alchaincyf/darwin-skill — reverse-engineered prompt

Build me a Claude Code compatible skill called Darwin that helps improve other agent skills automatically.

I want it to take a SKILL.md, score it, find the weakest area, make one focused improvement, test it, then only keep the change if the score actually goes up. If the change makes things worse, it should roll back cleanly. It should use separate reviewers so the same AI is not both editing and judging. It should pause at important moments and show me the score change and diff before continuing.

Please include a clear scoring rubric with 9 dimensions, including real output quality, clear instructions, known failure patterns, and a blacklist for dangerous actions. Add sample test prompts, simple docs, and a nice showcase page that explains the loop visually. Make it feel polished and easy to install as an agent skill. Look up current skill format docs online if you need to.