What problem does it solve? A shipped skill catalogue decays silently: claims stop being true, references stop resolving, and instructions stop being executable, and a single reviewer cannot reliably catch this across dozens of skills. This Skill runs a structured audit that separates deterministic checks from judgement calls and isolates reviewers from contaminated context. ## Core Features & Use Cases - Clean-room skill sweep: Copies the whole plugin into an isolated room and runs one reviewer per skill in parallel, writing a report per skill with verbatim-quote findings classified as WRONG, DANGLING, UNEXECUTABLE, or STALE. - Script audit mode: Reviews shipped Python and JavaScript (hooks, shims, skill scripts) with an evidence contract requiring runnable reproductions or traced line numbers, plus a deterministic pre-pass that separates settled facts from leads. - Triage guidance: Pairs every finding class with its usual false positive and ranks defects by what a reader does with them. - Use Case: Before a release touching many skills, run the sweep to catch a shipped example that no longer parses, a sibling reference that no longer resolves, or a documented flag the script no longer accepts. ## Quick Start Ask the AI to audit the shipped skill catalogue with meta-skill-audit, walling off recall first and running scripts/audit_skills.py against a clean room outside the knowledge tree.
