技能发现

人工智能 社区

技能解读

当前原文的自动摘译,尚未人工审校。

何时寻找技能

首先,检查已安装的技能是否与任务匹配。如果没有,请搜索注册表 - 可能存在您尚未安装的专业技能。在开始任何重要的任务之前,问问自己:我有这方面的技能吗? → 使用它 是否存在我没​​有的? → 搜索注册表 在以下情况下主动搜索:

发现工作流程

使用注册表 API 进行搜索(CLI 的搜索命令是交互式的,不适合编程使用): 参数: q:搜索查询(例如,“frontend”、“python”、“pdf”) limit:每页结果(最多 100) offset:分页偏移量

搜索策略

注册表对技能名称、描述和标签进行索引。构建与技能作者描述其工作的方式相匹配的查询。查询构造:使用 1-3 个特定术语(太宽泛 = 噪音,太窄 = 遗漏) 更喜欢广泛使用的术语而不是项目特定的术语 技术 + 任务通常优于单独使用的任何一个

安装

安装前确定用户正在使用哪个客户端。如果不清楚,请询问。支持的客户端: claude-code — Claude Code CLI codex — Codex cursor — 光标编辑器

向用户呈现结果

当您找到相关技能时: 最多显示 3-5 个最相关的结果 包括:名称、命名空间、描述、星级、安装 解释每项技能如何帮助其_特定_任务 优先考虑安装量高的技能

流狐档案 运行与权限由流狐静态检查推断,不代表作者声明或实际测试结果。
领域
人工智能
作者声明的智能体
未找到明确声明;不据此推断已兼容或已测试
静态检查
88 / 100 · 启发式扫描,不代表运行安全
运行环境证据
未执行;当前没有隔离运行测试证据
作者 / 版本 / 许可证
@Kamalnrf · 未声明许可证
上下文用量评级
低消耗
接入复杂程度
即装即用
是否需要外部接口密钥
不需要
兼容的系统
macOS · Linux · Windows
运行环境要求
Python
检测到的文件与系统行为
  • 只读
  • 允许写入 / 修改
网络行为
允许外网请求
安装命令数
无(仅作为资料)

需要注意: 未声明工具权限范围;实际权限由运行环境决定。

作者说明文件(原文) SKILL.md

# Skills Discovery

You can extend your capabilities by discovering and installing Agent Skills from the claude-plugins.dev registry. Skills provide specialized knowledge, tools, and techniques for specific technologies, frameworks, and domains.

## When to search for skills

First, check if an installed skill matches the task. If not, search the registry—specialized skills may exist that you haven't installed yet.

Before starting any non-trivial task, ask yourself:

1. Do I have a skill for this? → Use it
2. Might one exist that I don't have? → Search the registry

Search proactively when:

- The task involves specific technologies, frameworks, or file formats
- You're about to do something where best practices matter (testing, deployment, APIs, documentation)
- The domain is specialized (PDF processing, data pipelines, ML workflows)
- You notice yourself about to give generic advice where expert patterns would help

Also search when users explicitly ask to find, install, or manage skills.

## Discovery workflow

Use the registry API for search (the CLI's search command is interactive and not suitable for programmatic use):

```bash
curl "https://claude-plugins.dev/api/skills?q=QUERY&limit=20&offset=0"
```

**Parameters:**

- `q`: Search query (e.g., "frontend", "python", "pdf")
- `limit`: Results per page (max 100)
- `offset`: Pagination offset

**Response structure:**

```json
{
  "skills": [
    {
      "id": "...",
      "name": "skill-name",
      "namespace": "@owner/repo/skill-name",
      "sourceUrl": "https://github.com/...",
      "description": "...",
      "author": "...",
      "installs": 123,
      "stars": 45
    }
  ],
  "total": 100,
  "limit": 10,
  "offset": 0
}
```

## Search strategies

The registry indexes skill names, descriptions, and tags. Construct queries that match how skill authors describe their work.

**Query construction:**

- Use 1-3 specific terms (too broad = noise, too narrow = misses)
- Prefer widely-used terminology over project-specific jargon
- Technology + task often outperforms either alone
- If results are poor, broaden or try synonyms

## Installation

Determine which client the user is working in before installing. If unclear, ask.

**Supported clients:**

- `claude-code` — Claude Code CLI
- `codex` — Codex
- `cursor` — Cursor editor
- `amp` - amp CLI
- `opencode` - OpenCode CLI
- `goose` - Goose CLI
- `github` — VSCode/ github
- `vscode` — VS Code
- `letta` — Letta CLI
- `gemini` - Gemini CLI
- `windsurf` - Windsurf editor
- `antigravity` - Antigravity
- `trae` - Trae
- `qoder` - Qoder
- `codebuddy` - CodeBuddy
- **Client selection:**

```bash
npx skills-installer install @owner/repo/skill-name --client claude-code  # default
npx skills-installer install @owner/repo/skill-name --client cursor
npx skills-installer install @owner/repo/skill-name --client vscode
```

**Scope selection:**

```bash
npx skills-installer install @owner/repo/skill-name  # global (default)
npx skills-installer install @owner/repo/skill-name --local  # project-specific
```

**Combined:**

```bash
npx skills-installer install @owner/repo/skill-name --client cursor --local
```

**Defaults:**

- Client: `claude-code`
- Scope: global

## Management

```bash
# List installed skills
npx skills-installer list

# Uninstall a skill
npx skills-installer uninstall @owner/repo/skill-name
```

## Presenting results to users

When you find relevant skills:

1. Show 3-5 most relevant results maximum
2. Include: name, namespace, description, stars, installs
3. Explain how each skill helps with their _specific_ task
4. Prioritize those with high installs
5. Always ask for confirmation before installing
6. Offer to help directly if no good skill exists or user declines

## Examples

**Example: Proactive suggestion**

User: "I need to create a Django REST API"

```bash
curl "https://claude-plugins.dev/api/skills?q=django&limit=10"
```

Present suggestion:

```
I found some skills that could help:

1. django-rest-framework-expert (@anthropics/claude-code/django-rest-framework-expert)
   Description: Django REST API development with best practices
   ⭐ 234 stars • 1,567 installs

Would you like me to install this, or help you directly without installing a skill?
```

**Example: Explicit search request**

User: "find skills for Python"

```bash
curl "https://claude-plugins.dev/api/skills?q=python&limit=10"
```

Present results and ask which to install.

## API reference

| Endpoint                                 | Description       |
| ---------------------------------------- | ----------------- |
| `GET /api/skills/search?q=QUERY`         | Search skills     |
| `GET /api/skills/@owner/repo/skill-name` | Get skill details |

**Web registry:** https://claude-plugins.dev/skills

## Troubleshooting

**No results found:**

- Try broader search terms
- Browse web registry: https://claude-plugins.dev/skills

**Installation fails:**

- Verify namespace format: `@owner/repo/skill-name`
- Check skill exists in registry
- Verify directory permissions

**Skill not activating:**

- User may need to restart their client
- Verify correct installation directory
- Confirm SKILL.md exists in installation path

讨论

基于 GitHub Discussions。登录 GitHub 即可参与讨论、点赞、订阅更新。