Context7 Auto Research Skill

This skill automatically fetches current documentation from Context7 API when detecting library/framework-related queries, ensuring responses use up-to-date information instead of potentially outdated training data.

Automatic Activation Triggers

This skill should activate proactively when the user's message contains:

Implementation Queries (实现相关)

• "如何实现" / "怎么写" / "怎么做"
• "How do I..." / "How to..." / "How can I..."
• "Show me how to..." / "Write code for..."

Configuration & Setup (配置相关)

• "配置" / "设置" / "安装"
• "configure" / "setup" / "install"
• "初始化" / "initialize"

Documentation Requests (文档相关)

• "文档" / "参考" / "API"
• "documentation" / "docs" / "reference"
• "查看" / "look up"

Library/Framework Mentions (库/框架提及)

• React, Vue, Angular, Svelte, Solid
• Next.js, Nuxt, Remix, Astro
• Express, Fastify, Koa, Hono
• Prisma, Drizzle, TypeORM
• Supabase, Firebase, Clerk
• Tailwind, shadcn/ui, Radix
• Any npm package or GitHub repository

Code Generation Requests (代码生成)

• "生成代码" / "写一个" / "创建"
• "generate" / "create" / "build"
• "implement" / "add feature"

Research Process

When triggered, follow this workflow:

Step 1: Extract Library Information

Identify the library/framework from the user's query:

• Library name (e.g., "react", "next.js", "prisma")
• Version if specified (e.g., "React 19", "Next.js 15")
• Specific feature/API mentioned (e.g., "useEffect", "middleware", "relations")

Step 2: Search for Library

Use Task tool to call context7-fetcher sub-skill:

Task parameters:
- subagent_type: Bash
- description: "Search Context7 for library"
- prompt: node .claude/skills/context7-auto-research/context7-api.js search "<library-name>" "<user-query>"

Example:

Task: Search for Next.js
Prompt: node .claude/skills/context7-auto-research/context7-api.js search "next.js" "How to configure middleware in Next.js 15"

Response format:

{
  "libraries": [
    {
      "id": "/vercel/next.js",
      "name": "Next.js",
      "description": "The React Framework",
      "trustScore": 95,
      "versions": ["v15.1.8", "v14.2.0", "v13.5.0"]
    }
  ]
}

Why use Task tool?

• Uses context: fork from context7-fetcher sub-skill
• Avoids carrying conversation history to API calls
• Reduces Token consumption

Step 3: Select Best Match

From search results, choose the library based on:

1. Exact name match to user's query
2. Highest trust score (indicates quality/popularity)
3. Version match if user specified (e.g., "Next.js 15" → prefer v15.x)
4. Official packages over community forks

Step 4: Fetch Documentation

Use Task tool to call context7-fetcher sub-skill:

Task parameters:
- subagent_type: Bash
- description: "Fetch documentation from Context7"
- prompt: node .claude/skills/context7-auto-research/context7-api.js context "<library-id>" "<specific-query>"

Example:

Task: Fetch Next.js middleware docs
Prompt: node .claude/skills/context7-auto-research/context7-api.js context "/vercel/next.js" "middleware configuration"

Response format:

{
  "results": [
    {
      "title": "Middleware",
      "content": "Middleware allows you to run code before a request is completed...",
      "source": "docs/app/building-your-application/routing/middleware.md",
      "relevance": 0.95
    }
  ]
}

Why use Task tool?

• Independent context for API calls
• No conversation history overhead
• Faster execution

Step 5: Integrate into Response

Use the fetched documentation to:

1. Answer accurately with current information
2. Include code examples from the docs
3. Cite version when relevant
4. Provide context about the feature/API

Helper Script Usage

The context7-api.js script provides two commands:

Search Library

node context7-api.js search <libraryName> <query>

• Returns matching libraries with metadata
• Use for initial library resolution

Get Context

node context7-api.js context <libraryId> <query>

• Returns relevant documentation snippets
• Use after selecting a library

Environment Setup

The script supports two ways to configure the API key:

Option 1: .env File (Recommended)

Create a .env file in the skill directory:

# In .claude/skills/context7-auto-research/.env
CONTEXT7_API_KEY=your_api_key_here

You can copy from the example:

cp .env.example .env
# Then edit .env with your actual API key

Option 2: Environment Variable

export CONTEXT7_API_KEY="your-api-key"

Priority: Environment variable > .env file

Get API Key: Visit context7.com/dashboard to register and obtain your API key.

