What problem does it solve? Teams need to know whether an existing repository faithfully implements a given specification, but manual review is slow and self-review by the same AI that wrote the code is unreliable. This Skill audits a repo against a spec — supplied as free text or a file path — and produces a verifiable report of met, gap, and unverified criteria, all without modifying the code under audit. ## Core Features & Use Cases - Spec-to-criteria derivation: Converts a specification into numbered, observable criteria, each citing its source, and gates the run on user approval of that list. - Read-only multi-agent audit: Fans out read-only auditor agents per criteria cluster, runs a refutation pass on a different model for every claimed gap, and has a scribe write the final report. - Tamper-evident verification: Records a hash of the approved criteria and verifies the report cites that same hash, so softened criteria are detectable. - Use Case: Point it at a repo that never used the pipeline and a spec file like docs/specs/billing-api.md; it writes docs/audit/<slug>/ with criteria.md and report.md, listing each gap as an acceptance criterion ready for planning. ## Quick Start Ask the AI to audit this repository against the spec at docs/specs/billing-api.md and report where the implementation falls short.
