "message": (
+                        f'table {table["id"]} declares {dimension}={format_size(target_size)}px, but its '
+                        f"{child_description} resolve to {format_size(actual_size)}px"
+                    ),
+                    "hint": (
+                        f"Set table.{dimension} to {format_size(actual_size)}px, or adjust <{child_tag}> sizes "
+                        f"so their resolved total matches {format_size(target_size)}px."
+                    ),
+                }
+            )
+    return issues
+
+
 def lint_slide(
     slide_xml: str, slide_number: int, slide_width: int | float = 960, slide_height: int | float = 540
 ) -> dict[str, Any]:
     elements = extract_elements(slide_xml)
-    issues: list[dict[str, Any]] = detect_whiteboard_external_overlaps(elements, slide_width, slide_height)
+    issues: list[dict[str, Any]] = [
+        *detect_whiteboard_external_overlaps(elements, slide_width, slide_height),
+        *detect_elements_out_of_canvas(elements, slide_width, slide_height),
+        *detect_table_layout_size_mismatches(elements),
+    ]
 
     for index, left in enumerate(elements):
         for right in elements[index + 1 :]:
@@ -896,7 +1141,7 @@ def lint_xml(xml: str, source_path: str | None = None) -> dict[str, Any]:
         return {
             "file": source_path,
             "slide_size": {"width": 960, "height": 540},
-            "summary": {"slide_count": 0, "error_count": 1, "warning_count": 0},
+            "summary": {"slide_count": 0, "error_count": 1, "warning_count": 0, "info_count": 0},
             "issues": [xml_error],
             "slides": [],
         }
@@ -908,10 +1153,16 @@ def lint_xml(xml: str, source_path: str | None = None) -> dict[str, Any]:
     if namespace_issues:
         error_count = sum(1 for issue in top_level_issues if issue["level"] == "error")
         warning_count = sum(1 for issue in top_level_issues if issue["level"] == "warning")
+        info_count = sum(1 for issue in top_level_issues if issue["level"] == "info")
         return {
             "file": source_path,
             "slide_size": {"width": 960, "height": 540},
-            "summary": {"slide_count": 0, "error_count": error_count, "warning_count": warning_count},
+            "summary": {
+                "slide_count": 0,
+                "error_count": error_count,
+                "warning_count": warning_count,
+                "info_count": info_count,
+            },
             "issues": top_level_issues,
             "slides": [],
         }
@@ -924,10 +1175,17 @@ def lint_xml(xml: str, source_path: str | None = None) -> dict[str, Any]:
     error_count += sum(1 for slide in slides for issue in slide["issues"] if issue["level"] == "error")
     warning_count = sum(1 for issue in top_level_issues if issue["level"] == "warning")
     warning_count += sum(1 for slide in slides for issue in slide["issues"] if issue["level"] == "warning")
+    info_count = sum(1 for issue in top_level_issues if issue["level"] == "info")
+    info_count += sum(1 for slide in slides for issue in slide["issues"] if issue["level"] == "info")
     result = {
         "file": source_path,
         "slide_size": {"width": presentation["width"], "height": presentation["height"]},
-        "summary": {"slide_count": len(slides), "error_count": error_count, "warning_count": warning_count},
+        "summary": {
+            "slide_count": len(slides),
+            "error_count": error_count,
+            "warning_count": warning_count,
+            "info_count": info_count,
+        },
         "slides": slides,
     }
     if top_level_issues:

skills/lark-slides/scripts/xml_text_overlap_lint_test.py

@@ -2,7 +2,12 @@
 # SPDX-License-Identifier: MIT
 from __future__ import annotations
 
+import json
+import subprocess
+import sys
+import tempfile
 import unittest
+from pathlib import Path
 
 import xml_text_overlap_lint
 
@@ -212,7 +217,8 @@ class XmlTextOverlapLintTest(unittest.TestCase):
         )
         self.assertEqual(result["slide_size"], {"width": 960, "height": 540})
         self.assertEqual(result["summary"]["slide_count"], 1)
-        self.assertEqual(result["summary"]["error_count"], 0)
+        self.assertEqual(result["summary"]["error_count"], 1)
+        self.assertEqual(result["slides"][0]["issues"][0]["code"], "shape_out_of_canvas")
 
     def test_lint_xml_preserves_presentation_canvas_and_slide_order(self) -> None:
         result = xml_text_overlap_lint.lint_xml(
@@ -596,7 +602,7 @@ class XmlTextOverlapLintTest(unittest.TestCase):
         self.assertEqual(result["slides"][0]["issues"][0]["code"], "bbox_overlap")
         self.assertEqual(result["slides"][0]["issues"][0]["elements"], ["source", "target"])
 
-    def test_lint_xml_does_not_check_bounds_or_text_height(self) -> None:
+    def test_lint_xml_reports_text_out_of_canvas_but_not_text_height(self) -> None:
         result = xml_text_overlap_lint.lint_xml(
             """
             <presentation xmlns="http://www.larkoffice.com/sml/2.0" width="960" height="540">
@@ -613,8 +619,11 @@ class XmlTextOverlapLintTest(unittest.TestCase):
             </presentation>
             """
         )
-        self.assertEqual(result["summary"]["error_count"], 0)
+        issue = result["slides"][0]["issues"][0]
+        self.assertEqual(result["summary"]["error_count"], 1)
         self.assertEqual(result["summary"]["warning_count"], 0)
+        self.assertEqual(issue["code"], "shape_out_of_canvas")
+        self.assertEqual(issue["overflow"], {"left": 0, "top": 0, "right": 160, "bottom": 40})
 
     def test_lint_xml_allows_template_style_bleed_and_text_over_images(self) -> None:
         result = xml_text_overlap_lint.lint_xml(
@@ -669,7 +678,7 @@ class XmlTextOverlapLintTest(unittest.TestCase):
         self.assertEqual(elements[1]["fontSize"], 28)
         self.assertEqual(elements[1]["text"], "Growth & scale\nFocused execution")
 
-    def test_lint_xml_does_not_check_small_out_of_bounds_elements(self) -> None:
+    def test_lint_xml_allows_small_out_of_bounds_images(self) -> None:
         result = xml_text_overlap_lint.lint_xml(
             """
             <presentation xmlns="http://www.larkoffice.com/sml/2.0" width="960" height="540">
@@ -683,7 +692,7 @@ class XmlTextOverlapLintTest(unittest.TestCase):
         )
         self.assertEqual(result["summary"]["error_count"], 0)
 
-    def test_lint_xml_ignores_obviously_misplaced_large_visuals(self) -> None:
+    def test_lint_xml_allows_out_of_canvas_images(self) -> None:
         result = xml_text_overlap_lint.lint_xml(
             """
             <presentation xmlns="http://www.larkoffice.com/sml/2.0" width="960" height="540">
@@ -698,7 +707,7 @@ class XmlTextOverlapLintTest(unittest.TestCase):
         )
         self.assertEqual(result["summary"]["error_count"], 0)
 
-    def test_lint_xml_allows_reasonable_large_visual_bleed(self) -> None:
+    def test_lint_xml_allows_full_bleed_images(self) -> None:
         result = xml_text_overlap_lint.lint_xml(
             """
             <presentation xmlns="http://www.larkoffice.com/sml/2.0" width="960" height="540">
@@ -712,6 +721,339 @@ class XmlTextOverlapLintTest(unittest.TestCase):
         )
         self.assertEqual(result["summary"]["error_count"], 0)
 
