#304304th of 389 in Skills, plugins and prompts
darwin-skill
Agent skill that scores and improves other SKILL.md files, reverting regressions
- Stars
- 6.9k
- License
- MIT
- Last commit
- Sep 2026
- Language
- HTML
1,423 on 20 Apr 20266,888 on 9 Nov 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
Darwin Skill is a SKILL.md that runs an optimization loop over other agent skills. It scores each skill on a 9-dimension, 100-point rubric using 2 independent sub-agent judges, edits one dimension per round, and commits. It uses git revert if the score does not rise. It pauses for human confirmation between phases. Install with npx skills add.
Who it is for: Developers maintaining many agent skills who want measurable improvements
Strengths
- Keep-or-revert ratchet via git commits means a round never leaves a lower-scoring skill
- Fresh independent judges each round to limit self-grading bias
- Forced human checkpoints show diff and score change before each next round
- Works with agent tools that read SKILL.md, such as Claude Code and Codex
Weaknesses
- Scoring is done by LLM judges, so results depend on the model and can vary
- Main README is Chinese; the English version is in a separate file
- Needs a clean git repo; local skill edits must be committed or stashed first
- Reported gains come from the author's own skills, not an independent benchmark
What it needs
- no GPU
- Needs git, npx skills CLI, agent host that supports SKILL.md
- Models: Claude Code, Codex, OpenClaw, Trae, CodeBuddy
Quick answers
Is darwin-skill open source?
Yes. Its code is on GitHub under the MIT license.
Also in Skills, plugins and prompts
See all 389| Rank | Project | Score |
|---|---|---|
| 1 | claude-memPersistent cross-session memory plugin for Claude Code and other coding agents | 97 out of 100 |
| 2 | impeccableDesign skill with 24 slash commands for AI coding agents' frontend output | 97 out of 100 |
| 3 | gstackMarkdown slash-command skills that give Claude Code planning, review, QA and ship roles | 96 out of 100 |
| #4 | rufloAgent harness adding swarms, memory and hooks to Claude Code and Codex | 96 out of 100 |
| #5 | cavemanShortens coding-agent replies and compresses tool output before the model reads it | 95 out of 100 |