Replay the seeded IM failure cases. Most already give an agent enough signal to recover; deterministic search coverage requires stable fixtures, while the --head/--tail conflict lacked a concrete next action. Add the missing hint, record the inventory, and fold the search fixture prerequisite and dry-run coverage into coverage.md.
* docs(base): clarify form and file operation routing
* docs: clarify complete base role table rules
* docs: clarify base advanced permission status
* docs: clarify base form field lifecycle
* docs: guide base form question creation
* fix(base): address form dry-run review findings
* docs(base): add complete editable role example
* fix(base): validate form question create inputs
* feat(apps): add cache debug commands (+cache-get/-delete/-clear)
Add three apps-domain cache debug shortcuts for inspecting/clearing an app's
runtime cache:
- +cache-get: read a business key's value + metadata (hit/miss)
- +cache-delete: delete a single key (idempotent, write)
- +cache-clear: clear all cache in an environment (high-risk-write, --yes)
value renders raw on --format json, deserialized on --format pretty;
value_size_bytes is computed CLI-side; --environment auto-selects the branch
when omitted. Includes unit tests (hit/miss/dry-run/confirmation) and the
lark-apps cache skill reference.
* fix(apps): normalize cache numeric output fields and tidy comments
Follow-up hardening for the cache debug commands (+cache-get/-delete/-clear):
- Normalize ttl_ms / deleted_key_count via a new cacheInt() helper so
--format json emits a stable JSON number (or null) regardless of whether
the server sends the value as a number or a string. Aligns with the
repo convention that numeric wire fields may arrive as strings; previously
these were passed through raw, leaving the output type at the server's mercy.
- Add unit tests locking the string-wire -> JSON number contract for both
cache-get ttl_ms and cache-delete deleted_key_count.
- Tidy two comments: soften cacheBool's speculative "historical wire form"
claim to a defensive-tolerance note, and drop implementation jargon from
cache-delete's risk-level rationale.
* feat(drive): support Miaoda apps in permission shortcuts
Extend Drive permission shortcuts to accept Miaoda page URLs and the apps resource type while keeping each endpoint's accepted resource contract explicit.
Key features:
- Infer apps from /page/ URLs and accept explicit --type=apps in +apply-permission, +member-add, +member-list, and +permission-get-setting
- Decouple secure-label target parsing so expanding apply-permission does not widen secure-label support
- Align skill guidance and unit/dry-run coverage with the new resource type
* test(drive): cover apps permission target validation
Add focused coverage for Miaoda apps target handling across apply-permission and secure-label boundaries.
Exercise malformed page URLs, explicit apps bare tokens, typed validation errors, and command-level rejection so future resource-type changes cannot silently widen unsupported secure-label behavior.
* fix(drive): parse permission markers from URL paths
Keep drive +apply-permission resource inference aligned with URL component boundaries. Parse and validate URL inputs before extracting tokens so query strings and fragments cannot redirect permission requests to a different resource.
Key fixes:
- Match document and apps markers only against the parsed URL path
- Reject malformed URLs with a typed --token validation error
- Cover /page/ markers found only in query strings or fragments
* docs(skills): redact Miaoda page token example
Replace the concrete Miaoda page token with a representative pagcn placeholder. This keeps the token shape recognizable while avoiding exposure of a real resource identifier in the skill documentation.
* fix(drive): harden permission target resolution
Make Drive shortcut targets unambiguous before they reach read or write API paths. URL inputs now bind to a recognized root path and a single validated token segment, preventing encoded separators, dot segments, and type conflicts from silently changing the addressed resource.
Key fixes:
- Reject non-root URLs, dot/traversal tokens, and URL/type conflicts for secure-label and permission-apply writes
- Keep permission-setting URL parsing and pretty output reversible for every supported command-local resource kind
- Add unit and dry-run E2E regressions plus aligned permission-apply guidance
Promote the recurring-event rule to a pre-routing gate so it is read
before the specific operation flow, and require confirming the scope
(this event / all / this-and-following) when the user is ambiguous
instead of defaulting to this-event-only. Removes the redundant and
conflicting "edit existing event" row that hard-coded the single-
instance default.
Add comment-domain shortcuts: +batch-query-comments, +resolve-comment,
+restore-comment, +add-reply, +list-replies, +update-reply, +delete-reply
and +react-reply, sharing one target resolver with per-endpoint file_type
sets.
Flatten the comment reference docs by dropping the comments-guide routing
layer and folding its cross-command knowledge into the command refs:
comment-card model, comment/reply/interaction counting and sorting rules
into lark-drive-list-comments.md; the --solved-status prerequisite into
lark-drive-restore-comment.md; the apps exception into
lark-drive-add-comment.md. Comment intents now route straight from the
drive SKILL.md Shortcuts table to each command ref.
Cover the new shortcuts with unit tests, dry-run e2e and live workflow
e2e behind LARK_DRIVE_MD_COMMENT_E2E=1, and register them in
tests/cli_e2e/drive/coverage.md.
- add XSD-backed SXSD validation for tags, attributes, structure, scalar values, and namespaces
- preserve supported server-filled fields and readback namespace compatibility
- isolate SXSD failures by slide so valid slides continue through layout checks
- improve actionable lint diagnostics and suppress duplicate errors
- add regression coverage for schema validation and Slides readback cases
Validated with unit tests and real Slides create/readback round trips.