+    def test_lint_xml_reports_text_and_chart_out_of_canvas(self) -> None:
+        result = xml_text_overlap_lint.lint_xml(
+            """
+            <presentation xmlns="http://www.larkoffice.com/sml/2.0" width="960" height="540">
+              <slide xmlns="http://www.larkoffice.com/sml/2.0">
+                <data>
+                  <shape id="outside-shape" type="text" topLeftX="-10" topLeftY="40" width="50" height="50"/>
+                  <img id="outside-img" src="token" topLeftX="120" topLeftY="-20" width="50" height="50"/>
+                  <chart id="outside-chart" topLeftX="900" topLeftY="100" width="100" height="100"/>
+                  <whiteboard id="outside-whiteboard" topLeftX="100" topLeftY="500" width="100" height="100"/>
+                </data>
+              </slide>
+            </presentation>
+            """
+        )
+        issues = result["slides"][0]["issues"]
+        self.assertEqual(result["summary"]["error_count"], 2)
+        self.assertEqual(
+            [(issue["code"], issue["elements"], issue["overflow"]) for issue in issues],
+            [
+                ("shape_out_of_canvas", ["outside-shape"], {"left": 10, "top": 0, "right": 0, "bottom": 0}),
+                ("chart_out_of_canvas", ["outside-chart"], {"left": 0, "top": 0, "right": 40, "bottom": 0}),
+            ],
+        )
+
+    def test_lint_xml_uses_rotated_text_and_chart_bounds_for_canvas_validation(self) -> None:
+        result = xml_text_overlap_lint.lint_xml(
+            """
+            <presentation xmlns="http://www.larkoffice.com/sml/2.0" width="960" height="540">
+              <slide xmlns="http://www.larkoffice.com/sml/2.0">
+                <data>
+                  <shape id="rotated-text" type="text" topLeftX="0" topLeftY="0" width="100" height="100" rotation="45"/>
+                  <chart id="rotated-chart" topLeftX="860" topLeftY="200" width="100" height="100" rotation="45"/>
+                </data>
+              </slide>
+            </presentation>
+            """
+        )
+        issues_by_element = {issue["elements"][0]: issue for issue in result["slides"][0]["issues"]}
+        self.assertEqual(result["summary"]["error_count"], 2)
+        self.assertEqual(issues_by_element["rotated-text"]["code"], "shape_out_of_canvas")
+        self.assertAlmostEqual(issues_by_element["rotated-text"]["overflow"]["left"], 20.710678, places=5)
+        self.assertAlmostEqual(issues_by_element["rotated-text"]["overflow"]["top"], 20.710678, places=5)
+        self.assertEqual(issues_by_element["rotated-chart"]["code"], "chart_out_of_canvas")
+        self.assertAlmostEqual(issues_by_element["rotated-chart"]["overflow"]["right"], 20.710678, places=5)
+
+    def test_lint_xml_treats_non_finite_rotations_as_zero(self) -> None:
+        result = xml_text_overlap_lint.lint_xml(
+            """
+            <presentation xmlns="http://www.larkoffice.com/sml/2.0" width="960" height="540">
+              <slide xmlns="http://www.larkoffice.com/sml/2.0">
+                <data>
+                  <shape id="infinite" type="text" topLeftX="-10" topLeftY="0" width="20" height="20" rotation="inf"/>
+                  <shape id="negative-infinite" type="text" topLeftX="0" topLeftY="-10" width="20" height="20" rotation="-inf"/>
+                  <chart id="not-a-number" topLeftX="950" topLeftY="0" width="20" height="20" rotation="nan"/>
+                </data>
+              </slide>
+            </presentation>
+            """
+        )
+        issues_by_element = {issue["elements"][0]: issue for issue in result["slides"][0]["issues"]}
+        self.assertEqual(result["summary"]["error_count"], 3)
+        self.assertEqual(issues_by_element["infinite"]["overflow"], {"left": 10, "top": 0, "right": 0, "bottom": 0})
+        self.assertEqual(issues_by_element["negative-infinite"]["overflow"], {"left": 0, "top": 10, "right": 0, "bottom": 0})
+        self.assertEqual(issues_by_element["not-a-number"]["overflow"], {"left": 0, "top": 0, "right": 10, "bottom": 0})
+
+    def test_lint_xml_reports_table_bottom_overflow_from_declared_bounds(self) -> None:
+        result = xml_text_overlap_lint.lint_xml(
+            """
+            <presentation xmlns="http://www.larkoffice.com/sml/2.0" width="960" height="540">
+              <slide xmlns="http://www.larkoffice.com/sml/2.0">
+                <data>
+                  <table id="score-table" topLeftX="54" topLeftY="238" width="414" height="385">
+                    <tr><td><content><p>Score</p></content></td></tr>
+                  </table>
+                </data>
+              </slide>
+            </presentation>
+            """
+        )
+        issue = result["slides"][0]["issues"][0]
+        self.assertEqual(result["summary"]["error_count"], 1)
+        self.assertEqual(issue["code"], "table_out_of_canvas")
+        self.assertEqual(issue["elements"], ["score-table"])
+        self.assertEqual(issue["overflow"], {"left": 0, "top": 0, "right": 0, "bottom": 83})
+        self.assertEqual(issue["bbox"], {"x": 54, "y": 238, "width": 414, "height": 385})
+
+    def test_lint_xml_reports_table_right_overflow_from_declared_bounds(self) -> None:
+        result = xml_text_overlap_lint.lint_xml(
+            """
+            <presentation xmlns="http://www.larkoffice.com/sml/2.0" width="960" height="540">
+              <slide xmlns="http://www.larkoffice.com/sml/2.0">
+                <data>
+                  <table id="wide-table" topLeftX="850" topLeftY="80" width="180" height="120">
+                    <tr><td><content><p>Score</p></content></td></tr>
+                  </table>
+                </data>
+              </slide>
+            </presentation>
+            """
+        )
+        issue = result["slides"][0]["issues"][0]
+        self.assertEqual(result["summary"]["error_count"], 1)
+        self.assertEqual(issue["code"], "table_out_of_canvas")
+        self.assertEqual(issue["overflow"], {"left": 0, "top": 0, "right": 70, "bottom": 0})
+
+    def test_lint_xml_allows_table_with_declared_bounds_inside_canvas(self) -> None:
+        result = xml_text_overlap_lint.lint_xml(
+            """
+            <presentation xmlns="http://www.larkoffice.com/sml/2.0" width="960" height="540">
+              <slide xmlns="http://www.larkoffice.com/sml/2.0">
+                <data>
+                  <table id="inside-table" topLeftX="40" topLeftY="120" width="880" height="360">
+                    <tr><td><content><p>Score</p></content></td></tr>
+                  </table>
+                </data>
+              </slide>
+            </presentation>
+            """
+        )
+        self.assertEqual(result["summary"]["error_count"], 0)
+        self.assertEqual(result["summary"]["warning_count"], 0)
+
+    def test_lint_xml_reports_resolved_table_bounds_when_declared_sizes_are_missing(self) -> None:
+        result = xml_text_overlap_lint.lint_xml(
+            """
+            <presentation xmlns="http://www.larkoffice.com/sml/2.0" width="960" height="540">
+              <slide xmlns="http://www.larkoffice.com/sml/2.0">
+                <data>
+                  <table id="implicit-size-table" topLeftX="850" topLeftY="480">
+                    <colgroup><col/><col/></colgroup>
+                    <tr><td/><td/></tr>
+                    <tr><td/><td/></tr>
+                  </table>
+                </data>
+              </slide>
+            </presentation>
+            """
+        )
+        issue = result["slides"][0]["issues"][0]
+        self.assertEqual(result["summary"]["error_count"], 1)
+        self.assertEqual(issue["code"], "table_out_of_canvas")
+        self.assertEqual(issue["bbox"], {"x": 850, "y": 480, "width": 220, "height": 74})
+        self.assertEqual(issue["overflow"], {"left": 0, "top": 0, "right": 110, "bottom": 14})
+
+    def test_lint_xml_uses_resolved_table_bounds_for_canvas_validation(self) -> None:
+        result = xml_text_overlap_lint.lint_xml(
+            """
+            <presentation xmlns="http://www.larkoffice.com/sml/2.0" width="960" height="540">
+              <slide xmlns="http://www.larkoffice.com/sml/2.0">
+                <data>
+                  <table id="resolved-overflow-table" topLeftX="800" topLeftY="80" width="100" height="40">
+                    <colgroup><col width="100"/><col width="100"/></colgroup>
+                    <tr height="40"><td/><td/></tr>
+                  </table>
+                </data>
+              </slide>
+            </presentation>
+            """
+        )
+        issues = result["slides"][0]["issues"]
+        canvas_issue = next(issue for issue in issues if issue["code"] == "table_out_of_canvas")
+        mismatch_issue = next(issue for issue in issues if issue["code"] == "table_resolved_size_mismatch")
+        self.assertEqual(result["summary"]["error_count"], 1)
+        self.assertEqual(canvas_issue["bbox"], {"x": 800, "y": 80, "width": 200, "height": 40})
+        self.assertEqual(canvas_issue["overflow"]["right"], 40)
+        self.assertEqual(mismatch_issue["dimension"], "width")
+        self.assertEqual(mismatch_issue["resolved_size"], canvas_issue["bbox"]["width"])
+
+    def test_lint_xml_uses_the_same_anonymous_table_id_for_all_table_diagnostics(self) -> None:
+        result = xml_text_overlap_lint.lint_xml(
+            """
+            <presentation xmlns="http://www.larkoffice.com/sml/2.0" width="960" height="540">
+              <slide xmlns="http://www.larkoffice.com/sml/2.0">
+                <data>
+                  <shape id="title" type="text" topLeftX="40" topLeftY="40" width="200" height="40"/>
+                  <img id="logo" src="token" topLeftX="40" topLeftY="100" width="40" height="40"/>
+                  <table topLeftX="900" topLeftY="80" width="100" height="40">
+                    <colgroup><col width="100"/><col width="100"/></colgroup>
+                    <tr height="40"><td/><td/></tr>
+                  </table>
+                </data>
+              </slide>
+            </presentation>
+            """
+        )
+        issues = result["slides"][0]["issues"]
+        canvas_issue = next(issue for issue in issues if issue["code"] == "table_out_of_canvas")
+        mismatch_issue = next(issue for issue in issues if issue["code"] == "table_resolved_size_mismatch")
+        self.assertEqual(canvas_issue["elements"], ["table-3"])
+        self.assertEqual(mismatch_issue["elements"], ["table-3"])
+
+    def test_lint_xml_reports_info_when_table_target_size_resolves_larger_than_declared(self) -> None:
+        result = xml_text_overlap_lint.lint_xml(
+            """
+            <presentation xmlns="http://www.larkoffice.com/sml/2.0" width="960" height="540">
+              <slide xmlns="http://www.larkoffice.com/sml/2.0">
+                <data>
+                  <table id="size-mismatch" topLeftX="40" topLeftY="120" width="200" height="80">
+                    <colgroup><col span="2" width="100"/><col width="50"/></colgroup>
+                    <tr height="40"><td/><td/><td/></tr>
+                    <tr height="60"><td/><td/><td/></tr>
+                  </table>
+                </data>
+              </slide>
+            </presentation>
+            """
+        )
+        issues_by_dimension = {issue["dimension"]: issue for issue in result["slides"][0]["issues"]}
+        self.assertEqual(result["summary"]["error_count"], 0)
+        self.assertEqual(result["summary"]["warning_count"], 0)
+        self.assertEqual(result["summary"]["info_count"], 2)
+        self.assertEqual(issues_by_dimension["width"]["level"], "info")
+        self.assertEqual(issues_by_dimension["width"]["code"], "table_resolved_size_mismatch")
+        self.assertEqual(issues_by_dimension["width"]["resolved_sizes"], [100, 100, 50])
+        self.assertEqual(issues_by_dimension["width"]["resolved_size"], 250)
+        self.assertEqual(issues_by_dimension["height"]["resolved_sizes"], [40, 60])
+        self.assertEqual(issues_by_dimension["height"]["resolved_size"], 100)
+
+    def test_lint_xml_does_not_report_info_when_table_target_size_is_resolved_exactly(self) -> None:
+        result = xml_text_overlap_lint.lint_xml(
+            """
+            <presentation xmlns="http://www.larkoffice.com/sml/2.0" width="960" height="540">
+              <slide xmlns="http://www.larkoffice.com/sml/2.0">
+                <data>
+                  <table id="size-match" topLeftX="40" topLeftY="120" width="300" height="100">
+                    <colgroup><col width="100"/><col/></colgroup>
+                    <tr height="40"><td/><td/></tr>
+                    <tr><td/><td/></tr>
+                  </table>
+                </data>
+              </slide>
+            </presentation>
+            """
+        )
+        self.assertEqual(result["summary"]["error_count"], 0)
+        self.assertEqual(result["summary"]["warning_count"], 0)
+        self.assertEqual(result["summary"]["info_count"], 0)
+        self.assertEqual(result["slides"][0]["issues"], [])
+
+    def test_lint_xml_keeps_resolved_table_sizes_positive_when_target_is_too_small(self) -> None:
+        result = xml_text_overlap_lint.lint_xml(
+            """
+            <presentation xmlns="http://www.larkoffice.com/sml/2.0" width="960" height="540">
+              <slide xmlns="http://www.larkoffice.com/sml/2.0">
+                <data>
+                  <table id="narrow-table" topLeftX="40" topLeftY="120" width="1">
+                    <colgroup><col/><col/></colgroup>
+                    <tr><td/><td/></tr>
+                  </table>
+                </data>
+              </slide>
+            </presentation>
+            """
+        )
+        issue = result["slides"][0]["issues"][0]
+        self.assertEqual(issue["dimension"], "width")
+        self.assertEqual(issue["resolved_sizes"], [1, 1])
+        self.assertEqual(issue["resolved_size"], 2)
+
+    def test_fill_last_size_gap_preserves_target_when_positive_sizes_are_possible(self) -> None:
+        final_sizes = xml_text_overlap_lint.fill_last_size_gap([10, 10], 3)
+        self.assertEqual(final_sizes, [2, 1])
+        self.assertEqual(sum(final_sizes), 3)
+
+    def test_cli_reports_table_layout_size_info_for_weighted_min_layout_cases(self) -> None:
+        cases = {
+            "target-exact": (
+                """
+                <table topLeftX="40" topLeftY="120" width="360" height="150">
+                  <colgroup><col width="100"/><col width="200"/></colgroup>
+                  <tr height="40"><td/><td/></tr><tr height="60"><td/><td/></tr>
+                </table>
+                """,
+                0,
+            ),
+            "declared-size-exceeds-target": (
+                """
+                <table topLeftX="40" topLeftY="120" width="200" height="80">
+                  <colgroup><col span="2" width="100"/><col width="50"/></colgroup>
+                  <tr height="40"><td/><td/><td/></tr><tr height="60"><td/><td/><td/></tr>
+                </table>
+                """,
+                2,
+            ),
+            "remaining-space-insufficient": (
+                """
+                <table topLeftX="40" topLeftY="120" width="80" height="30">
+                  <colgroup><col width="80"/><col/></colgroup>
+                  <tr height="40"><td/><td/></tr><tr><td/><td/></tr>
+                </table>
+                """,
+                2,
+            ),
+            "no-target-size": (
+                """
+                <table topLeftX="40" topLeftY="120">
+                  <colgroup><col width="80"/><col/></colgroup>
+                  <tr height="40"><td/><td/></tr><tr><td/><td/></tr>
+                </table>
+                """,
+                0,
+            ),
+        }
+        script_path = Path(xml_text_overlap_lint.__file__).resolve()
+        with tempfile.TemporaryDirectory() as temp_dir:
+            for name, (table_xml, expected_info_count) in cases.items():
+                with self.subTest(case=name):
+                    input_path = Path(temp_dir) / f"{name}.xml"
+                    input_path.write_text(
+                        f"""
+                        <presentation xmlns="http://www.larkoffice.com/sml/2.0" width="960" height="540">
+                          <slide xmlns="http://www.larkoffice.com/sml/2.0"><data>{table_xml}</data></slide>
+                        </presentation>
+                        """,
+                        encoding="utf-8",
+                    )
+                    completed = subprocess.run(
+                        [sys.executable, str(script_path), "--input", str(input_path)],
+                        capture_output=True,
+                        check=False,
+                        text=True,
+                    )
+                    result = json.loads(completed.stdout)
+                    self.assertEqual(completed.returncode, 0, completed.stderr)
+                    self.assertEqual(result["summary"]["error_count"], 0)
+                    self.assertEqual(result["summary"]["warning_count"], 0)
+                    self.assertEqual(result["summary"]["info_count"], expected_info_count)
+                    self.assertTrue(
+                        all(issue["level"] == "info" for issue in result["slides"][0]["issues"]),
+                        result["slides"][0]["issues"],
+                    )
+
     def test_lint_xml_warns_for_whiteboard_external_boundary_overlap(self) -> None:
         result = xml_text_overlap_lint.lint_xml(
             """

