darwin-skill
Darwin Skill (达尔文.skill): autonomous skill optimizer inspired by Karpathy's autoresearch.
Skill by xkzs2007
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/xkzs2007-darwin-skill
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: testing
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
- Runs as
- Node.js package
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/darwin-skill
What's inside
- Skill name
darwin-skill- Description
- Darwin Skill (达尔文.skill): autonomous skill optimizer inspired by Karpathy's autoresearch. Evaluates SKILL.md files using an 8-dimension rubric (structure + effectiveness), runs hill-climbing with git version control, validates improvements through test prompts, 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".
- License
- MIT
- Instructions
- 398 lines in SKILL.md · view on GitHub
Before installing
- Ships executable scripts
Files (33)
- README.md
- README_EN.md
- SKILL.md
- aso-hero.png · assets/
- banner-check.png · assets/
- banner-en-check.png · assets/
- banner-en.svg · assets/
- banner.svg · assets/
- chart-loop-en.html · assets/
- chart-loop-en.png · assets/
- chart-loop.html · assets/
- chart-loop.png · assets/
- chart-phases-en.html · assets/
- chart-phases-en.png · assets/
- chart-phases.html · assets/
- chart-phases.png · assets/
- chart-ratchet-en.html · assets/
- chart-ratchet-en.png · assets/
- chart-ratchet.html · assets/
- chart-ratchet.png · assets/
- chart-rubric-en.html · assets/
- chart-rubric-en.png · assets/
- chart-rubric.html · assets/
- chart-rubric.png · assets/
- hero.gif · assets/
- result-card-example.png · assets/
- index.html · docs/
- screenshot.mjs · scripts/
- showcase.html
- result-card-dark.html · templates/
- result-card-white.html · templates/
- result-card.html · templates/
- result-card.png · templates/
SKILL.md (first lines)
# Darwin Skill
> 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**。
---
## 评估 Rubric(8维度,总分100)
### 结构维度(60分)— 静态分析
| # | 维度 | 权重 | 评分标准 |
|---|------|------|---------|
| 1 | **Frontmatter质量** | 8 | name规范、description包含做什么+何时用+触发词、≤1024字符 |
| 2 | **工作流清晰度** | 15 | 步骤明确可执行、有序号、每步有明确输入/输出 |
| 3 | **边界条件覆盖** | 10 | 处理异常情况、有fallback路径、错误恢复 |
| 4 | **检查点设计** | 7 | 关键决策前有用户确认、防止自主失控 |
| 5 | **指令具体性** | 15 | 不模糊、有具体参数/格式/示例、可直接执行 |
| 6 | **资源整合度** | 5 | references/scripts/assets引用正确、路径可达 |
### 效果维度(40分)— 需要实测
| # | 维度 | 权重 | 评分标准 |
|---|------|------|---------|
| 7 | **整体架构** | 15 | 结构层次清晰、不冗余不遗漏、与花叔生态一致 |
| 8 | **实测表现** | 25 | 用测试prompt跑一遍,输出质量是否符合skill宣称的能力 |
### 评分规则
- 维度1-7:每个维度打 1-10 分,乘以权重得到该维度得分
- 维度8(实测表现):跑2-3个测试prompt,按输出质量打1-10分
- **总分 = Σ(维度分 × 权重) / 10**,满分100
- 改进后总分必须 **严格高于** 改进前才保留
### 关于「实测表现」维度
这是与纯结构评分最大的区别。评分方式:
1. 为每个skill设计2-3个**典型用户prompt**(不是边缘case,是最常见的使用场景)
2. 用子agent执行:一个带skill跑,一个不带skill跑(baseline)
3. 对比输出质量,从以下角度打分:
- 输出是否完成了用户意图?
- 相比不带skill的baseline,质量提升明显吗?
- 有没有skill引入的负面影响(过度冗余、跑偏、格式奇怪)?
