agentleFS
lzx-Billlzx-Bill/Project-Factory-Core/.agents/skills/skill-creator/SKILL.md

skill-creator

Skill0 starsChanged 4 months agoMITView on GitHub

Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit or optimize an existing skill, run evals to test a skill, or optimize a skill description for better triggering accuracy. Also use when a gap in the current skill set is identified.

189 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use taskflow-inbox-triage, mermaid-diagram-generator and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • AtomicBot-aiSkill

    taskflow-inbox-triage

    Also in 6 projects that use this

  • shalombSkill

    mermaid-diagram-generator

    Guide on creating and editing Mermaid charts safely. Use when asked to draw a diagram, flowchart, or architecture chart. Triggers on: 'mermaid', 'diagram', 'flowchart', 'visualize'.

    Also in 5 projects that use this

  • naman10parikhSkill

    ecosystem-reference

    Quick-access directory of Claude Code ecosystem repos, skills, plugins, MCPs, and patterns. Use when improving the harness.

    Also in 5 projects that use this

  • naman10parikhSkill

    fractal-delegation

    Also in 5 projects that use this

  • Liang-MHSkill

    stock_analysis

    获取 A 股、基金和港股的实时行情和技术分析数据

    Also in 5 projects that use this

  • nicepkgSkill

    community-led-growth

    Expert in community-led growth (CLG) - leveraging user communities to drive acquisition, retention, and expansion. Covers building developer communities, user groups, ambassador programs, and turning customers into advocates. Knows the difference between community as a feature vs community as a growth engine, and how to measure community ROI. Use when "community-led, CLG, ambassador program, champion program, user community, developer community, word of mouth, user group, " mentioned.

    Also in 5 projects that use this

  • +6 more from naman10parikh

The file

Files
SKILL.md

Skill Creator

用于创建新 skill 和迭代改进已有 skill。

When to Use

  • 需要从零创建一个新 skill
  • 需要修改或优化已有 skill
  • 需要运行测试验证 skill 效果
  • 需要优化 skill description 以提升触发准确性
  • 发现当前 skill 集合存在空白

NOT When to Use

  • 只需要运行已有 skill — 直接调用该 skill 即可
  • 只需要编辑单页文档 — 直接用 Write/Edit 工具

Input

来源内容
用户消息需求描述、触发场景、目标输出
当前项目现有 skill 集合(确认是否重复或冲突)

Output Schema

文件类型说明
.agents/skills/<skill-name>/SKILL.mdSkill 定义仓库统一 Skill 位置
.agents/skills/<skill-name>/references/参考文档可选,包含技能相关文档
.agents/skills/<skill-name>/scripts/脚本可选,包含自动化脚本
reports/skill-evals/<skill-name>/测试工作区eval 测试用例和结果

Minimum Viable Output

  • SKILL.md 含:name、description、argument-hint、When to Use、Procedure、Target Pages
  • description 长度 ≥100 字,含具体触发场景
  • 至少 1 轮用户反馈迭代

Complete Output

  • SKILL.md:完整含 Input/Output Schema、Dependencies、Min/Complete Output
  • 含自测报告(测试用例执行结果)
  • description 经过人工评审优化

Dependencies

类型说明
前置无(可随时创建新 skill)
后置新 skill 需经过测试和用户确认后再正式加入路由
依赖读取现有 skill 集合(确认无重复)

Procedure

高层流程

理解意图 → 访谈调研 → 写 SKILL.md 草稿 → 写测试用例 → 跑测试 → 用户评审 → 迭代改进 → 描述优化 → 交付

1. 创建 Skill

Capture Intent

从对话历史中提取用户意图:skill 应该做什么?何时触发?期望的输出格式?

Interview and Research

主动询问边界情况、输入输出格式、示例文件、成功标准。

Write the SKILL.md

按统一结构填写:

  • name: Skill 标识符(kebab-case)
  • description: 触发机制核心,包含 skill 做什么和何时使用。要"pushy"——包含具体触发场景,≥100 字
  • argument-hint: 中文触发词提示
  • When to Use: 触发场景清单
  • NOT When to Use: 何时不用此 skill(避免误触发)
  • Input/Output Schema: 输入来源和目标输出文件
  • Dependencies: 前置/后置/并行 skill
  • Procedure: 标准工作步骤(用 do/not-do 清单)
  • Minimum Viable Output / Complete Output: 两级输出标准
  • Target Pages: 产出的目标页面
  • Enriched Behavior: 增强行为指导(可选)
  • Changelog: 变更记录

Skill 写作规范

Skill Anatomy:

skill-name/
├── SKILL.md (required)
│   ├── YAML frontmatter (name, description required)
│   └── Markdown instructions
└── Bundled Resources (optional)
    ├── scripts/    - Executable code
    ├── references/ - Docs loaded as needed
    └── assets/     - Templates, icons, fonts

页面结构规范:

  • SKILL.md 保持在 500 行以内;超过时拆分为 SKILL.md + 子文档
  • Procedure 使用 do/not-do 清单格式(如 incubation-discovery)
  • 引用文件时明确标注何时需要读取