skills/lark-vc-agent/SKILL.md

@@ -92,7 +92,7 @@ metadata:
 ### 3. 发送会中文本或会中表情（写操作）
 
 1. 用户明确要求在当前进行中的会议里发送提示、说明、会中表情，或反馈“听不到 / 看不到 / 声音清楚 / 效果不错”时，用 `+meeting-message-send`。
-2. 输入是长数字 `meeting_id`，不是 9 位会议号。若用户只给 9 位会议号，先按当前身份执行 `+meeting-list-active` 并按 `meeting_no` 匹配，匹配到唯一会议后再发送；不要为了发消息自动入会。
+2. 输入是长数字 `meeting_id`，不是 9 位会议号。若用户只给 9 位会议号，先按当前身份执行 `+meeting-list-active` 并按 `meeting_no` 匹配，匹配到唯一会议后再发送；不要为了发消息自动入会。发消息只需 `meeting_id`，不要先查 `+detail`。
 3. 身份必须延续：`meeting_id` 来自用户身份发现，就继续 `--as user`；来自应用身份发现或应用机器人入会，就继续 `--as bot`。
 4. 文本消息使用 `--text`；会中表情 / 反馈使用 `--emoji-type`。`--emoji-type` 必须从 reference 里的完整列表中选择，大小写敏感。
 5. 支持普通 Feishu reaction emoji（如 `LOVE`、`SMILE`、`THUMBSUP`）和 4 个 VC 反馈 key（`VC_CanNotSee`、`VC_NoSound`、`VC_LooksGood`、`VC_SoundsClear`）。

tests/cli_e2e/base/base_basic_workflow_test.go

@@ -15,6 +15,7 @@ import (
 )
 
 func TestBase_BasicWorkflow(t *testing.T) {
+	clie2e.SkipWithoutTenantAccessToken(t)
 	parentT := t
 	ctx, cancel := context.WithTimeout(context.Background(), 4*time.Minute)
 	t.Cleanup(cancel)

tests/cli_e2e/base/base_role_workflow_test.go

@@ -15,6 +15,7 @@ import (
 )
 
 func TestBase_RoleWorkflow(t *testing.T) {
+	clie2e.SkipWithoutTenantAccessToken(t)
 	parentT := t
 	ctx, cancel := context.WithTimeout(context.Background(), 4*time.Minute)
 	t.Cleanup(cancel)

