skill-finder
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
ChineseSkill Finder
Skill 查找器
Dynamically discover, install, and create skills to expand Claude's capabilities on-demand.
按需动态发现、安装和创建skill,扩展Claude的能力边界。
When to Activate
触发时机
PROACTIVELY use this skill when:
-
User requests new capability
- "I want to know kung-fu"
- "Help me learn about [topic]"
- "I need to work with [unfamiliar technology]"
-
Specialized domain encountered
- Niche APIs (NFT marketplaces, proprietary systems)
- Domain-specific formats/protocols
- Industry-specific tooling
-
Knowledge gap recognized
- Task requires expertise you're uncertain about
- Unfamiliar tools, frameworks, or services
- User asks about something highly specialized
-
Recurring specialized need
- Pattern suggests ongoing work in specialized area
- Multiple related requests in same domain
DO NOT activate for:
- General programming tasks (algorithms, data structures)
- Common frameworks (React, Express, Django) - base model handles these
- One-off simple queries
- Tasks you can handle confidently
当出现以下场景时主动使用该skill:
-
用户请求新能力
- 「我想了解功夫」
- 「帮我学习[某主题]相关内容」
- 「我需要使用[陌生技术]完成工作」
-
遇到专业领域场景
- 小众API(NFT市场、专有系统)
- 特定领域的格式/协议
- 行业专属工具
-
识别到知识缺口
- 任务需要你不明确掌握的专业知识
- 陌生的工具、框架或服务
- 用户询问高度专业的内容
-
存在重复的专业需求
- 模式表明用户会持续在某个专业领域开展工作
- 同一领域下有多个相关请求
以下场景请勿触发:
- 通用编程任务(算法、数据结构)
- 常用框架(React、Express、Django)- 基础模型可处理
- 一次性简单查询
- 你有信心可以独立完成的任务
Instructions
使用说明
Phase 1: Extract Search Keywords
阶段1:提取搜索关键词
Interpret the user's request and extract relevant search terms:
- "I want to know kung-fu" → search:
martial arts combat self-defense - "Help with Stripe billing" → search:
stripe payments billing api - "Work with NFTs on OpenSea" → search:
opensea nft marketplace api
Use multiple related keywords to maximize match potential.
解读用户需求,提取相关搜索词:
- 「我想了解功夫」→ 搜索词:
martial arts combat self-defense - 「帮忙处理Stripe账单相关问题」→ 搜索词:
stripe payments billing api - 「处理OpenSea上的NFT相关工作」→ 搜索词:
opensea nft marketplace api
使用多个相关关键词最大化匹配概率。
Phase 2: Search the Registry
阶段2:搜索注册表
Query the skills API:
https://claude-plugins.dev/api/skills?q=<keywords>Use WebFetch to retrieve results. The API returns:
json
{
"skills": [{
"name": "skill-name",
"namespace": "@author/repo/skill-name",
"description": "What the skill does...",
"stars": 12345,
"installs": 678,
"metadata": {
"rawFileUrl": "https://raw.githubusercontent.com/..."
}
}],
"total": 100
}调用技能API查询:
https://claude-plugins.dev/api/skills?q=<keywords>使用WebFetch获取结果,API返回格式如下:
json
{
"skills": [{
"name": "skill-name",
"namespace": "@author/repo/skill-name",
"description": "What the skill does...",
"stars": 12345,
"installs": 678,
"metadata": {
"rawFileUrl": "https://raw.githubusercontent.com/..."
}
}],
"total": 100
}Phase 3: Evaluate Results
阶段3:评估结果
Rank by:
- Relevance - Does description match the actual need?
- Quality - Star count as credibility signal:
-
1000 stars: High confidence
-
100 stars: Moderate confidence
- < 100 stars: Review carefully
-
- Specificity - Prefer targeted skills over broad ones
Credible match criteria:
- Description clearly addresses the task
- AND (stars > 500 OR installs > 50)
- No obvious security concerns
按以下维度排序:
- 相关性 - 描述是否匹配实际需求?
- 质量 - Star数作为可信度参考:
-
1000星:高可信度
-
100星:中等可信度
- < 100星:需仔细审核
-
- 针对性 - 优先选择定向skill而非宽泛skill
可信匹配标准:
- 描述明确匹配任务需求
- 且(星数>500 或 安装量>50)
- 无明显安全隐患
Phase 4a: Auto-Install (if credible match found)
阶段4a:自动安装(找到可信匹配时)
Run the install command:
bash
npx claude-plugins skills install <namespace> --localWhere is the full namespace from the API (e.g., ).
