Skill A/B Optimizer

Automates A/B testing and continuous improvement of Claude Code skills. Instead of guessing whether a skill modification actually helps, this skill provides hard quantitative data — and uses the Auto Research ML loop to automatically approve only changes that measurably improve outcomes.

Reference Files:

• workflow.md — Full A/B testing lifecycle and phase breakdown
• metrics.md — Metric definitions, scoring rubrics, and evaluation criteria
• examples.md — Practical examples and real optimization runs

Core Problem This Solves

When creating or modifying a skill you have no quantitative feedback:

• Does the new prompt actually produce better results?
• Is the reorganized SKILL.md clearer than the original?
• Did adding an examples file improve accuracy?