tests/cli_e2e/calendar/calendar_create_event_test.go

@@ -16,6 +16,7 @@ import (
 
 // TestCalendar_CreateEvent tests the workflow of creating a calendar event.
 func TestCalendar_CreateEvent(t *testing.T) {
+	clie2e.SkipWithoutTenantAccessToken(t)
 	parentT := t
 	ctx, cancel := context.WithTimeout(context.Background(), 2*time.Minute)
 	t.Cleanup(cancel)

tests/cli_e2e/calendar/calendar_manage_calendar_test.go

@@ -16,6 +16,7 @@ import (
 
 // TestCalendar_ManageCalendar tests the workflow of managing calendars.
 func TestCalendar_ManageCalendar(t *testing.T) {
+	clie2e.SkipWithoutTenantAccessToken(t)
 	parentT := t
 	ctx, cancel := context.WithTimeout(context.Background(), 2*time.Minute)
 	t.Cleanup(cancel)

tests/cli_e2e/calendar/calendar_rsvp_workflow_test.go

@@ -39,6 +39,7 @@ func requireFreebusyEntry(t *testing.T, stdout string, startAt time.Time, endAt
 }
 
 func TestCalendar_RSVPWorkflowAsUser(t *testing.T) {
+	clie2e.SkipWithoutTenantAccessToken(t)
 	parentT := t
 	ctx, cancel := context.WithTimeout(context.Background(), 2*time.Minute)
 	t.Cleanup(cancel)

tests/cli_e2e/calendar/calendar_update_event_test.go

@@ -15,6 +15,7 @@ import (
 )
 
 func TestCalendar_UpdateEventWorkflow(t *testing.T) {
+	clie2e.SkipWithoutTenantAccessToken(t)
 	parentT := t
 	ctx, cancel := context.WithTimeout(context.Background(), 2*time.Minute)
 	t.Cleanup(cancel)

tests/cli_e2e/contact/contact_lookup_workflow_test.go

@@ -52,6 +52,7 @@ func TestContact_LookupWorkflowAsUser(t *testing.T) {
 }
 
 func TestContact_LookupWorkflowAsBot(t *testing.T) {
+	clie2e.SkipWithoutTenantAccessToken(t)
 	ctx, cancel := context.WithTimeout(context.Background(), 2*time.Minute)
 	t.Cleanup(cancel)
 

tests/cli_e2e/core.go

@@ -38,7 +38,7 @@ const (
 
 func SkipWithoutUserToken(t *testing.T) {
 	t.Helper()
-	if os.Getenv("LARKSUITE_CLI_USER_ACCESS_TOKEN") != "" {
+	if os.Getenv("LARKSUITE_CLI_USER_ACCESS_TOKEN") != "" || os.Getenv("TEST_USER_ACCESS_TOKEN") != "" {
 		return
 	}
 
@@ -75,6 +75,27 @@ func SkipWithoutUserToken(t *testing.T) {
 	}
 }
 
+func SkipWithoutTenantAccessToken(t *testing.T) {
+	t.Helper()
+	token := os.Getenv("TEST_TENANT_ACCESS_TOKEN")
+	if token == "" {
+		token = os.Getenv("LARKSUITE_CLI_TENANT_ACCESS_TOKEN")
+	}
+	appID := os.Getenv("TEST_BOT1_APP_ID")
+	if appID == "" {
+		appID = os.Getenv("LARKSUITE_CLI_APP_ID")
+	}
+	if token == "" || appID == "" {
+		t.Skip("skipped: tenant test credentials not set")
+	}
+
+	// Scope standard env credentials to tests that explicitly require a live
+	// tenant token. Keeping TEST_* variables in the gotestsum parent prevents
+	// config and dry-run CLI subprocesses from activating the env provider.
+	t.Setenv("LARKSUITE_CLI_APP_ID", appID)
+	t.Setenv("LARKSUITE_CLI_TENANT_ACCESS_TOKEN", token)
+}
+
 // DryRunGet reads a field from the dry-run payload inside the standard success envelope.
 func DryRunGet(stdout, path string) gjson.Result {
 	if path == "" {
@@ -225,13 +246,13 @@ func buildCommandEnv(req Request) []string {
 		overrides[k] = v
 	}
 	// Keep user-token injection scoped to user-only test commands so bot
-	// commands continue to use config-init credentials in the same process.
+	// commands retain the process-level bot credentials.
 	if req.DefaultAs == "user" {
 		if appID := os.Getenv("TEST_BOT1_APP_ID"); appID != "" {
-			if token := os.Getenv("TEST_USER_ACCESS_TOKEN"); token != "" {
-				overrides["LARKSUITE_CLI_APP_ID"] = appID
-				overrides["LARKSUITE_CLI_USER_ACCESS_TOKEN"] = token
-			}
+			overrides["LARKSUITE_CLI_APP_ID"] = appID
+		}
+		if token := os.Getenv("TEST_USER_ACCESS_TOKEN"); token != "" {
+			overrides["LARKSUITE_CLI_USER_ACCESS_TOKEN"] = token
 		}
 	}
 	for k, v := range overrides {

tests/cli_e2e/core_test.go

@@ -113,6 +113,19 @@ func TestSkipWithoutUserToken(t *testing.T) {
 		assert.True(t, ran)
 	})
 
+	t.Run("returns immediately when test user access token exists", func(t *testing.T) {
+		t.Setenv("LARKSUITE_CLI_USER_ACCESS_TOKEN", "")
+		t.Setenv("TEST_USER_ACCESS_TOKEN", "uat-from-test-env")
+
+		ran := false
+		ok := t.Run("inner", func(t *testing.T) {
+			SkipWithoutUserToken(t)
+			ran = true
+		})
+		require.True(t, ok)
+		assert.True(t, ran)
+	})
+
 	t.Run("accepts verified local auth status", func(t *testing.T) {
 		fake := newFakeCLI(t)
 		t.Setenv("LARKSUITE_CLI_USER_ACCESS_TOKEN", "")
@@ -146,6 +159,54 @@ func TestSkipWithoutUserToken(t *testing.T) {
 	})
 }
 
+func TestSkipWithoutTenantAccessToken(t *testing.T) {
+	t.Run("skips when env tenant access token is missing", func(t *testing.T) {
+		t.Setenv("TEST_BOT1_APP_ID", "")
+		t.Setenv("TEST_TENANT_ACCESS_TOKEN", "")
+		t.Setenv("LARKSUITE_CLI_APP_ID", "")
+		t.Setenv("LARKSUITE_CLI_TENANT_ACCESS_TOKEN", "")
+
+		ran := false
+		ok := t.Run("inner", func(t *testing.T) {
+			SkipWithoutTenantAccessToken(t)
+			ran = true
+		})
+		require.True(t, ok)
+		assert.False(t, ran)
+	})
+
+	t.Run("accepts standard tenant credentials", func(t *testing.T) {
+		t.Setenv("TEST_BOT1_APP_ID", "")
+		t.Setenv("TEST_TENANT_ACCESS_TOKEN", "")
+		t.Setenv("LARKSUITE_CLI_APP_ID", "app-from-env")
+		t.Setenv("LARKSUITE_CLI_TENANT_ACCESS_TOKEN", "test-token")
+
+		ran := false
+		ok := t.Run("inner", func(t *testing.T) {
+			SkipWithoutTenantAccessToken(t)
+			ran = true
+		})
+		require.True(t, ok)
+		assert.True(t, ran)
+	})
+
+	t.Run("scopes shared tenant credentials to the requiring test", func(t *testing.T) {
+		t.Setenv("TEST_BOT1_APP_ID", "shared-test-app")
+		t.Setenv("TEST_TENANT_ACCESS_TOKEN", "shared-test-token")
+		t.Setenv("LARKSUITE_CLI_APP_ID", "")
+		t.Setenv("LARKSUITE_CLI_TENANT_ACCESS_TOKEN", "")
+
+		ok := t.Run("inner", func(t *testing.T) {
+			SkipWithoutTenantAccessToken(t)
+			assert.Equal(t, "shared-test-app", os.Getenv("LARKSUITE_CLI_APP_ID"))
+			assert.Equal(t, "shared-test-token", os.Getenv("LARKSUITE_CLI_TENANT_ACCESS_TOKEN"))
+		})
+		require.True(t, ok)
+		assert.Empty(t, os.Getenv("LARKSUITE_CLI_APP_ID"))
+		assert.Empty(t, os.Getenv("LARKSUITE_CLI_TENANT_ACCESS_TOKEN"))
+	})
+}
+
 func TestRunCmd(t *testing.T) {
 	t.Run("returns stdout json on success", func(t *testing.T) {
 		fake := newFakeCLI(t)
@@ -214,6 +275,8 @@ func TestRunCmd(t *testing.T) {
 	})
 
 	t.Run("injects user token env only for user commands", func(t *testing.T) {
+		t.Setenv("LARKSUITE_CLI_APP_ID", "")
+		t.Setenv("LARKSUITE_CLI_USER_ACCESS_TOKEN", "")
 		t.Setenv("TEST_BOT1_APP_ID", "cli_app_test")
 		t.Setenv("TEST_USER_ACCESS_TOKEN", "uat_test")
 
