← ClaudeAtlas

auditlisted

Master codebase audit for claude-annotate. Runs all focused sub-audits — `/audit-engine-boundary`, `/audit-http-surface`, `/audit-plugin-manifest`, `/audit-docs-truth`, `/audit-code-health` — then presents one unified actionable report. Use when the user says "/audit", "audit the codebase", "run a full audit", "full sweep", or wants everything checked instead of one targeted sub-audit.
petmakris/claude-annotate · ★ 0 · Code & Development · score 75
Install: claude install-skill petmakris/claude-annotate
# /audit — master orchestrator The user runs this for the **full sweep**. This skill owns no checks itself. It dispatches the focused sub-audits and stitches their reports into one delivery. Each sub-audit is independently invocable — when the user already knows where the drift lives, run that sub-audit directly instead. This master exists for the "check everything" case. ## The audit contract (binding on every sub-audit) An audit is only useful if its output is **actionable without hand-waving**. Every item any sub-audit returns must be **exactly one of two kinds**: - **Violation** — objectively wrong against a written rule. A Violation must hold **100% of the time**: if a knowledgeable reader could reasonably say "that's fine because…", it is **not** a Violation and must not be reported as one. **A false positive is a bug in the sub-audit, not a nuance to explain at delivery.** When a reported item turns out to be fine, the fix is to **harden that sub-audit's SKILL.md** (add the case to its allowlist / tighten its check) so it never fires again — not to footnote it. - **Decision** — a genuine either/or that needs the user's judgment. Still actionable — the user acts by deciding — but it lives in the **Optional / Decision** bucket, clearly labeled as a question, never mixed into the Violation buckets. Two corollaries the sub-audits must honor: 1. **Check against a source of truth, not a regex hunch.** 2. **Build the file universe from `git ls-files`** so nothing is str