home / skills / enoch-robinson / agent-skill-collection / skill-creator

skill-creator skill

/skills/meta/skill-creator

This skill guides creating high-quality Skills, enabling users to design, structure, and validate modular capability packs for Claude.

npx playbooks add skill enoch-robinson/agent-skill-collection --skill skill-creator

Review the files below or copy the command above to add this skill to your agents.

Files (1)
SKILL.md
4.6 KB
---
name: skill-creator
description: Skill 创建指南。当用户需要创建新的 Skill、更新现有 Skill、或学习如何设计高质量 Skill 时使用此技能。
---

# Skill Creator

指导创建高质量的 Skills,扩展 Claude 的专业能力。

## 什么是 Skill?

Skill 是模块化、自包含的能力包,通过提供专业知识、工作流程和工具来扩展 Claude 的能力。它们就像特定领域的"入职指南"——将Claude 从通用助手转变为特定任务的专家。

### Skill 能提供什么

1. **专业工作流** - 特定领域的多步骤流程
2. **工具集成** - 特定文件格式或API 的操作指南
3. **领域知识** - 公司特定的知识、模式、业务逻辑
4. **捆绑资源** - 脚本、参考文档、模板资源

## 核心原则

### 1. 简洁至上

上下文窗口是公共资源。Skill 与系统提示、对话历史、用户请求共享空间。

**默认假设:Claude 已经很聪明。** 只添加 Claude 不知道的内容。对每条信息提问:
- "Claude 真的需要这个解释吗?"
- "这段内容值得占用 token 吗?"

**优先使用简洁示例,而非冗长解释。**

### 2. 设置适当的自由度

根据任务的脆弱性和可变性匹配具体程度:

| 自由度 | 形式 | 适用场景 |
|--------|------|----------|
| 高 | 文本指令 | 多种方法有效、依赖上下文决策 |
| 中 |伪代码/带参数脚本 | 存在首选模式、允许一定变化 |
| 低 | 具体脚本、少量参数 | 操作脆弱易错、一致性关键 |

### 3. 渐进式披露

Skill 使用三级加载系统管理上下文:

1. **元数据** (name + description) - 始终在上下文中 (~100词)
2. **SKILL.md 正文** - 触发时加载 (<5k词)
3. **捆绑资源** - 按需加载 (无限制)

## Skill 结构

每个 Skill 由必需的 SKILL.md 文件和可选的捆绑资源组成:

```
skill-name/
├── SKILL.md (必需)
│   ├── YAML frontmatter (必需)
│   │   ├── name: (必需)
│   │   └── description: (必需)
│   └── Markdown 指令 (必需)
└── 捆绑资源 (可选)
    ├── scripts/      - 可执行脚本 (Python/Bash等)
    ├── references/   - 按需加载的参考文档
    └── assets/       - 输出中使用的文件 (模板、图标等)
```

### SKILL.md 组成

#### Frontmatter (YAML)

```yaml
---
name: my-skill-name
description: 清晰描述功能和使用场景。这是触发机制的关键。
---
```

- `name`: 唯一标识符(小写,用连字符分隔)
- `description`: 完整描述功能和触发条件

#### Body (Markdown)

Skill 的具体指令、示例和指南。仅在Skill 触发后加载。

##创建流程

### 步骤 1:理解需求

通过具体示例理解 Skill 的使用场景:
- "这个 Skill 应该支持什么功能?"
- "用户会如何使用它?"
- "什么情况下应该触发这个 Skill?"

### 步骤 2:规划资源

分析每个示例,确定需要的可复用资源:
- `scripts/` - 重复编写的代码 →脚本化
- `references/` - 需要参考的文档 → 文档化
- `assets/` - 输出中使用的文件 → 模板化

### 步骤 3:创建目录结构

```bash
mkdir -p my-skill/{scripts,references,assets}
touch my-skill/SKILL.md
```

### 步骤 4:编写 SKILL.md

1. 编写 frontmatter(name + description)
2. 编写核心指令和工作流
3. 添加示例和指南
4. 引用捆绑资源

### 步骤 5:测试验证

- 在实际场景中测试 Skill
- 验证触发条件是否准确
- 检查输出是否符合预期

### 步骤 6:迭代优化

根据使用反馈持续改进:
- 补充遗漏的场景
- 优化指令清晰度
- 精简冗余内容

## 最佳实践

### Description编写要点

Description 是 Skill 的主要触发机制,必须包含:
- **功能描述**:Skill 做什么
- **触发条件**:何时使用这个 Skill
- **使用场景**:具体的应用示例

```yaml
#❌ 不好的 description
description: 处理 PDF 文件

# ✅ 好的 description
description: PDF 处理工具包,用于提取文本和表格、创建新PDF、合并拆分文档、处理表单。当需要程序化处理、生成或分析 PDF 文档时使用。
```

### 内容组织原则

1. **SKILL.md 保持精简** - 控制在 500 行以内
2. **详细内容放references/** - 按需加载
3. **避免重复** - 信息只在一处存在
4. **清晰引用** - 明确说明何时读取哪个文件

## 检查清单

创建 Skill 前确认:

- [ ] `name` 使用小写和连字符
- [ ] `description` 包含功能和触发条件
- [ ] 指令简洁,无冗余解释
- [ ] 示例具体可操作
- [ ] 资源文件有清晰引用
- [ ] 在实际场景中测试过

Overview

This skill is a concise guide for creating, updating, and testing high-quality Skills that extend an AI assistant’s capabilities. It explains required metadata, structure, resource packaging, and a practical creation workflow to produce reliable, reusable Skills. The goal is fast iteration and predictable triggers for real-world use.

How this skill works

The skill inspects design choices, required metadata (name and description), and recommended directory layout to ensure each Skill is discoverable and lightweight. It prescribes a three-tier content loading approach (minimal metadata, main instructions loaded on trigger, and optional bundled resources loaded on demand) and a step-by-step creation and validation workflow. It also provides checklists and examples to speed testing and iteration.

When to use it

  • You need to create a new Skill to add domain-specific capabilities to an assistant.
  • You want to update an existing Skill to improve triggers, examples, or resources.
  • You are designing triggers and descriptions so the assistant reliably selects the right Skill.
  • You must package reusable scripts, references, or templates for on-demand loading.
  • You want a checklist to validate a Skill before production deployment.

Best practices

  • Keep the manifest metadata short and explicit: unique lowercase name plus a trigger-focused description.
  • Favor concise examples over long explanations; tokens are shared with conversation context.
  • Use progressive disclosure: keep only essential metadata always loaded and load detailed content or assets on demand.
  • Match instruction specificity to task fragility: high-level text for flexible tasks, structured scripts for semi-deterministic tasks, and concrete scripts for brittle operations.
  • Store large or rarely used reference material as separate resources and reference them clearly from the main instructions.

Example use cases

  • Create a PDF processing Skill with a clear description of triggers and sample commands for extraction and merging.
  • Package a data-cleaning Skill that exposes reusable Python scripts and a short usage example for different datasets.
  • Build a deployment Skill that includes step-by-step runbooks and on-demand logs or templates.
  • Update a Skill by tightening its trigger description and adding a failing test case to the validation checklist.

FAQ

What must the manifest description include?

A short functional summary, clear trigger conditions, and one or two practical usage scenarios so the assistant can detect when to run the Skill.

How do I keep context usage low?

Prioritize minimal metadata, use short concrete examples, and move detailed documentation into referenced resources loaded only when needed.