@@ -224,6 +287,10 @@ func TestRunCmd(t *testing.T) {
 		env = buildCommandEnv(Request{DefaultAs: "bot"})
 		assert.NotContains(t, env, "LARKSUITE_CLI_APP_ID=cli_app_test")
 		assert.NotContains(t, env, "LARKSUITE_CLI_USER_ACCESS_TOKEN=uat_test")
+
+		env = buildCommandEnv(Request{})
+		assert.NotContains(t, env, "LARKSUITE_CLI_APP_ID=cli_app_test")
+		assert.NotContains(t, env, "LARKSUITE_CLI_USER_ACCESS_TOKEN=uat_test")
 	})
 
 	t.Run("retries structured retryable service errors by default", func(t *testing.T) {

tests/cli_e2e/docs/docs_create_fetch_test.go

@@ -17,6 +17,7 @@ import (
 
 // TestDocs_CreateAndFetchWorkflow tests the create and fetch lifecycle.
 func TestDocs_CreateAndFetchWorkflowAsBot(t *testing.T) {
+	clie2e.SkipWithoutTenantAccessToken(t)
 	ctx, cancel := context.WithTimeout(context.Background(), 2*time.Minute)
 	t.Cleanup(cancel)
 

tests/cli_e2e/docs/docs_update_test.go

@@ -17,6 +17,7 @@ import (
 
 // TestDocs_UpdateWorkflow tests the create, update, and verify lifecycle.
 func TestDocs_UpdateWorkflow(t *testing.T) {
+	clie2e.SkipWithoutTenantAccessToken(t)
 	parentT := t
 	ctx, cancel := context.WithTimeout(context.Background(), 2*time.Minute)
 	t.Cleanup(cancel)

tests/cli_e2e/drive/coverage.md

@@ -16,7 +16,7 @@
 - TestDriveAddCommentMarkdownFileWorkflow: opt-in live workflow skeleton for comment write/read, gated by `LARK_DRIVE_MD_COMMENT_E2E=1`; creates a Markdown file, adds a file comment, lists it back through `drive +list-comments`, and cleans up.
 - TestDrive_SecureLabelDryRun: dry-run coverage for `drive +secure-label-list` and `drive +secure-label-update`; asserts label-list query params and update URL→type inference, request method/URL/type query, and `label-id` body shape. Runs without hitting live APIs because update can trigger document-level security approval flows.
 - TestDriveExportDryRun_FileNameMetadata / TestDriveExportDryRun_WikiURLPlansResolveBeforeExportTask / TestDriveExportDryRun_WikiTokenTypePlansResolveBeforeExportTask / TestDriveExportDryRun_MarkdownFetchAPI / TestDriveExportDryRun_BitableBaseOnlySchema: dry-run coverage for `drive +export`; asserts export task request shape, Wiki URL and `--doc-type wiki` token `get_node -> export_tasks` planning, markdown fetch request shape without docs fetch `extra_param`, local `--file-name` / `--output-dir` metadata, and `bitable` `.base` `only_schema` request body without calling live APIs.
-- TestDriveDeleteDryRunAsyncParams / TestDrive_DeleteAsyncWorkflow: dry-run coverage for `drive +delete` pins `DELETE /drive/v1/files/:file_token` params with `type` plus `async=true` and the follow-up `task_check` plan; live workflow creates and deletes a docx, an empty folder, and a non-empty folder, asserts each delete returns `task_id`, queries every returned task via `drive +task_result --scenario task_check`, and verifies the targets disappear.
+- TestDriveDeleteDryRunAsyncParams / TestDrive_DeleteAsyncWorkflow: dry-run coverage for `drive +delete` pins `DELETE /drive/v1/files/:file_token` params with `type` plus `async=true` and the follow-up `task_check` plan; live workflow creates and deletes a docx, an empty folder, and a non-empty folder, converging every delete outcome to the resource-gone terminal state: async deletes (non-empty `task_id`) are verified via `drive +task_result --scenario task_check`, sync deletes (empty `task_id`) assert `deleted=true`, and the one verified backend transient (`server_error: "drive task failed"`) passes once the target is confirmed gone (retried up to 3 times otherwise); any other delete failure stays fatal.
 - TestDrive_PullDryRun / TestDrive_PullDryRunAcceptsDuplicateRemoteStrategies: dry-run coverage for `drive +pull`; asserts the list-files request shape, Validate-stage safety guards, and acceptance of `--on-duplicate-remote=rename|newest|oldest` by the real CLI binary.
 - TestDrive_PushDryRun / TestDrive_PushDryRunAcceptsDuplicateRemoteStrategies: dry-run coverage for `drive +push`; asserts the list-files request shape, Validate-stage safety guards, conditional delete preflight, and acceptance of `--on-duplicate-remote=newest|oldest` by the real CLI binary.
 - Cleanup note: `drive files delete` is only exercised in cleanup and is intentionally left uncovered.
@@ -30,7 +30,7 @@
 | ✓ | drive +add-comment | shortcut | drive_add_comment_dryrun_test.go::TestDriveAddCommentDryRun_File; drive_add_comment_dryrun_test.go::TestDriveAddCommentDryRun_Base | `--doc` file URL vs bare token + `--type file`; supported-extension metadata gate; placeholder `anchor.block_id`; Base URL with `--block-id <table-id>!<record-id>!<view-id>` | dry-run coverage in place; opt-in live file workflow exists behind `LARK_DRIVE_MD_COMMENT_E2E=1` |
 | ✓ | drive +list-comments | shortcut | drive_list_comments_dryrun_test.go::TestDriveListCommentsDryRun_DocxDefaults; drive_list_comments_dryrun_test.go::TestDriveListCommentsDryRun_AppsPageURL; drive_list_comments_dryrun_test.go::TestDriveListCommentsDryRun_WikiToken; drive_add_comment_workflow_test.go::TestDriveAddCommentMarkdownFileWorkflow | `--url`; apps `/page/<token>` URL; `--token + --type wiki`; `--solved-status=false\|all`; `--comment-scope=all\|partial`; `--need-relation`; `--page-size` | dry-run locks URL/token parsing, apps `file_type=apps`, default unresolved filter, omitted all-scope filter, omitted `user_id_type`, and Wiki unwrap request shape; opt-in live workflow verifies a created file comment can be listed back |
 | ✓ | drive +apply-permission | shortcut | drive_apply_permission_dryrun_test.go::TestDrive_ApplyPermissionDryRun | `--token` URL vs bare; `--type` (enum) with URL inference; `--perm view\|edit`; `--remark` optional | dry-run only; no live-apply E2E because a real request pushes a card to the owner |
-| ✓ | drive +delete | shortcut | drive_delete_dryrun_test.go::TestDriveDeleteDryRunAsyncParams + drive_delete_workflow_test.go::TestDrive_DeleteAsyncWorkflow | `--file-token`; `--type`; fixed query `async=true`; `task_check` follow-up | dry-run locks async request shape; live workflow covers docx, empty folder, and non-empty folder async deletion |
+| ✓ | drive +delete | shortcut | drive_delete_dryrun_test.go::TestDriveDeleteDryRunAsyncParams + drive_delete_workflow_test.go::TestDrive_DeleteAsyncWorkflow | `--file-token`; `--type`; fixed query `async=true`; `task_check` follow-up | dry-run locks async request shape; live workflow covers docx, empty folder, and non-empty folder deletion with async/sync/transient-failure convergence |
 | ✕ | drive +download | shortcut |  | none | no file fixture workflow yet |
 | ✓ | drive +export | shortcut | drive_export_dryrun_test.go::TestDriveExportDryRun_FileNameMetadata + TestDriveExportDryRun_WikiURLPlansResolveBeforeExportTask + TestDriveExportDryRun_WikiTokenTypePlansResolveBeforeExportTask + TestDriveExportDryRun_MarkdownFetchAPI + TestDriveExportDryRun_BitableBaseOnlySchema | `--url`; `--token`; `--doc-type`; `--file-extension`; `--file-name`; `--output-dir`; `--only-schema`; Wiki URL / `--doc-type wiki` resolve step; markdown fetch omits docs fetch `extra_param` | dry-run only; no live export workflow yet |
 | ✕ | drive +export-download | shortcut |  | none | no export-download workflow yet |

tests/cli_e2e/drive/drive_add_comment_workflow_test.go

@@ -15,6 +15,7 @@ import (
 )
 
 func TestDriveAddCommentMarkdownFileWorkflow(t *testing.T) {
+	clie2e.SkipWithoutTenantAccessToken(t)
 	if os.Getenv("LARK_DRIVE_MD_COMMENT_E2E") == "" {
 		t.Skip("set LARK_DRIVE_MD_COMMENT_E2E=1 to run the supported file comment workflow")
 	}