如果无法跑子agent(时间/资源限制),可以退化为「干跑验证」:读完skill后模拟一个典型prompt的执行思路,判断流程是否合理。但要在results.tsv中标注 `dry_run`。
---
## 自主优化循环
### Phase 0: 初始化
```
1. 确认优化范围:
- 全部skills → 扫描 .claude/skills/*/SKILL.md
- 指定skills → 用户指定列表
2. 创建 git 分支:auto-optimize/YYYYMMDD-HHMM
3. 初始化 results.tsv(如不存在)
4. 读取现有 results.tsv 了解历史优化记录
```
### Phase 0.5: 测试Prompt设计
在评估之前,为每个skill设计测试prompt。这步很关键——没有测试prompt,「实测表现」维度就打不了分。
```
for each skill:
1. 读取 SKILL.md,理解它做什么
2. 设计2-3个测试prompt,覆盖:
- 最典型的使用场景(happy path)
- 一个稍复杂或有歧义的场景
3. 保存到 skill目录/test-prompts.json:
[
{"id": 1, "prompt": "用户会说的话", "expected": "期望输出的简短描述"},
{"id": 2, "prompt": "...", "expected": "..."}
]
```
展示所有测试prompt给用户,**确认后再进入评估**。测试prompt的质量决定了优化方向是否正确。
### Phase 1: 基线评估(Baseline)
```
for each skill in 优化范围:
# 结构评分(主agent可以做)
1. 读取 SKILL.md 全文
2. 按维度1-7逐项打分(附简短理由)
# 效果评分(用子agent做,独立于主agent)
3. 对每个测试prompt,spawn子agent:
- with_skill: 带着SKILL.md执行测试prompt
- baseline: 不带skill执行同一prompt
4. 对比两组输出,打维度8的分
# 汇总
5. 计算加权总分
6. 记录到 results.tsv
```
**如果子agent不可用**(超时、环境限制),维度8用干跑验证打分,标注 `dry_run`。不要因为跑不了测试就跳过这个维度——哪怕是模拟推演也比完全不看效果好。
基线评估完成后,展示评分卡:
```From xkzs2007/GalaxyOS-desktop, MIT.
More from xkzs2007/GalaxyOS-desktop
- 0.564score
concentrate-forces
xkzs2007
触发:当多个任务同时争夺时间、注意力、算力或预算,必须确定主攻方向并停止分散用力时调用;常见信号包括优先级过多、资源紧张、推进分散、需要决定先做什么。 English: Trigger when limited resources are being split across too many tasks and…
SkillBFreshness: B★ 3Claude Code +9 - 0.564score
criticism-self-criticism
xkzs2007
触发:当一项工作已经完成、进入阶段验收、收到批评反馈,或反复出现同类错误需要系统纠偏时调用;常见信号包括 review、audit、retrospective、quality check、纠错与复盘。 English: Trigger after delivery or at a review checkpoint…
SkillBFreshness: B★ 3Claude Code +9 - 0.564score
arming-thought
xkzs2007
触发:在每次新的顶层对话开始时自动调用,用于建立“实事求是”的总原则,并在明确适用时为后续任务选择下游 skill;如果你是被派遣执行单一具体任务的子 agent,则跳过此 skill。 English: Trigger at the start of each new top-level conversation…
SkillBFreshness: B★ 3Claude Code +9 - 0.564score
contradiction-analysis
xkzs2007
触发:当问题复杂、存在多个冲突因素、优先级不清,或你不知道应该先解决什么时调用;常见信号包括 trade-off、瓶颈、根因不明、主次不清、多个问题互相牵制。 English: Trigger when a problem contains competing forces, unclear priorities…
SkillBFreshness: B★ 3Claude Code +9 - 0.561score
autonomous-tasks
xkzs2007
Self-driven AI worker.
