AI技能创建指南(免费版)
免费AI技能创建指南,指导SKILL.md结构设计、核心原则与Progressive Disclosure分层加载,助力高效准确编写基础技能。
天轰穿
@thcjp
Install
$ openclaw skills install @thcjp/skill-creator-freeAI技能创建指南(免费版)
创建有效AI Skill的基础指南,覆盖Core Principles与SKILL.md结构设计.
输入输出
输入: 用户提供skill创建需求,包括具体的concrete examples(skill应支持的功能描述与触发场景)、预期触发关键词或短语、skill名称建议、所需的bundled resources类型((请参考skill目录中的脚本文件))、skill的复杂度与自由度偏好(High/Medium/Low freedom).
输出: SKILL.md结构设计指导(含frontmatter中 name 与 description 字段的编写规范、body内容组织建议)、Progressive Disclosure三级加载系统设计建议(Level 1 Metadata约100 words / Level 2 SKILL.md body <5k words / Level 3 Bundled resources按需加载)、Bundled Resources基础组织方案((请参考skill目录中的脚本文件))、Core Principles三大原则应用建议。免费版不含 init_skill.py/package_skill.py 脚本集成与完整规范(付费版专享).
依赖说明
运行环境
- Agent平台: 支持SKILL.md的任意AI Agent(Claude Code / Cursor / Codex / Gemini CLI等)
- 操作系统: Windows / macOS / Linux
依赖项
| 依赖项 | 类型 | 是否必需 | 获取方式 |
|---|---|---|---|
| LLM API | API | 必需 | 由Agent内置LLM提供 |
API Key 配置
- 本Skill基于指令驱动,无需额外API Key
可用性分类
- 分类: MD+EXEC()
- 说明: 基于Markdown的AI Skill,通过自然语言指令驱动Agent执行skill创建任务
核心能力
Core Principles指导
提供skill创建的三大核心原则指导:
- Concise is Key:context window是公共资源,skill仅添加Claude不具备的知识。默认假设"Claude already is very smart",对每段内容质疑"Does Claude really need this explanation?"与"Does this paragraph justify its token cost?"。优先用简洁examples替代冗长explanations.
- Set Appropriate Degrees of Freedom:根据任务的fragility和variability匹配specificity级别
- High freedom(text-based instructions):多种方法valid、决策依赖context时使用
- Medium freedom(pseudocode or scripts with parameters):有preferred pattern但允许variation时使用
- Low freedom(specific scripts, few parameters):操作fragile且error-prone、consistency关键时使用
- Anatomy of a Skill:每个skill由required的SKILL.md与optional的Bundled Resources组成
SKILL.md结构设计
提供SKILL.md的标准结构与编写规范:
skill-name/
├── SKILL.md (required)
│ ├── YAML frontmatter metadata (required)
│ │ ├── name: (required)
│ │ └── description: (required)
│ └── Markdown instructions (required)
└── Bundled Resources (optional)
├── scripts/ - Executable code (Python/Bash/etc.)
├── references/ - Documentation intended to be loaded into context as needed
└── assets/ - Files used in output (templates, icons, fonts, etc.)
- Frontmatter(YAML):包含
name和description字段。这是Claude读取以判断何时使用skill的唯一依据,必须清晰全面地描述skill是什么以及何时使用。description是primary triggering mechanism,应包含skill做什么与具体触发场景. - Body(Markdown):使用skill的instructions和guidance。仅在skill触发后加载.
Progressive Disclosure设计
提供三级加载系统的设计原则:
- Level 1 - Metadata(name + description):始终在context中(约100 words)
- Level 2 - SKILL.md body:skill触发时加载(<5k words)
- Level 3 - Bundled resources:Claude按需加载(Unlimited,因scripts可不读入context即执行)
保持SKILL.md body在essentials以内且<500 lines,接近限制时拆分内容。重要准则:references保持一级深度;长reference文件(>100 lines)顶部包含table of contents.
Bundled Resources基础组织
提供三类bundled resources的基础说明:
- Scripts(
scripts/):Executable code用于需要deterministic reliability或反复重写的任务 - References(
references/):按需加载的文档与参考材料,大文件(>10k words)在SKILL.md中包含grep search patterns - Assets(
assets/):用于输出而非加载入context的文件(templates、icons、fonts等)
快速开始
- 确认运行环境满足依赖说明中的要求
- 在AI Agent对话中调用本技能,提供必要的输入参数
- 检查输出结果,根据需要进行后续处理
详细的输入输出格式请参考下方章节说明。
使用流程
- 理解需求:通过concrete examples明确skill应支持的功能与触发场景
- 规划资源:分析examples,列出reusable的scripts、references、assets清单
- 编辑SKILL.md:编写frontmatter(name + description)与body
- 应用Progressive Disclosure:保持SKILL.md <500 lines,detailed info移至references
- 迭代优化:实际使用后识别改进点并更新
结果验证: 任务完成后,查看输出确认状态。成功时返回摘要和数据;失败时根据错误信息排查,参考恢复章节获取修复步骤.