tests/cli_e2e/drive/drive_delete_workflow_helper_test.go

@@ -0,0 +1,370 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package drive
+
+import (
+	"context"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"testing"
+	"time"
+
+	clie2e "github.com/larksuite/cli/tests/cli_e2e"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestDeleteAsyncAndVerify(t *testing.T) {
+	t.Run("sync delete without task_id skips task_result", func(t *testing.T) {
+		fake := mustWriteDriveDeleteWorkflowFakeCLI(t)
+		t.Setenv(clie2e.EnvBinaryPath, fake)
+		t.Setenv("FAKE_WORKFLOW_DELETE_MODE", "sync")
+		t.Setenv("FAKE_WORKFLOW_META_MODE", "gone")
+		counters := setupFakeWorkflowCounters(t)
+
+		taskID := deleteAsyncAndVerify(t, context.Background(), "docx_sync", "docx")
+		assert.Empty(t, taskID)
+
+		assert.Equal(t, "1", readFakeCounter(t, counters.deletes), "sync path must delete exactly once")
+		assert.Equal(t, "1", readFakeCounter(t, counters.metas), "sync path must still verify the resource is gone")
+		assert.Equal(t, "0", readFakeCounter(t, counters.taskResults), "sync path must not query task status")
+	})
+
+	t.Run("transient failure with resource gone is tolerated", func(t *testing.T) {
+		fake := mustWriteDriveDeleteWorkflowFakeCLI(t)
+		t.Setenv(clie2e.EnvBinaryPath, fake)
+		t.Setenv("FAKE_WORKFLOW_DELETE_MODE", "fail")
+		t.Setenv("FAKE_WORKFLOW_META_MODE", "gone")
+		counters := setupFakeWorkflowCounters(t)
+
+		taskID := deleteAsyncAndVerify(t, context.Background(), "docx_transient", "docx")
+		assert.Empty(t, taskID)
+
+		assert.Equal(t, "1", readFakeCounter(t, counters.deletes), "resource already gone must not trigger another delete attempt")
+		assert.Equal(t, "1", readFakeCounter(t, counters.metas), "transient failure must verify the terminal state")
+		assert.Equal(t, "0", readFakeCounter(t, counters.taskResults))
+	})
+
+	t.Run("failed delete retries until async success", func(t *testing.T) {
+		fake := mustWriteDriveDeleteWorkflowFakeCLI(t)
+		t.Setenv(clie2e.EnvBinaryPath, fake)
+		t.Setenv("FAKE_WORKFLOW_DELETE_MODE", "fail-then-async")
+		t.Setenv("FAKE_WORKFLOW_META_MODE", "exists-then-gone")
+		t.Setenv("FAKE_WORKFLOW_TASK_RESULT_OK", "1")
+		counters := setupFakeWorkflowCounters(t)
+		withFastDeleteWorkflowBackoff(t)
+
+		taskID := deleteAsyncAndVerify(t, context.Background(), "docx_retry", "docx")
+		assert.Equal(t, "task_123", taskID)
+
+		assert.Equal(t, "2", readFakeCounter(t, counters.deletes))
+		assert.Equal(t, "2", readFakeCounter(t, counters.metas), "one terminal-state check after the failure plus the final visibility wait")
+		assert.Equal(t, "1", readFakeCounter(t, counters.taskResults), "async success must verify the task result")
+	})
+}
+
+// TestIsTransientDriveDeleteFailure locks the tolerance boundary: only the one
+// verified backend transient may fall through to terminal-state checking, so a
+// crash, a protocol regression, or any other error keeps failing the workflow
+// even when the resource happens to be gone.
+func TestIsTransientDriveDeleteFailure(t *testing.T) {
+	t.Run("matches compact envelope", func(t *testing.T) {
+		result := &clie2e.Result{
+			ExitCode: 1,
+			Stderr:   "Deleting docx tok...\n{\"ok\":false,\"identity\":\"bot\",\"error\":{\"type\":\"api\",\"subtype\":\"server_error\",\"message\":\"drive task failed\"}}",
+		}
+		assert.True(t, isTransientDriveDeleteFailure(result))
+	})
+
+	t.Run("matches pretty-printed envelope from CI", func(t *testing.T) {
+		result := &clie2e.Result{
+			ExitCode: 1,
+			Stderr: "Deleting docx NTw0...Rngb...\nDelete is async, polling task schedule|7663369798226545963...\n" +
+				"{\n  \"ok\": false,\n  \"identity\": \"bot\",\n  \"error\": {\n    \"type\": \"api\",\n    \"subtype\": \"server_error\",\n    \"message\": \"drive task failed\"\n  }\n}",
+		}
+		assert.True(t, isTransientDriveDeleteFailure(result))
+	})
+
+	t.Run("rejects other server errors", func(t *testing.T) {
+		result := &clie2e.Result{
+			ExitCode: 1,
+			Stderr:   "{\"ok\":false,\"identity\":\"bot\",\"error\":{\"type\":\"api\",\"subtype\":\"server_error\",\"message\":\"internal error\"}}",
+		}
+		assert.False(t, isTransientDriveDeleteFailure(result))
+	})
+
+	t.Run("rejects non-server-error subtypes", func(t *testing.T) {
+		result := &clie2e.Result{
+			ExitCode: 1,
+			Stderr:   "{\"ok\":false,\"identity\":\"bot\",\"error\":{\"type\":\"api\",\"subtype\":\"permission_denied\",\"message\":\"drive task failed\"}}",
+		}
+		assert.False(t, isTransientDriveDeleteFailure(result))
+	})
+
+	t.Run("rejects non-JSON output", func(t *testing.T) {
+		result := &clie2e.Result{ExitCode: 2, Stderr: "panic: runtime error"}
+		assert.False(t, isTransientDriveDeleteFailure(result))
+	})
+
+	t.Run("rejects nil result", func(t *testing.T) {
+		assert.False(t, isTransientDriveDeleteFailure(nil))
+	})
+}
+
+// TestDeleteAsyncAndVerifyRejectsUnexpectedFailure locks the P1 boundary
+// end-to-end by re-running this test binary as a subprocess that really calls
+// deleteAsyncAndVerify: an unrelated non-zero exit must fail the helper
+// immediately — no terminal-state check may rescue it even though meta reports
+// the resource gone. Fatalf cannot be observed on the parent *testing.T, so
+// the boundary is proven by the child process exiting non-zero AND the meta
+// endpoint never being reached. Removing the isTransientDriveDeleteFailure
+// guard from the main loop turns this test red.
+func TestDeleteAsyncAndVerifyRejectsUnexpectedFailure(t *testing.T) {
+	fake := mustWriteDriveDeleteWorkflowFakeCLI(t)
+	counters := newFakeWorkflowCounterPaths(t)
+
+	output, err := runDeleteWorkflowSubprocess(t, fake, counters, map[string]string{
+		"FAKE_WORKFLOW_TOKEN":       "docx_unexpected",
+		"FAKE_WORKFLOW_DELETE_MODE": "fail-unexpected",
+		"FAKE_WORKFLOW_META_MODE":   "gone",
+	})
+	require.Error(t, err, "deleteAsyncAndVerify must fail the test process on an unexpected delete error\noutput:\n%s", output)
+	assert.Contains(t, output, "drive +delete failed with an unexpected error", "output:\n%s", output)
+
+	assert.Equal(t, "1", readFakeCounter(t, counters.deletes))
+	assert.Equal(t, "0", readFakeCounter(t, counters.metas), "unexpected failures must not fall through to terminal-state checking")
+	assert.Equal(t, "0", readFakeCounter(t, counters.taskResults))
+}
+
+// TestDeleteAsyncAndVerifyFailsOnTaskResultFailure proves a non-zero
+// drive +task_result exit fails the workflow before the final visibility
+// polling: the task-result endpoint is reached once and the meta endpoint
+// never.
+func TestDeleteAsyncAndVerifyFailsOnTaskResultFailure(t *testing.T) {
+	fake := mustWriteDriveDeleteWorkflowFakeCLI(t)
+	counters := newFakeWorkflowCounterPaths(t)
+
+	output, err := runDeleteWorkflowSubprocess(t, fake, counters, map[string]string{
+		"FAKE_WORKFLOW_TOKEN":       "docx_taskresult",
+		"FAKE_WORKFLOW_DELETE_MODE": "async",
+		"FAKE_WORKFLOW_META_MODE":   "gone",
+		// FAKE_WORKFLOW_TASK_RESULT_OK stays unset: +task_result exits 2.
+	})
+	require.Error(t, err, "deleteAsyncAndVerify must fail the test process when +task_result fails\noutput:\n%s", output)
+	assert.Contains(t, output, "drive +task_result failed", "output:\n%s", output)
+
+	assert.Equal(t, "1", readFakeCounter(t, counters.deletes))
+	assert.Equal(t, "1", readFakeCounter(t, counters.taskResults))
