| name | comment-accuracy-review |
| description | Review comment, JSDoc, TODO, README, and docs-to-code factual consistency, stale or low-value comments, and Chinese comment/variable appropriateness; do not use for general code review, broad documentation rewrites, or auto-editing comments without approval. |
Comment Accuracy Review
Use this skill when ROSE needs a narrow, read-only pass over whether explanatory text still matches the code it describes.
Trigger
- Comments, JSDoc, TODOs, README snippets, or nearby docs may be stale, misleading, redundant, or contradicted by implementation.
- A review asks whether docs-to-code claims, examples, parameters, return values, errors, environment notes, or operational steps are factually accurate.
- Chinese comments, Chinese variable names, or bilingual text need appropriateness checks for clarity, tone, maintainability, and consistency with repository language patterns.
Near Misses
- General correctness, architecture, performance, or security review: use
code-reviewer or the relevant review lane.
- Large documentation authoring or product-facing docs: use
doc-researcher or the repository documentation workflow.
- Translation, copywriting, or style-only polishing without code consistency risk: use a writing/documentation workflow instead.
- Implementing fixes: route to
implementer only after a scoped change is approved.
Required Routing
- Owner lane:
subagent:review when used as a review lane.
- Preferred agents:
code-reviewer for code-adjacent comment accuracy; doc-researcher for README/docs-to-code evidence when available.
- Default mode: read-only; report findings and recommended edits, but do not modify comments automatically.
Boundaries
- Compare text claims to the current repository evidence, not to intent or stale plans.
- Flag comments that restate obvious code, describe old behavior, overpromise guarantees, hide uncertainty, or use TODOs as unmanaged task trackers.
- Preserve useful intent comments that explain non-obvious constraints, trade-offs, invariants, data formats, or external behavior.
- Treat generated, vendored, third-party, or upstream-copied comments as out of scope unless the task explicitly includes them.
Verification
- Cite file paths and line anchors for both the questioned text and the code evidence that confirms or contradicts it.
- Classify each finding as inaccurate, stale, unverifiable, redundant, language/terminology concern, or acceptable.
- If edits are later approved, require focused diff inspection and any relevant docs/build/typecheck command for the touched surface.