Documentation Freshness Review
SkillDocs & knowledgeUse when auditing repository documentation for drift, stale statements, broken references, or README mismatches, especially across published docs under `doc/` and key entry surfaces such as `README.md` (English default), `README.zh-CN.md`, `mcp/README.md`, and `mcp/README.zh-CN.md` during periodic maintenance or before preparing corrective PRs.
Available today. Use it from your connected AI after setup.
No other account needed.
Connect ahel once, and every AI you use reads what you have installed.
Then ask your AI: use the Documentation Freshness Review skill
What this skill tells your AI
The instructions your AI receives, as published by tencentcloudbase/cloudbase-ai-toolkit in skills/doc-freshness-review/SKILL.md and read by ahel’s review.
Review whether repository documentation still matches the current code, generated assets, and published entry points.
When to use this skill
Use this skill when you need to:
- Audit published documentation under
doc/for stale or wrong statements - Check whether
README.md,README.zh-CN.md,mcp/README.md, ormcp/README.zh-CN.mddrifted away from current behavior - Verify that linked scripts, files, commands, and workflows still exist and still match the repo
- Run a periodic docs health check before opening corrective PRs
- Separate mechanical doc drift from product ambiguity that still needs human confirmation
Do NOT use for:
- Marketing copy polish without a freshness or correctness question
- API contract review that depends on CloudBase control-plane docs first
- Feature planning or architecture design
- Rewriting large docs from scratch when the main need is a targeted drift audit
Workflow
Phase 1 — Scope the surfaces
- Read
references/review-scope.mdfirst. - Build the audit surface in two groups:
- published documentation under
doc/ - README surfaces such as
README.md,README.zh-CN.md, andmcp/README.md
- published documentation under
- Note which code, scripts, generated files, or workflows those docs claim to describe.
Phase 2 — Evidence-based comparison
- Verify whether referenced files, scripts, commands, and workflow names still exist.
- Compare the docs against the current implementation, generated outputs, and source-of-truth files.
- Mark each mismatch as one of:
- stale statement
- missing or stale reference
- outdated command
- drift between published docs and README surfaces
- drift between docs and actual implementation
Phase 3 — Severity and action
- Prefer concrete findings with exact file paths and the smallest useful rewrite direction.
- If the fix is mechanical and low-risk, prepare the doc updates for a focused PR.
- If the docs may be wrong because product behavior is unclear, write a report and issue instead of guessing.
- If the drift originates from a CloudBase API contract problem, hand off to
api-contract-reviewbefore updating the prose.
Phase 4 — Follow-through
- Keep report sections separate for:
- published documentation
- README surfaces
- broken or missing references
- Prefer one focused PR per coherent doc drift batch.
- Route open-PR repair work to
pr-review-fixafter the doc finding is confirmed.
Routing
| Task | Read |
|---|---|
| Review docs and README freshness scope | references/review-scope.md |
| Review CloudBase API contract drift before changing related prose | api-contract-review |
| Fix an already-open PR after confirming the doc drift | pr-review-fix |
Evaluation prompts
Should-trigger
- Audit
doc/and the main README files for stale commands, missing files, and broken references. - Review whether
mcp/README.mdstill matches the current MCP tool behavior and linked files. - Help me prepare a corrective PR for docs that still reference scripts or workflows that no longer exist.
Should-not-trigger
- Review this TypeScript module for runtime validation gaps.
- Compare CloudBase action parameters against the official control-plane docs.
- Write a brand-new tutorial for a feature that does not exist yet.
Minimum self-check
- Did I separately review published documentation and README surfaces?
- Can I point to exact file paths for every stale or broken reference?
- Did I compare the docs against current code, scripts, and generated outputs instead of relying on memory?
- If I recommend a fix, is it a small, reviewable doc batch suitable for a PR?
- If the underlying product behavior was unclear, did I stop at report or issue instead of rewriting confidently?
Signals
- GitHub stars
- 1k
- Forks
- 138
- Last commit
- Sep 2026
Advanced
- Catalog kind
- skill
- Gateway key
doc-freshness-review- Source
- github.com/tencentcloudbase/cloudbase-ai-toolkit