canonical-audit - SKILL.md Agent Skill

name: canonical-audit description: Run a full-codebase canonical-pattern audit and write a markdown report to .canonical-audit/. Use when the user asks for a canonical-style scan, a #7458 progress check, or a periodic audit between cron runs.

/canonical-audit

Runs the canonical-check Python and infra runners in --all mode and writes both markdown and JSON sidecars to .canonical-audit/ (gitignored). Then summarizes the report inline.

Steps

Verify the runners exist:

test -f tools/lint/canonical_check.py && test -f tools/lint/canonical_check_infra.py || echo "ERROR: canonical-check not installed"

Run the audit via the Make target (also creates .canonical-audit/ if missing):
```
make canonical-audit
```
Read the most recent markdown reports:
```
ls -1t .canonical-audit/*.md | head -3
```
Open each and surface the Summary table to the user.
If any rule has BLOCK-severity violations, also surface the top 5 offending files for that rule.
Do NOT auto-file GitHub issues. Report findings inline. Filing is a separate user-driven step.

Notes

Frontend audit (canonical-check-fe) does not yet support --all mode (deferred to Wave 3). It is skipped here.
Reports go to .canonical-audit/ locally and to GitHub Actions artifacts on the weekly cron run. Neither is committed.