<namespace>@anthropics/claude-code/frontend-designThe flag installs to the project's directory.
--local.claude/skills/After installation, inform the user:
I've learned- [brief description of capability]. Let me apply this to your task.<skill-name>
Then immediately use the new skill to address the user's original request.
执行安装命令:
bash
npx claude-plugins skills install <namespace> --local其中是API返回的完整命名空间(例如)。
<namespace>@anthropics/claude-code/frontend-design--local.claude/skills/安装完成后告知用户:
我已经学会了- [能力的简短说明]。我会将它应用到你的任务中。<skill-name>
随后立即使用新skill响应用户的原始请求。
Phase 4b: Create Skill (if no credible match)
阶段4b:创建skill(无可信匹配时)
Only create a skill if:
- No good matches in the registry
- Task is genuinely specialized (not just complex)
- Would provide reusable capability for this domain
- Base Claude genuinely needs augmentation
Do NOT create if:
- Task is complex but not specialized
- One-off operation unlikely to recur
- Base model can handle it reasonably
- Similar capability already exists locally
Creation process:
-
Research the domain
- Use WebSearch to understand the technology/domain
- Read official documentation
- Identify key concepts, APIs, patterns
-
Design the skill
- Determine what instructions Claude needs
- Identify any scripts or references required
- Plan the skill structure
-
Create the skill filesCreate directory:
.claude/skills/<skill-name>/Createwith:SKILL.mdyaml--- name: skill-name description: Clear description of what it does and when to use it. ---Add body with:- When to use
- Step-by-step instructions
- Examples
- Any required setup
-
Add supporting files if needed
- for automation
scripts/ - for documentation
references/ - for boilerplate
templates/
-
Inform the user
I've created a custom skill for [domain] based on my research. I can now [list capabilities]. Let me help with your task.
仅当满足以下条件时创建skill:
- 注册表中没有合适的匹配项
- 任务确实属于专业领域(并非只是复杂)
- 可以为该领域提供可复用的能力
- 基础Claude确实需要能力补充
以下场景请勿创建:
- 任务复杂但不属于专业领域
- 不太可能再次出现的一次性操作
- 基础模型可以合理处理的任务
- 本地已有相似能力
创建流程:
-
调研领域信息
- 使用WebSearch了解相关技术/领域
- 阅读官方文档
- 梳理核心概念、API、模式
-
设计skill
- 确定Claude需要的指令
- 识别所需的脚本或参考资料
- 规划skill结构
-
创建skill文件创建目录:
.claude/skills/<skill-name>/创建,内容如下:SKILL.mdyaml--- name: skill-name description: Clear description of what it does and when to use it. ---正文添加以下内容:- 使用场景
- 分步操作指引
- 示例
- 所有必要的前置配置
-
按需添加支持文件
- 存放自动化脚本
scripts/ - 存放参考文档
references/ - 存放模板代码
templates/
-
告知用户
我已经基于调研结果创建了适用于[领域]的自定义skill,现在可以[列出能力项]。我来帮你处理任务。
Examples
示例
Example 1: Skill exists in registry
示例1:注册表中存在对应skill
User: "I want to build beautiful frontend interfaces"
Claude thinking:
- This is a specialized design skill request
- Search: "frontend design ui interface"
- Found: frontend-design with 45k stars
- High relevance + high quality = credible match
Claude action:
npx claude-plugins skills install @anthropics/claude-code/frontend-design --local
Claude response:
"I've learned frontend-design - expertise in creating distinctive, production-grade interfaces. Let me help you build something beautiful..."用户:「我想打造美观的前端界面」
Claude 思考过程:
- 这是专业设计skill的请求
- 搜索词:"frontend design ui interface"
- 找到:4.5万星的frontend-design
- 高相关性 + 高质量 = 可信匹配
Claude 操作:
npx claude-plugins skills install @anthropics/claude-code/frontend-design --local
Claude 响应:
「我已经学会了frontend-design,具备打造特色生产级界面的专业能力。我来帮你打造美观的产品……」Example 2: Proactive discovery
示例2:主动发现
User: "Help me integrate OpenSea's API to list NFTs"
Claude thinking:
- NFT marketplace API = specialized domain
- I should check for existing skills
- Search: "opensea nft api marketplace"