2. 编写测试用例

写 2-3 个真实测试 prompt,保存到 evals/evals.json:

{
  "skill_name": "example-skill",
  "evals": [
    {
      "id": 1,
      "prompt": "用户的任务描述",
      "expected_output": "期望产出的描述",
      "files": []
    }
  ]
}

测试用例原则:

  • 每个测试用例代表一个真实触发场景
  • 避免过于简单或过于复杂的边界 case
  • 含期望产出描述,用于后续判定是否通过

3. 执行测试

Codex 执行方式(无 Python 工具链)

按以下步骤手动执行测试:

Step 1: 读取被测 skill 读取待测 skill 的 SKILL.md,理解其预期行为。

Step 2: 逐个执行测试用例

对每个测试用例:

  1. 构造符合 prompt 字段的输入
  2. 带着 skill 上下文执行(读取 skill 文件)
  3. 记录实际产出

Step 3: 人工判定

对每个测试用例:

  1. 对照 expected_output 判定 PASS/FAIL
  2. 记录具体差异(PASS 的证据或 FAIL 的原因)
  3. 将结果写入 grading.json:
{
  "eval_id": 1,
  "result": "PASS|FAIL",
  "evidence": "具体证据",
  "gap": "如果 FAIL,描述差距"
}

Step 4: 汇总报告

生成 <workspace>/iteration-1/report.md:

# Skill Eval Report: <skill-name>

## 测试结果

| 测试用例 | 结果 | 证据 |
|---------|------|------|
| eval-1 | PASS | 成功产出预期文件 |
| eval-2 | FAIL | 缺少 X 字段 |

## 改进建议

- 问题 1: ...
- 问题 2: ...

有 Python 工具链的环境(可选)

如果有 scripts/ 目录,可选地使用自动化工具:

# 聚合评分
python -m scripts.aggregate_benchmark <workspace>/iteration-N --skill-name <name>

# 启动评测查看器
python <skill-creator-path>/eval-viewer/generate_review.py <workspace>/iteration-N --skill-name "my-skill"

⚠️ Codex 默认没有这些 Python 工具链。使用上方的手动评测流程。

4. 迭代改进

根据测试结果改进 skill:

  1. 不要过度拟合 — 从具体反馈中提炼通用规律
  2. 保持 prompt 精简 — 删除没有实际作用的描述
  3. 解释 WHY — 用"因为..."替代生硬的 MUST 规则
  4. 寻找重复工作 — 如果所有测试用例都写了同一个辅助函数,考虑 bundling

迭代循环:

应用改进 → 重新跑测试 → 用户评审 → 继续迭代

停止条件:用户满意、反馈为空、或无明显改进空间。

5. Description 优化

优化 description 字段以提升触发准确性:

Step 1: 生成测试查询

创建 20 个测试查询 — 包含应触发和不应触发的场景。重点关注模糊边界。

Step 2: 手动对比

对每个测试查询,对比原 description 和候选 description 的触发效果。选出在准确性和覆盖面间平衡最好的版本。

Step 3: 用户评审

呈现候选 description 给用户,选择或进一步修改。

6. 打包交付

完成所有迭代后:

  1. 确认 .agents/skills/<skill-name>/SKILL.md 结构完整
  2. 确认 Changelog 已更新
  3. 将 skill 目录复制到目标位置
  4. 向用户说明 skill 的触发词和使用方式

Codex 评测工作流(总结)

1. 读取 skill SKILL.md
2. 构造测试输入
3. 带 skill 上下文执行
4. 对照 expected_output 判定 PASS/FAIL
5. 记录到 grading.json
6. 汇总到 report.md
7. 根据反馈迭代改进

无需 Python 脚本、无需 subagent、无需浏览器。全部手动完成。


Reference Files(人工参考)

以下文件可在人工评测时阅读参考:

文件用途
agents/grader.md如何对输出做 assertion 判定
agents/comparator.md如何做 A/B 盲测对比
agents/analyzer.md如何分析评测结果模式
references/schemas.mdevals.json、grading.json 的 JSON 结构规范

这些文件是给人工参考的概念指南,不需要自动化工具即可理解和使用。

Target Pages

  • .agents/skills/<skill-name>/SKILL.md
  • .agents/skills/<skill-name>/references/
  • .agents/skills/<skill-name>/scripts/
  • reports/skill-evals/<skill-name>/

Changelog

日期变更原因
2026-07-15输出路径统一为仓库实际使用的 .agents/skills/,评测工作区统一到 reports/skill-evals/修复 Skill 自身与仓库约定不一致
2026-04-29v2.0: 移除 Python 工具链依赖,重写为 Codex 兼容的手动评测流程;新增 NOT When to Use;Procedure 改为 do/not-do 清单风格Codex 无 Python eval 脚本,需手动评测
2026-04-26统一结构增强(Input/Output/Dependencies/Min-Complete/Changelog)+ 保留完整工作流修复首次增强时意外裁剪核心流程的问题

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.