What problem does it solve? Whole-codebase reviews are slow, shallow, and easy to get wrong: hidden bugs like silent error swallowing, registry key mismatches, and frontend-backend type drift slip through manual review, and findings from one audit are never compared against the last. This Skill runs an adaptive, multi-agent deep audit that verifies Critical/High findings adversarially and tracks issues across audits via a persistent ledger. ## Core Features & Use Cases - Adaptive parallel analysis: Detects the stack (TypeScript, Python, Rust, Go, full-stack) and dispatches 2-5 read-only finder agents across dimensions like frontend-backend contract, data integrity, error handling/security, architecture, and config/persistence, with optional test-quality and concurrency dimensions. - Adversarial verification and ledger diffing: Critical/High findings must survive a refutation pass before entering the report, and every finding is classified as new, still-open, or resolved against the previous audit's .audit/findings.json ledger. - Deterministic dependency audits: Runs cargo audit, npm audit, pip-audit, or govulncheck from the target directory and classifies CVEs by reachability. - Use Case: Ask for a full audit of a FastAPI + React project before a release; receive a severity-sorted Chinese-language report with a phased repair roadmap, a delta versus the previous audit, and a quick mode for fast health checks. ## Quick Start Ask the agent to run a full codebase audit on your project directory, or say "quick 快速体检" for a two-agent fast health check.
