takehiro177

Cost-weighted A/B performance testing for Claude Code skills — one skill, or several combined with interaction (synergy/redundancy/conflict) analysis. Blind LLM-as-judge quality scoring, transcript-sourced token economics, required deepeval cross-check, and a zero-dependency Skill Harness Dashboard: every run also writes a versioned summary.json and refreshes a self-contained reports/dashboard.html governance view (open from file://, no server).