SkillBFreshness: B★ 3Claude Code +9 - 0.560score
brain
xkzs2007
Personal knowledge base for capturing and retrieving information about people, places, restaurants, games, tech, events, media, ideas, and organizations.
SkillBFreshness: B★ 3Claude Code +9 - 0.558score
caveman
xkzs2007
超压缩通信模式。砍掉约 75% 的废话 token,保留全部技术信息。触发词:caveman 模式、简洁回复、省 token。
SkillBFreshness: B★ 3Claude Code +9 - 0.558score
best-minds
xkzs2007
模拟器思维:不问"你怎么看",而是问"世界上谁最懂这个?TA 会怎么说?"。触发词:最强大脑、顶级专家、世界级、best minds、谁最懂这个
SkillBFreshness: B★ 3Claude Code +9 - 0.556score
galaxyos
xkzs2007
GalaxyOS 认知增强型 AI Agent 桌面引擎 v0.3.0 Cognitive Nexus
SkillBFreshness: B★ 3Claude Code +9 - 0.555score
decision-engine
xkzs2007
决策引擎。结构化决策分析,权衡利弊,量化风险。触发词:决策、选择、权衡、决策引擎。
SkillBFreshness: B★ 3Claude Code +9 - 0.555score
analogical-thinking
xkzs2007
类比思维。从其他领域的成功模式中借鉴解决方案。触发词:类比、借鉴、像...一样、跨领域。
SkillBFreshness: B★ 3Claude Code +9 - 0.555score
backward-thinking
xkzs2007
逆向思维。从目标倒推起点,识别关键路径。触发词:倒推、逆向、从终点出发、逆向思维。
SkillBFreshness: B★ 3Claude Code +9
Signals
- Stars
- 3
- Forks
- 0
- Open issues
- 0
- Last commit (this folder)
- Jun 12, 2026 (4 months ago)
- Last release
- Aug 1, 2026 (2 months ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/xkzs2007/GalaxyOS-desktop
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from xkzs2007
- 0.564score
criticism-self-criticism
xkzs2007
触发:当一项工作已经完成、进入阶段验收、收到批评反馈,或反复出现同类错误需要系统纠偏时调用;常见信号包括 review、audit、retrospective、quality check、纠错与复盘。 English: Trigger after delivery or at a review checkpoint…
SkillBFreshness: B★ 3Claude Code +9 - 0.564score
concentrate-forces
xkzs2007
触发:当多个任务同时争夺时间、注意力、算力或预算,必须确定主攻方向并停止分散用力时调用;常见信号包括优先级过多、资源紧张、推进分散、需要决定先做什么。 English: Trigger when limited resources are being split across too many tasks and…
SkillBFreshness: B★ 3Claude Code +9 - 0.564score
contradiction-analysis
xkzs2007
触发:当问题复杂、存在多个冲突因素、优先级不清,或你不知道应该先解决什么时调用;常见信号包括 trade-off、瓶颈、根因不明、主次不清、多个问题互相牵制。 English: Trigger when a problem contains competing forces, unclear priorities…
SkillBFreshness: B★ 3Claude Code +9 - 0.564score
arming-thought
xkzs2007
触发:在每次新的顶层对话开始时自动调用,用于建立“实事求是”的总原则,并在明确适用时为后续任务选择下游 skill;如果你是被派遣执行单一具体任务的子 agent,则跳过此 skill。 English: Trigger at the start of each new top-level conversation…
SkillBFreshness: B★ 3Claude Code +9 - 0.561score
autonomous-tasks
xkzs2007
Self-driven AI worker.
SkillBFreshness: B★ 3Claude Code +9 - 0.560score
brain
xkzs2007
Personal knowledge base for capturing and retrieving information about people, places, restaurants, games, tech, events, media, ideas, and organizations.
SkillBFreshness: B★ 3Claude Code +9
Cite this page
darwin-skill. Agent Mart. https://agentmart.dev/skills/xkzs2007-darwin-skill (last updated 2026-10-24)