Claude Perfectionist audits your Claude Code instruction harness — CLAUDE.md, rules, hooks, settings, MCP config — and finds contradictions, missing verification, stale commands, and context bloat you didn't know you had. Key features: - Verdict-first reports with adaptive templates (audit, bootstrap, contradiction debug) - Evidence-grounded findings with honest confidence levels — no fake certainty - Fix-first remediation — resolves contradictions before deleting anything - Grades with human meaning (A — production-ready through F — actively misleading) - Pure markdown, zero dependencies — works in Claude Code and Claude.ai