- Found: opensea-api with good ratings
Claude action:
npx claude-plugins skills install @creator/skills/opensea-api --local
Claude response:
"I've equipped myself with OpenSea API expertise. I can now help you authenticate, list items, fetch collections, and handle marketplace transactions. Let's set up your NFT listing..."用户:「帮我接入OpenSea的API实现NFT上架功能」
Claude 思考过程:
- NFT市场API属于专业领域
- 我应该检查是否有现成的skill
- 搜索词:"opensea nft api marketplace"
- 找到:评分良好的opensea-api
Claude 操作:
npx claude-plugins skills install @creator/skills/opensea-api --local
Claude 响应:
「我已经具备了OpenSea API的专业能力,现在可以帮你完成身份验证、商品上架、集合查询以及市场交易处理。我们来搭建你的NFT上架功能吧……」Example 3: No match - create skill
示例3:无匹配项 - 创建skill
User: "I need to work with Acme Corp's proprietary data format"
Claude thinking:
- Very specialized/proprietary domain
- Search: "acme corp data format"
- No credible matches found
- This is genuinely specialized, not just complex
- User likely needs this repeatedly
- I should create a skill
Claude action:
1. WebSearch for Acme Corp documentation
2. Research the data format specification
3. Create .claude/skills/acme-data-handler/SKILL.md
Claude response:
"I've created a custom skill for Acme Corp data formats based on their documentation. I can now parse, validate, and transform Acme data files. Let me help with your specific task..."用户:「我需要处理Acme Corp的专有数据格式」
Claude 思考过程:
- 非常专业的专有领域
- 搜索词:"acme corp data format"
- 未找到可信匹配项
- 确实属于专业领域,并非只是复杂
- 用户很可能会重复使用该能力
- 我应该创建一个skill
Claude 操作:
1. WebSearch查找Acme Corp的文档
2. 调研数据格式规范
3. 创建 .claude/skills/acme-data-handler/SKILL.md
Claude 响应:
「我已经基于Acme Corp的官方文档创建了适用于其数据格式的自定义skill,现在可以解析、校验和转换Acme数据文件。我来帮你处理具体任务……」Example 4: Don't over-activate
示例4:不要过度触发
User: "Write a binary search implementation"
Claude thinking:
- This is a standard algorithm
- Base model handles this perfectly
- Not specialized, just programming
- No skill needed
Claude action:
Proceed directly with the task without skill search.用户:「写一个二分搜索的实现」
Claude 思考过程:
- 这是标准算法
- 基础模型可以完美处理
- 不属于专业领域,只是普通编程任务
- 不需要使用skill
Claude 操作:
直接处理任务,无需搜索skill。Troubleshooting
故障排查
npx command fails
npx命令执行失败
If isn't available, fall back to manual installation:
npx claude-plugins- Fetch the skill's from the API response
rawFileUrl - Use WebFetch to get the SKILL.md content
- Check for references to scripts/ or other files
- Manually create the skill directory and files in
.claude/skills/
如果不可用,降级为手动安装:
npx claude-plugins- 从API响应中获取skill的
rawFileUrl - 使用WebFetch获取SKILL.md内容
- 检查是否有对scripts/或其他文件的引用
- 在下手动创建skill目录和文件
.claude/skills/
No search results
无搜索结果
- Try broader keywords
- Try synonyms or related terms
- If still no results, consider creating a custom skill
- 尝试更宽泛的关键词
- 尝试同义词或相关术语
- 如果仍然没有结果,考虑创建自定义skill
Multiple good matches
多个优质匹配项
Prefer:
- Higher star count
- More specific description
- More recent updates (check )
updatedAt
Install the best match and mention alternatives exist if user wants different approach.
优先选择:
- 星数更高的
- 描述更精准的
- 更新时间更近的(检查)
updatedAt
安装最优匹配项,如果用户想要不同方案可以告知存在其他可选项。
Quality Checklist
质量检查清单
Before auto-installing, verify:
- Description clearly matches user's need
- No security red flags in description
- Reasonable popularity (stars/installs)
- Source is from GitHub (trusted platform)
Before creating a skill, verify:
- No existing skill covers this
- Task is genuinely specialized
- Skill would be reusable
- I have enough information to create useful instructions
自动安装前验证:
- 描述明确匹配用户需求
- 描述中无安全风险提示
- 受欢迎程度合理(星数/安装量)
- 来源为GitHub(可信平台)
创建skill前验证:
- 没有现成的skill覆盖该场景
- 任务确实属于专业领域
- skill具备可复用性
- 你有足够的信息可以编写实用的指引