README_EN.md 5.0 KB
The Core Loop
Evaluate → Improve → Test → Human Confirm → Keep or Revert. Repeat.
Why This Exists
When you have 10 skills, you can maintain them by hand. When you have 60+, you need a system.
Traditional skill review is purely structural: does the frontmatter look right? Are the steps numbered? Do the file paths exist? But a perfectly formatted skill can still produce terrible output.
darwin.skill evaluates both structure and real-world effectiveness, then keeps only the changes that actually improve things.
From autoresearch to Skill Optimization
This project maps Karpathy's autoresearch directly onto skill optimization:
| autoresearch | darwin.skill | Why |
|---|---|---|
program.md |
This SKILL.md | Defines evaluation criteria and constraints |
train.py |
Each target SKILL.md | The single editable asset per experiment |
val_bpb |
8-dimension weighted score (max 100) | Quantifiable optimization target |
git ratchet |
keep / revert mechanism | Only improving commits survive |
test set |
test-prompts.json | Validates whether improvements are real |
| Fully autonomous | Human in the loop | Skill quality is more subjective than loss |
The key difference: autoresearch is fully autonomous (loss is just a number). Skill quality sometimes needs human judgment. So darwin.skill pauses after each skill's optimization cycle, shows you the diff and score delta, and waits for your confirmation.
Five Core Principles
| # | Principle | Details |
|---|---|---|
| 01 | Single editable asset | One SKILL.md per experiment. One change, one measurement, one decision |
| 02 | Dual evaluation | Structure scoring (static analysis) + effectiveness scoring (live test execution) |
| 03 | Ratchet mechanism | Score can only go up. Regressions are auto-reverted |
| 04 | Independent scoring | The agent that edits is never the agent that scores |
| 05 | Human in the loop | System pauses after each skill. You review, then continue |
8-Dimension Evaluation Rubric
Total: 100 points. Structure (60) + Effectiveness (40).
Live test performance has the highest weight (25 points). A beautifully written skill that produces bad output is still a bad skill.
The Optimization Cycle
Five phases. Only one is the core.
Phase 2 (the heart):
- Find the lowest-scoring dimension
- Generate one targeted improvement
- Edit SKILL.md, git commit
- Independent sub-agent re-scores
- Score up → keep. Score down → git revert
- Pause. Show diff + score delta. Wait for human confirmation
The Ratchet
Scores can only go up. Failed experiments are cleanly reverted. No regressions accumulate over time.
Round 2 scored 75, below the current best of 78. Auto-reverted. Effective baseline stays at 78. Subsequent improvements build from 78, not 75.
Quick Start
npx skills add alchaincyf/darwin-skill
After installation, tell Claude Code: "optimize all skills" or "optimize [skill-name]".
Can't access GitHub? Download the zip: darwin-skill.zip. Extract and place SKILL.md in ~/.claude/skills/darwin-skill/.
Design Inspiration
Directly inspired by Andrej Karpathy's autoresearch.
The core mechanism is identical: keep only measurable improvements, revert everything else.
About the Author
| 🌐 Website | bookai.top · huasheng.ai |
| 𝕏 Twitter | @AlchainHust |
| 📺 Bilibili | 花叔 |
| ▶️ YouTube | @Alchain |
| 📕 Xiaohongshu | 花叔 |
| Search "花叔" |
License
MIT
**Darwin** makes them evolve.
*Keep only improvements. Time is on your side.*
MIT License © [花叔 Huashu](https://github.com/alchaincyf)