Глубокий аудит репозитория с автоисправлением

general

deep-audit is a Claude Code skill that performs a comprehensive consistency audit of the entire repository infrastructure by running mechanical checks followed by four parallel specialist agents to detect factual errors, code bugs, count mismatches, and cross-document inconsistencies, then fixes all findings in a loop until the repository is clean. Phase 0 runs `check-skill-integrity.py` to catch `allowed-tools`–to–tool-invocation parity issues, broken markdown anchors, undocumented flags, and rule `paths:` mismatches before any agent is spawned. The four agents cover guide content accuracy, executable code quality across all hooks and scripts (error handling, exit codes, docstring-to-implementation parity), skills and rules frontmatter validity, and catalog-wide metric consistency. It is designed for use after broad changes, before releases, or whenever a thorough cross-repository consistency check is needed.