What problem does it solve? Loading a rule into an AI agent's context does not guarantee the rule was followed. This Skill closes that gap by reviewing the actual diff produced by a completed task against every rule and skill that was active during the session, making violations visible and recoverable instead of silent. ## Core Features & Use Cases - Rule-by-rule diff audit: Gathers active contexts, rules, and invoked skills, then checks the commit diff against each one with explicit PASS/VIOLATION/UNCLEAR verdicts and severity ratings. - Resource management scanning: Automatically scans JS/TS diffs for anti-patterns like missing cleanup, unbounded Maps, HMR-unsafe module state, and SDK-per-request instantiation, including verification that fixes use valid patterns. - PR review thread auditing: Fetches GitHub PR review threads via GraphQL to validate HITL tier classification and deferral paths in Copilot review rounds, with a degraded fallback mode. - Self-improving loop: Updates the skill or rule inline when a violation reveals an ambiguous rule, and appends results to a persistent compliance log. - Use Case: After a TDD cycle completes and produces a commit, run the audit to confirm the diff follows surgical-changes and framework rules, flag any violations with severity, and log the result for compliance-rate tracking over time. ## Quick Start Ask the agent to run a compliance audit on the most recent commit against the rules and skills active in this session.