示例
示例1:创建pdf-editor skill基础结构
需求: 用户频繁请求"帮我旋转这个PDF"
Step 1 - Understand:
- 功能: PDF旋转
- 触发: "rotate this PDF"、"rotate PDF"
Step 2 - Plan:
- 分析: 旋转PDF每次都需重写相同代码
- 资源: (请参考skill目录中的脚本文件)
Step 3 - 编写 SKILL.md:
frontmatter:
name: pdf-editor
description: "PDF manipulation toolkit. Use when user needs to rotate, split, merge PDFs..."
body:
- 旋转PDF的使用说明
- 引用 (请参考skill目录中的脚本文件)
Step 4 - Progressive Disclosure 检查:
- SKILL.md <500 lines ✓
- description含触发信息 ✓
错误处理
| 错误场景 | 原因 | 处理方式 |
|---|---|---|
| frontmatter格式错误 | YAML缩进或字段缺失 | 确保name和description字段存在且格式正确,description不为空 |
| skill未触发 | description缺少触发信息 | 在description中补充"when to use"信息与具体触发场景 |
| context window溢出 | SKILL.md过长或未使用Progressive Disclosure | 将detailed info移至references/,保持SKILL.md <5k words |
常见问题
Q1: 免费版与付费版有何区别?
A: 免费版提供基础Core Principles、SKILL.md结构与Progressive Disclosure设计指导。不含完整Bundled Resources规范、Skill Creation Process(init_skill.py/package_skill.py脚本集成)、三种Progressive Disclosure Patterns详解与不应包含内容指南.
Q2: description应该包含哪些内容?
A: description是skill的primary triggering mechanism,必须包含skill做什么与具体触发场景(when to use)。不应在body中放"When to Use"sections,因为body仅在触发后加载.
Q3: Progressive Disclosure的三个层级如何划分?
A: Level 1是Metadata(name + description),始终在context(约100 words);Level 2是SKILL.md body,skill触发时加载(<5k words);Level 3是Bundled resources,Claude按需加载(unlimited,因scripts可执行不读入context).
已知限制
- 不含init_skill.py与package_skill.py脚本集成,需手动创建skill目录结构
- 不含三种Progressive Disclosure Patterns(High-level guide、Domain-specific organization、Conditional details)详解
- 不含完整Bundled Resources组织规范与不应包含内容指南
- 依赖用户提供准确的concrete examples以理解skill需求
- SKILL.md的<500 lines限制对于复杂skill可能需要大量内容拆分
输出格式
{
"success": true,
"data": {
"result": "AI技能创建指南(免费版)处理结果",
"execution_time": "0.5s",
"metadata": {
"version": "1.0",
"processor": "skill-creator"
}
},
"execution_log": [
"解析输入参数",
"执行核心处理",
"格式化输出结果"
],
"error": null
}
Top skills in this category
self-improving agent
@pskoettCaptures learnings, errors, and corrections to enable continuous improvement. Use when: (1) A command or operation fails unexpectedly, (2) User corrects Claude ('No, that's wrong...', 'Actually...'), (3) User requests a capability that doesn't exist, (4) An external API or tool fails, (5) Claude rea
Skill Vetter
@spclaudehomeSecurity-first skill vetting for AI agents. Use before installing any skill from ClawdHub, GitHub, or other sources. Checks for red flags, permission scope, and suspicious patterns.
Self-Improving + Proactive Agent
@ivangdavilaSelf-reflection + Self-criticism + Self-learning + Self-organizing memory. Agent evaluates its own work, catches mistakes, and improves permanently. Use when...
Proactive Agent
@halthelobsterTransform AI agents from task-followers into proactive partners that anticipate needs and continuously improve. Now with WAL Protocol, Working Buffer, Autonomous Crons, and battle-tested patterns. Part of the Hal Stack 🦞
Agent Browser
@matrixyHeadless browser automation CLI optimized for AI agents with accessibility tree snapshots and ref-based element selection