Review Docs
Overview
After code changes, verify that documentation still matches reality. Stale docs are worse than no docs.
Steps
1. Identify Changed Code
Review the diff or recent commits to understand what changed — new features, renamed APIs, changed behavior, removed functionality.
2. Find Affected Docs
Search for documentation that references the changed code:
bash
rg "changed_thing" docs/ README.md *.md
Include docstrings, inline comments, and config files — not just standalone docs.
3. Review Categories
Check each affected document against these lenses:
- •Codebase match — Does it describe current behavior? File paths correct?
- •Cross-doc consistency — Do different docs agree with each other?
- •API & interface assumptions — Are assumptions about APIs, tools, external services still correct?
- •Examples — Do code examples still work?
- •Project standards — Does it align with CLAUDE.md and documented conventions?
4. Report (severity-weighted)
code
## Doc Review: [scope] ### Summary X docs checked | Y issues (Z critical, W high) ### CRITICAL - **[Issue]** `file:line` — [what's wrong] → [fix] ### HIGH - **[Issue]** `file:line` — [what's wrong] → [fix] ### MEDIUM / LOW - ... ### Missing Coverage - New feature X has no documentation ### Clean - `docs/setup.md` — still accurate
5. Fix (after user reviews findings)
Review phase is read-only. Present findings first, then fix only after user says go.
Red Flags
- •Assuming docs are fine because tests pass
- •Only checking README and ignoring inline docs/comments
- •Skipping docstrings in changed functions
- •Updating code without searching for affected docs
- •Fixing docs before presenting findings