Darwin Skill

Install

npx skills add alchaincyf/darwin-skill

This is third-party code. Review the repository files before installing.

What it does

Darwin Skill carries Karpathy's autoresearch loop from model training over to skill maintenance. It scores a SKILL.md on nine criteria: structure, workflow clarity, failure handling, checkpoints, specificity, architecture, results on test prompts and a list of forbidden actions. It then picks the weakest weighted area, edits only that, commits to git and has two independent subagents re-score it. If the result does not improve, the edit is undone with git revert, and the loop stops when gains get small. After each skill the agent waits for user confirmation.

Who it is for. For people who write and maintain dozens of their own skills and want to improve them against measurable criteria.

Example request

Score and improve the release-notes skill, showing the diff and score change after each edit

Limitations

SKILL.md and the main README are in Chinese; an English README exists. It must run inside a git repository, and local skill changes should be committed first. Subagent judging and test prompt runs consume many tokens. Some criteria reflect the author's style preferences.

How to disable. Remove the darwin-skill folder from the skills directory, for example ~/.claude/skills/darwin-skill.

Security check

• Edits skill files and makes git commits and reverts
• Spawns subagents and runs test prompts

README in short

The README explains the link to Karpathy's autoresearch and the 2.0 changes based on two Microsoft Research papers on skill quality. It covers five principles, nine scoring criteria, a five-phase optimization loop and the ratchet that only keeps improvements. Install with one npx skills command or a ZIP. MIT licensed.

SKILL.md

---
name: darwin-skill
description: "Darwin Skill 2.0 (达尔文.skill 2.0): autonomous skill optimizer, v2.0 integrates Microsoft Research SkillLens (arXiv 2605.23899) 9-dim rubric + SkillOpt (arXiv 2605.23904) validation-gated design + human-in-the-loop checkpoints. Evaluates SKILL.md files using a 9-dimension rubric (structure + effectiveness + meta-skill blacklists), runs hill-climbing with git version control, spawns independent judge agents for blind evaluation, validates improvements through test prompts with auto-break on diminishing returns, and generates visual result cards. Use when user mentions "优化skill", "skill评分", "自动优化", "auto optimize", "skill质量检查", "达尔文", "darwin", "帮我改改skill", "skill怎么样", "提升skill质量", "skill review", "skill打分"."
---

# Darwin Skill 2.0

> **v2.0 · 2026-05-28** — 吸收 Microsoft Research SkillLens（arXiv 2605.23899）的 9 维评分药方 + SkillOpt（arXiv 2605.23904）的 validation-gated 验证机制 + human in the loop 三层守关。
>
> 借鉴 Karpathy autoresearch 的自主实验循环，对 skills 进行持续优化。
> 核心理念：**评估 → 改进 → 实测验证 → 人类确认 → 保留或回滚 → 生成成果卡片**
> GitHub: https://github.com/alchaincyf/darwin-skill

---

## 设计哲学

autoresearch 的精髓：
1. **单一可编辑资产** — 每次只改一个 SKILL.md
2. **双重评估** — 结构评分（静态分析）+ 效果验证（跑测试看输出）
3. **棘轮机制** — 只保留改进，自动回滚退步
4. **独立评分** — 评分用子agent，避免「自己改自己评」的偏差
5. **人在回路** — 每个skill优化完后暂停，用户确认再继续

与纯结构审查的区别：不只看 SKILL.md 写得规不规范，更看改完后**实际跑出来的效果是否更好**。

Related