If not set, the API will use public rate limits (lower quota).

Best Practices

Query Specificity

• Pass the full user question as the query parameter for better relevance
• Include specific feature names (e.g., "useEffect cleanup" vs just "useEffect")

Version Awareness

• When users mention versions, use version-specific library IDs
• Example: /vercel/next.js/v15.1.8 instead of /vercel/next.js

Error Handling

• If library search returns no results, inform user and suggest alternatives
• If API fails, fall back to training data but mention it may be outdated
• Handle rate limits gracefully (429 errors)

Response Quality

• Don't dump entire documentation - extract relevant parts
• Combine multiple doc snippets if needed for complete answer
• Always include practical code examples

Example Workflows

Example 1: React Hook Question

User: "How do I use useEffect to fetch data in React 19?"

Skill Actions:

1. Detect trigger: "How do I use" + "useEffect" + "React 19"
2. Search: node context7-api.js search "react" "useEffect fetch data"
3. Select: /facebook/react/v19.0.0 (version match)
4. Fetch: node context7-api.js context "/facebook/react/v19.0.0" "useEffect data fetching"
5. Respond with current React 19 patterns (e.g., using use() hook if applicable)

Example 2: Next.js Configuration

User: "配置 Next.js 15 的中间件"

Skill Actions:

1. Detect trigger: "配置" + "Next.js 15" + "中间件"
2. Search: node context7-api.js search "next.js" "middleware configuration"
3. Select: /vercel/next.js/v15.1.8
4. Fetch: node context7-api.js context "/vercel/next.js/v15.1.8" "middleware"
5. Respond with Next.js 15 middleware setup

Example 3: Prisma Relations

User: "Show me how to define one-to-many relations in Prisma"

Skill Actions:

1. Detect trigger: "Show me how" + "Prisma"
2. Search: node context7-api.js search "prisma" "one-to-many relations"
3. Select: /prisma/prisma (highest trust score)
4. Fetch: node context7-api.js context "/prisma/prisma" "one-to-many relations"
5. Respond with Prisma schema examples

Architecture: Context Separation

Why Split into Two Skills?

This skill adopts a two-phase architecture:

1. Main Skill (context7-auto-research) - Needs conversation context:

   • Detect trigger keywords in user message
   • Extract user query intent
   • Select best matching library (version, name, trust score)
   • Integrate documentation into response

2. Sub-Skill (context7-fetcher) - Independent context (context: fork):

   • Execute API calls to Context7
   • Pure HTTP requests, no conversation history needed
   • Reduce Token consumption

Benefits

Call Flow

User Query → Main Skill (detect + analyze)
                ↓
           Task Tool → Sub-Skill (API search)
                ↓
           Main Skill (select best match)
                ↓
           Task Tool → Sub-Skill (API fetch docs)
                ↓
           Main Skill (integrate + respond)

Integration with Existing Skills

This skill complements the existing documentation-lookup skill:

• auto-research: Proactive, automatic activation
• documentation-lookup: Manual, user-invoked via /context7:docs

Both can coexist - use auto-research for seamless UX, documentation-lookup for explicit queries.

Performance Considerations

• Cache responses: Documentation changes infrequently
• Parallel requests: If user asks about multiple libraries, fetch in parallel using multiple Task calls
• Timeout handling: Set reasonable timeouts (5-10s) for API calls
• Fallback strategy: If API unavailable, use training data with disclaimer
• Context efficiency: Sub-skill uses fork context to minimize Token consumption

Limitations

• Requires internet connection for API access
• Subject to Context7 API rate limits
• May not have documentation for very new or obscure libraries
• Documentation quality depends on source repository structure