/CLAUDE

Manages eval-driven development workflow with subcommands to define, check, report, and list evals. Tracks capability and regression criteria with pass rates.

Runs meta-evaluation Python script on evaluation skills, checking recursive quality, TOC, tests, and docs; produces severity reports, pass rates, and action items.

Evaluates Claude Skills for quality, frontmatter compliance, token efficiency, activation reliability, and security. Generates scores, breakdowns, prioritized improvements, and roadmaps for all or specific skills.

Audits the repository's agent harness setup across tool coverage, context efficiency, quality gates, and more, producing an overall score out of 70, category breakdowns, findings, top 3 actions with file paths, and suggested ECC skills. Supports optional scopes and JSON format.

Evaluates a plugin or skill directory for quality via static analysis and optional LLM judging, producing overall score with badge, dimension breakdowns, anti-patterns, and recommendations.

Runs three-stage verification pipeline to evaluate execution, checking for correctness, safety, and performance in the current context.