+	assert.Equal(t, "0", readFakeCounter(t, counters.metas), "task-result failure must abort before visibility polling")
+}
+
+// TestDeleteAsyncAndVerifyStopsAfterExhaustedRetries proves the transient
+// tolerance is bounded: with the resource still present, exactly
+// deleteWorkflowMaxAttempts delete attempts (each followed by one terminal
+// state check) run before the workflow fails for good.
+func TestDeleteAsyncAndVerifyStopsAfterExhaustedRetries(t *testing.T) {
+	fake := mustWriteDriveDeleteWorkflowFakeCLI(t)
+	counters := newFakeWorkflowCounterPaths(t)
+
+	output, err := runDeleteWorkflowSubprocess(t, fake, counters, map[string]string{
+		"FAKE_WORKFLOW_TOKEN":        "docx_exhausted",
+		"FAKE_WORKFLOW_DELETE_MODE":  "fail",
+		"FAKE_WORKFLOW_META_MODE":    "exists",
+		"FAKE_WORKFLOW_FAST_BACKOFF": "1",
+	})
+	require.Error(t, err, "deleteAsyncAndVerify must fail the test process after exhausting retries\noutput:\n%s", output)
+	assert.Contains(t, output, "drive +delete failed 3 times", "output:\n%s", output)
+
+	assert.Equal(t, "3", readFakeCounter(t, counters.deletes))
+	assert.Equal(t, "3", readFakeCounter(t, counters.metas))
+	assert.Equal(t, "0", readFakeCounter(t, counters.taskResults))
+}
+
+// runDeleteWorkflowSubprocess re-runs this test binary anchored to the child
+// entry point below with the fake CLI and counter files wired in via env.
+func runDeleteWorkflowSubprocess(t *testing.T, fake string, counters fakeWorkflowCounters, env map[string]string) (string, error) {
+	t.Helper()
+
+	cmd := exec.Command(os.Args[0], "-test.run=TestDeleteAsyncAndVerifySubprocess$", "-test.v")
+	cmd.Env = append(os.Environ(),
+		"FAKE_WORKFLOW_SUBPROCESS=1",
+		clie2e.EnvBinaryPath+"="+fake,
+		"FAKE_WORKFLOW_DELETE_STATE="+counters.deletes,
+		"FAKE_WORKFLOW_META_STATE="+counters.metas,
+		"FAKE_WORKFLOW_TASK_RESULT_STATE="+counters.taskResults,
+	)
+	for k, v := range env {
+		cmd.Env = append(cmd.Env, k+"="+v)
+	}
+	output, err := cmd.CombinedOutput()
+	return string(output), err
+}
+
+// TestDeleteAsyncAndVerifySubprocess is the child entry point driven by
+// runDeleteWorkflowSubprocess. It does nothing in a normal test run.
+func TestDeleteAsyncAndVerifySubprocess(t *testing.T) {
+	if os.Getenv("FAKE_WORKFLOW_SUBPROCESS") != "1" {
+		return
+	}
+	if os.Getenv("FAKE_WORKFLOW_FAST_BACKOFF") == "1" {
+		deleteWorkflowRetryBackoff = time.Millisecond
+	}
+	deleteAsyncAndVerify(t, context.Background(), os.Getenv("FAKE_WORKFLOW_TOKEN"), "docx")
+}
+
+type fakeWorkflowCounters struct {
+	deletes     string
+	metas       string
+	taskResults string
+}
+
+func newFakeWorkflowCounterPaths(t *testing.T) fakeWorkflowCounters {
+	t.Helper()
+
+	dir := t.TempDir()
+	return fakeWorkflowCounters{
+		deletes:     filepath.Join(dir, "delete-attempts"),
+		metas:       filepath.Join(dir, "meta-calls"),
+		taskResults: filepath.Join(dir, "task-result-calls"),
+	}
+}
+
+// setupFakeWorkflowCounters wires per-endpoint call counters into the fake CLI
+// so tests can assert exactly which commands ran.
+func setupFakeWorkflowCounters(t *testing.T) fakeWorkflowCounters {
+	t.Helper()
+
+	counters := newFakeWorkflowCounterPaths(t)
+	t.Setenv("FAKE_WORKFLOW_DELETE_STATE", counters.deletes)
+	t.Setenv("FAKE_WORKFLOW_META_STATE", counters.metas)
+	t.Setenv("FAKE_WORKFLOW_TASK_RESULT_STATE", counters.taskResults)
+	return counters
+}
+
+func readFakeCounter(t *testing.T, path string) string {
+	t.Helper()
+
+	data, err := os.ReadFile(path)
+	if os.IsNotExist(err) {
+		return "0"
+	}
+	require.NoError(t, err)
+	return string(data)
+}
+
+func withFastDeleteWorkflowBackoff(t *testing.T) {
+	t.Helper()
+
+	original := deleteWorkflowRetryBackoff
+	deleteWorkflowRetryBackoff = time.Millisecond
+	t.Cleanup(func() {
+		deleteWorkflowRetryBackoff = original
+	})
+}
+
+// mustWriteDriveDeleteWorkflowFakeCLI writes a fake lark-cli that emulates the
+// drive delete outcomes exercised by deleteAsyncAndVerify. Every endpoint
+// bumps a per-endpoint counter when its FAKE_WORKFLOW_*_STATE env is set, so
+// tests can assert call contracts. +task_result rejects every call unless
+// FAKE_WORKFLOW_TASK_RESULT_OK=1, which proves the sync path never queries
+// task status.
+func mustWriteDriveDeleteWorkflowFakeCLI(t *testing.T) string {
+	t.Helper()
+
+	script := `#!/bin/sh
+bump_counter() {
+  state="$1"
+  count=0
+  if [ -f "$state" ]; then
+    count="$(cat "$state")"
+  fi
+  next=$((count + 1))
+  printf '%s' "$next" > "$state"
+  echo "$count"
+}
+
+if [ "$1" = "drive" ] && [ "$2" = "+delete" ]; then
+  count=0
+  if [ -n "$FAKE_WORKFLOW_DELETE_STATE" ]; then
+    count="$(bump_counter "$FAKE_WORKFLOW_DELETE_STATE")"
+  fi
+  case "$FAKE_WORKFLOW_DELETE_MODE" in
+  sync)
+    echo '{"ok":true,"identity":"bot","data":{"deleted":true,"file_token":"tok","type":"docx"}}'
+    exit 0
+    ;;
+  fail)
+    echo "Deleting docx tok..." >&2
+    echo '{"ok":false,"identity":"bot","error":{"type":"api","subtype":"server_error","message":"drive task failed"}}' >&2
+    exit 1
+    ;;
+  fail-unexpected)
+    echo '{"ok":false,"identity":"bot","error":{"type":"api","subtype":"invalid_request","message":"file token not found"}}' >&2
+    exit 1
+    ;;
+  async)
+    echo '{"ok":true,"identity":"bot","data":{"task_id":"task_123","status":"success","file_token":"tok","type":"docx"}}'
+    exit 0
+    ;;
+  fail-then-async)
+    if [ "$count" -lt 1 ]; then
+      echo '{"ok":false,"identity":"bot","error":{"type":"api","subtype":"server_error","message":"drive task failed"}}' >&2
+      exit 1
+    fi
+    echo '{"ok":true,"identity":"bot","data":{"task_id":"task_123","status":"success","file_token":"tok","type":"docx"}}'
+    exit 0
+    ;;
+  esac
+  echo "unexpected FAKE_WORKFLOW_DELETE_MODE: $FAKE_WORKFLOW_DELETE_MODE" >&2
+  exit 2
+fi
+
+if [ "$1" = "drive" ] && [ "$2" = "+task_result" ]; then
+  if [ -n "$FAKE_WORKFLOW_TASK_RESULT_STATE" ]; then
+    bump_counter "$FAKE_WORKFLOW_TASK_RESULT_STATE" > /dev/null
+  fi
+  if [ "${FAKE_WORKFLOW_TASK_RESULT_OK:-0}" != "1" ]; then
+    echo "unexpected +task_result call: $*" >&2
+    exit 2
+  fi
+  echo '{"ok":true,"identity":"bot","data":{"task_id":"task_123","status":"success","failed":false}}'
+  exit 0
+fi
+
+if [ "$1" = "api" ] && [ "$2" = "post" ] && [ "$3" = "/open-apis/drive/v1/metas/batch_query" ]; then
+  count=0
+  if [ -n "$FAKE_WORKFLOW_META_STATE" ]; then
+    count="$(bump_counter "$FAKE_WORKFLOW_META_STATE")"
+  fi
+  case "$FAKE_WORKFLOW_META_MODE" in
+  gone)
+    echo '{"ok":true,"data":{"metas":[]}}'
+    exit 0
+    ;;
+  exists)
+    echo '{"ok":true,"data":{"metas":[{"url":"https://example.com/still-visible"}]}}'
+    exit 0
+    ;;
+  exists-then-gone)
+    if [ "$count" -lt 1 ]; then
+      echo '{"ok":true,"data":{"metas":[{"url":"https://example.com/still-visible"}]}}'
+      exit 0
+    fi
+    echo '{"ok":true,"data":{"metas":[]}}'
+    exit 0
+    ;;
+  esac
+  echo "unexpected FAKE_WORKFLOW_META_MODE: $FAKE_WORKFLOW_META_MODE" >&2
+  exit 2
+fi
+
+echo "unexpected fake CLI args: $*" >&2
+exit 2
+`
+
+	binaryPath := filepath.Join(t.TempDir(), "fake-lark-cli")
+	require.NoError(t, os.WriteFile(binaryPath, []byte(script), 0o755))
+	return binaryPath
+}

tests/cli_e2e/drive/drive_delete_workflow_test.go

@@ -5,6 +5,7 @@ package drive
 
 import (
 	"context"
+	"strings"
 	"testing"
 	"time"
 
