AI Inspiration Worlddarwin-skill
Scores, experiments on and verifies your existing Skills, keeping changes that help and rolling back the rest.
- Ecosystem
- Agent Skills
- Skill category
- Coding
- Creator / organization
- 花叔（alchaincyf）
What it is
darwin-skill is a Skill optimizer by Huashu (alchaincyf), named after the idea of evolution. After installing, say "optimize all skills" or name one Skill. It first scores the current Skill as a baseline, then proposes changes and runs experiments. Each step has a human checkpoint: changes stay only if they pass verification, and anything worse is rolled back. What sets it apart is treating Skill edits as a controlled experiment, with a baseline and verification. It works with Claude Code, Codex, OpenClaw, Trae, CodeBuddy and other tools that read SKILL.md.
Who it's for:People with a growing pile of Skills who want to tidy and improve them regularly.
Before you start: Install it with the prompt or method in the README. No background service is needed. It rewrites the Skills you choose, so back them up first and read each checkpoint before approving. Model costs follow your host. Docs are mainly in Chinese. License: MIT.
Based on the project's public documentation. Not tested by AI Inspiration World.