What problem does it solve? Third-party audit protocols for Claude Code setups produce scores and findings that reflect the protocol author's preferences, not your repository's actual needs. This Skill runs such protocols faithfully, then filters each finding through a multi-lens discussion review so only genuinely applicable improvements become Issues, implementations, and merged PRs. ## Core Features & Use Cases - Faithful protocol execution: Fetches the audit protocol fresh on every run, executes all phases without modification, and records before/after scorecards using the same protocol version. - Discussion-based triage: Routes every finding through a 4-lens adversarial review (spec compliance, design coherence, security, cost-benefit) that returns adopt/defer/reject verdicts instead of blindly chasing score improvements. - Autonomous follow-through: Converts adopted findings into GitHub Issues via MCP tools, implements them on branches, creates and merges PRs, then re-audits and reports the before/after diff including improvements the score cannot capture. - Use Case: On a monthly schedule, the Skill fetches the latest claude-code-ultimate-guide audit prompt, audits your repository setup, debates the 12 findings, adopts 4 real defects, rejects 5 author-specific conventions, merges the fixes, and reports that sandboxing now actually works even though the score barely moved. ## Quick Start Ask the AI to run the audit-runner skill to execute the setup audit protocol, review the findings, implement the adopted ones, and re-audit for a before/after report.
