| name | audit-docs |
| description | Use when asked to audit docs, review docs accuracy, or validate documentation claims against actual code, commands, APIs, configs, and diagrams. |
| metadata | {"short-description":"Verify every doc claim against the code"} |
Docs Audit
Docs lie by going stale. Verify claims against code; never grade prose style.
1) Inventory claims
Walk each doc file and extract every checkable claim:
- Commands and flags (do they run? do the flags exist in the CLI source?)
- File paths and module names referenced (do they exist?)
- API/config examples (do the fields match the current types/schema?)
- Mermaid/architecture diagrams (does each node and edge correspond to a real module and a real dependency?)
- Version numbers, install instructions, env vars.
2) Verify
- Cheap checks first: paths via
ls/glob, flags via --help or the arg-parser source, config fields via the schema/types.
- Runnable examples: actually run them in a scratch dir when safe; otherwise trace the code path and say "traced, not executed".
- Diagrams: grep imports to confirm each drawn edge; list edges that exist in code but are missing from the diagram (coverage), and drawn edges with no code counterpart (drift).
3) Report
Per file: ✅ verified / ❌ wrong (with the correct current value) / ⚠️ unverifiable (why).
- Lead with the wrong ones — those actively mislead.
- End with a coverage note: significant code areas (public commands, exported APIs) that no doc mentions.
Notes:
- Fixing is a separate pass — audit output must stay a faithful report, not a mixed diff.