What problem does it solve? Agent directive files accumulate contradictions, stale rules, redundant guidance, and broken routing over time, silently degrading how AI agents behave. This Skill audits directive systems against explicit dimensions so you can find and fix authority conflicts, unsafe gaps, and drift before they cause failures. ## Core Features & Use Cases - Read-only audit mode: Traverses every in-scope directive file, router, projection, and consumed entrypoint, classifying each material rule as KEEP, CUT, REWRITE, MOVE, SPLIT, or ENFORCE with severity and conflict fields. - Gated housekeeping mode: Applies selected findings only after explicit approval, with canonical-source resolution, symlink rejection, atomic writes, recoverable preimages, and partial-failure rollback rules. - Parity and behavior verification: Claims source-to-consumed parity only from target-exclusive comparison and consumed behavior only from discriminating controls, avoiding unsupported claims. - Use Case: Your CLAUDE.md and routed conditional rule files have grown for a year. Run the audit to surface contradictions, stale project-specific rules, and missing fallback behavior, then selectively housekeep the findings. ## Quick Start Ask the agent to audit my directive files for authority conflicts, stale rules, and routing gaps using /directives-audit.
