darwin-skill
达尔文.skill —— 一个让你的Skill无限进化的系统：评估→改进→测试→保留或回滚 | Autoresearch-inspired autonomous skill optimization for Claude Code. Evaluate, improve, test, keep or revert.
Darwin-skill is an autonomous agent skill optimizer that continuously improves SKILL.md files through evaluation, testing, and selective retention of improvements. Inspired by Andrej Karpathy's autoresearch and incorporating Microsoft's SkillLens and SkillOpt research, it evaluates skills across nine dimensions including failure mechanism encoding, actionable specificity, and high-risk action blacklists. The system uses a ratchet mechanism that only keeps measurable improvements and automatically reverts changes that don't enhance performance, with human-in-the-loop checkpoints at critical phases to ensure quality control.
Key Features
Use Cases
- 01Systematically improving a library of 60+ agent skills when manual maintenance becomes impractical
- 02Eliminating vague language and ambiguous instructions from existing skills to improve agent execution reliability
- 03Encoding known failure modes directly into skills to prevent recurring errors
- 04Testing skill improvements against real prompts to validate effectiveness beyond structural correctness
- 05Maintaining skill quality over time by automatically reverting performance regressions
- 06Optimizing Claude Code, Codex, OpenClaw, Trae, and CodeBuddy compatible SKILL.md files
Related Skills
View moresuperpowers
An agentic skills framework & software development methodology that works.
skills
Skills for Real Engineers. Straight from my .agents directory.
skills
Public repository for Agent Skills
ponytail
Makes your AI agent think like the laziest senior dev in the room. The best code is the code you never wrote.
darwin-skill — FAQ
What is darwin-skill and what does it do?+
Darwin-skill is an autonomous agent skill optimizer that evaluates your existing SKILL.md files across nine quality dimensions, proposes targeted improvements, tests them, and only keeps changes that measurably improve performance. It uses a ratchet mechanism inspired by Andrej Karpathy's autoresearch to ensure skills only get better over time.
How do I install darwin-skill?+
Run 'npx skills add alchaincyf/darwin-skill' to install. Alternatively, download the zip package from the provided R2 URL and extract the SKILL.md file to ~/.claude/skills/darwin-skill/ directory. The skill works with any agent tool that supports the SKILL.md format.
Which AI clients and agents does darwin-skill work with?+
Darwin-skill is compatible with Claude Code, Codex, OpenClaw, Trae, CodeBuddy, and any other agent tool that supports the SKILL.md skill format. Once installed, you can activate it by saying 'optimize all skills' or 'optimize [specific skill name]' to your agent.
Does darwin-skill require API keys or special prerequisites?+
No API keys are required. You must run the optimizer inside a Git repository and ensure your local skill modifications are committed or stashed first, so darwin-skill can cleanly preserve or revert experimental changes. The skill itself manages Git operations for version control.
Is darwin-skill free to use?+
Yes, darwin-skill is open-source software released under the MIT License and is free to use for any purpose.
What makes darwin-skill different from SkillOpt?+
While darwin-skill integrates the validation-gated framework from Microsoft's SkillOpt research, it differs by enforcing human-in-the-loop checkpoints at critical phases. SkillOpt is fully autonomous, but darwin-skill pauses for human review after baseline evaluation and single-dimension optimization because skill quality requires nuanced human judgment beyond automated metrics.
How do I install darwin-skill?+
Open the source repository on GitHub and follow its README. darwin-skill is a skill — MCP Agents Market links you directly to the official repo.
Is darwin-skill free?+
darwin-skill is an open-source project hosted on GitHub. Check the repository for its license and any usage requirements.