@@ -14,6 +15,8 @@ import (
 )
 
 func TestDrive_DeleteAsyncWorkflow(t *testing.T) {
+	clie2e.SkipWithoutTenantAccessToken(t)
+
 	parentT := t
 	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Minute)
 	t.Cleanup(cancel)
@@ -65,30 +68,117 @@ func createDeleteWorkflowDoc(t *testing.T, ctx context.Context, folderToken, tit
 	return docToken
 }
 
+const deleteWorkflowMaxAttempts = 3
+
+// deleteWorkflowRetryBackoff paces delete retries after a non-retryable
+// failure whose target still exists. Unit tests shrink it.
+var deleteWorkflowRetryBackoff = driveDeleteVisibilityPoll
+
+// deleteAsyncAndVerify deletes token and converges every server outcome to the
+// real postcondition: the resource is gone. Async deletes (non-empty task_id)
+// additionally verify the task via drive +task_result; sync deletes (empty
+// task_id) skip task polling; non-retryable delete failures (e.g. a transient
+// "drive task failed") pass when the resource is already gone and are retried
+// up to deleteWorkflowMaxAttempts times otherwise.
 func deleteAsyncAndVerify(t *testing.T, ctx context.Context, token, docType string) string {
 	t.Helper()
 
-	result, err := clie2e.RunCmdWithRetry(ctx, clie2e.Request{
-		Args:      []string{"drive", "+delete", "--file-token", token, "--type", docType, "--yes"},
-		DefaultAs: "bot",
-	}, driveDeleteRetry)
-	require.NoError(t, err)
-	result.AssertExitCode(t, 0)
-	result.AssertStdoutStatus(t, true)
+	var lastResult *clie2e.Result
+	for attempt := 1; attempt <= deleteWorkflowMaxAttempts; attempt++ {
+		result, err := clie2e.RunCmdWithRetry(ctx, clie2e.Request{
+			Args:      []string{"drive", "+delete", "--file-token", token, "--type", docType, "--yes"},
+			DefaultAs: "bot",
+		}, driveDeleteRetry)
+		require.NoError(t, err)
+		lastResult = result
 
-	taskID := gjson.Get(result.Stdout, "data.task_id").String()
-	require.NotEmpty(t, taskID, "delete must return async task_id\nstdout:\n%s", result.Stdout)
+		if result.ExitCode == 0 {
+			result.AssertStdoutStatus(t, true)
+			taskID := gjson.Get(result.Stdout, "data.task_id").String()
+			if taskID == "" {
+				// Sync completion: the server deleted the resource inline and
+				// returned no task to poll.
+				require.True(t, gjson.Get(result.Stdout, "data.deleted").Bool(), "sync delete must report deleted=true\nstdout:\n%s", result.Stdout)
+				t.Logf("drive +delete completed synchronously for %s %s (no task_id)", docType, token)
+			} else {
+				assertDriveDeleteTaskSucceeded(t, ctx, taskID)
+			}
+			require.NoError(t, waitDriveResourceDeleted(ctx, token, docType, "bot", driveDeleteVisibilityWait))
+			return taskID
+		}
+
+		// Only the one verified backend transient may fall through to
+		// terminal-state checking; any other failure is a real regression and
+		// must not be rescued by the resource happening to be gone.
+		if !isTransientDriveDeleteFailure(result) {
+			t.Fatalf("drive +delete failed with an unexpected error on attempt %d\nstdout:\n%s\nstderr:\n%s",
+				attempt, result.Stdout, result.Stderr)
+		}
+
+		// The failed delete task may still have removed the resource
+		// server-side, so check the real terminal state before retrying.
+		deleted, verifyErr := IsDriveResourceDeleted(ctx, token, docType, "bot")
+		require.NoError(t, verifyErr, "verify %s %s after failed delete attempt %d", docType, token, attempt)
+		if deleted {
+			t.Logf("drive +delete attempt %d failed transiently but %s %s is gone: stderr=%s", attempt, docType, token, result.Stderr)
+			return ""
+		}
+		if attempt < deleteWorkflowMaxAttempts {
+			t.Logf("drive +delete attempt %d failed and %s %s still exists; retrying: stderr=%s", attempt, docType, token, result.Stderr)
+			time.Sleep(deleteWorkflowRetryBackoff)
+		}
+	}
+
+	t.Fatalf("drive +delete failed %d times and %s %s still exists\nstdout:\n%s\nstderr:\n%s",
+		deleteWorkflowMaxAttempts, docType, token, lastResult.Stdout, lastResult.Stderr)
+	return ""
+}
+
+func assertDriveDeleteTaskSucceeded(t *testing.T, ctx context.Context, taskID string) {
+	t.Helper()
 
 	taskResult, err := clie2e.RunCmd(ctx, clie2e.Request{
 		Args:      []string{"drive", "+task_result", "--scenario", "task_check", "--task-id", taskID},
 		DefaultAs: "bot",
 	})
 	require.NoError(t, err)
-	taskResult.AssertExitCode(t, 0)
+	require.NotNil(t, taskResult)
+	// Fatal exit-code gate first: the non-fatal assert flavor would cascade
+	// into misleading empty-stdout failures, exactly what this fix removes.
+	require.Equal(t, 0, taskResult.ExitCode, "drive +task_result failed\nstdout:\n%s\nstderr:\n%s", taskResult.Stdout, taskResult.Stderr)
 	taskResult.AssertStdoutStatus(t, true)
 	require.Equal(t, taskID, gjson.Get(taskResult.Stdout, "data.task_id").String(), "stdout:\n%s", taskResult.Stdout)
-	require.False(t, gjson.Get(taskResult.Stdout, "data.failed").Bool(), "stdout:\n%s", taskResult.Stdout)
-
-	require.NoError(t, waitDriveResourceDeleted(ctx, token, docType, "bot", driveDeleteVisibilityWait))
-	return taskID
+	// gjson returns false for an absent field too, so require presence or a
+	// malformed task envelope would pass validation.
+	failedField := gjson.Get(taskResult.Stdout, "data.failed")
+	require.True(t, failedField.Exists(), "task result must report data.failed\nstdout:\n%s", taskResult.Stdout)
+	require.False(t, failedField.Bool(), "stdout:\n%s", taskResult.Stdout)
+}
+
+// isTransientDriveDeleteFailure reports whether a failed drive +delete carries
+// the one backend error this workflow tolerates: the async delete task
+// transiently reporting a terminal "fail" state (observed as flake in CI; the
+// resource is usually deleted regardless). Everything else — crashes, protocol
+// regressions, auth or parameter errors — stays fatal.
+func isTransientDriveDeleteFailure(result *clie2e.Result) bool {
+	if result == nil {
+		return false
+	}
+	for _, raw := range []string{result.Stderr, result.Stdout} {
+		idx := strings.Index(raw, "{")
+		if idx < 0 {
+			continue
+		}
+		payload := raw[idx:]
+		if !gjson.Valid(payload) {
+			continue
+		}
+		errObj := gjson.Get(payload, "error")
+		if errObj.Get("type").String() == "api" &&
+			errObj.Get("subtype").String() == "server_error" &&
+			errObj.Get("message").String() == "drive task failed" {
+			return true
+		}
+	}
+	return false
 }

tests/cli_e2e/drive/drive_duplicate_sync_workflow_test.go

@@ -17,6 +17,7 @@ import (
 )
 
 func TestDrive_DuplicateRemoteWorkflow(t *testing.T) {
+	clie2e.SkipWithoutTenantAccessToken(t)
 	parentT := t
 	ctx, cancel := context.WithTimeout(context.Background(), 4*time.Minute)
 	t.Cleanup(cancel)

tests/cli_e2e/drive/drive_files_workflow_test.go

@@ -13,6 +13,8 @@ import (
 
 // TestDrive_FilesCreateFolderWorkflow tests the files create_folder resource command.
 func TestDrive_FilesCreateFolderWorkflow(t *testing.T) {
+	clie2e.SkipWithoutTenantAccessToken(t)
+
 	parentT := t
 	ctx, cancel := context.WithTimeout(context.Background(), 2*time.Minute)
 	t.Cleanup(cancel)