library-docs - SKILL.md Agent Skill

name: library-docs description: CRITICAL - Fetch real-time, official library documentation instead of relying on potentially outdated training data. Use WHENEVER working with ANY external library, framework, or package (React, Vue, Express, etc.) to get the latest APIs, usage patterns, and best practices directly from authoritative sources. Essential for accurate implementation - your training data may be months or years old, but this fetches current docs. allowed-tools: mcpcontext7resolve-library-id, mcpcontext7get-library-docs

Library Documentation Research

What I Do

Fetch up-to-date, official library documentation to ensure you're using current APIs and patterns rather than potentially outdated training data. Training data can be months or years old - this skill retrieves the latest documentation directly from authoritative sources.

Core Process

1. Library Resolution Pattern

Always resolve the library ID first unless explicitly provided:

1. Call mcp__context7__resolve-library-id with the library name
2. Evaluate matches based on:
   - Name similarity (exact matches preferred)
   - Description relevance
   - Documentation coverage (higher snippet counts = better)
   - Trust score (7-10 indicates authoritative source)
3. Select the most relevant match

2. Documentation Retrieval Pattern

Once you have the library ID:

1. Call mcp__context7__get-library-docs with:
   - context7CompatibleLibraryID: The resolved ID
   - topic: Specific feature/API if needed (optional)
   - tokens: 5000 default, up to 10000 for comprehensive coverage
2. Process the returned documentation
3. Extract relevant sections for the query

Search Strategies

For API Documentation

Libraries: "react", "next.js", "vue", "express", "fastify"
Topics: "hooks", "routing", "middleware", "components"
Token Strategy: Use 7000-10000 for comprehensive API coverage

For Implementation Patterns

Focus topics: "authentication", "state-management", "error-handling"
Look for: Code examples, usage patterns, best practices
Token Strategy: 5000 usually sufficient for focused topics

For Version-Specific Information

Format: /org/project/version (e.g., /vercel/next.js/v14.3.0)
Key areas: Breaking changes, migration guides, new features
Token Strategy: Increase tokens for migration guides

Common Library IDs

Claude Code Documentation

Claude Code: /websites/claude_en_claude-code (410+ code snippets for Claude Code features)

Frontend Frameworks & State Management

React: /websites/react_dev (1923 snippets, trust 9) or /reactjs/react.dev (2836 snippets, trust 10)
TanStack Router: /tanstack/router (1307 snippets, trust 8)
TanStack Query: /tanstack/query (749 snippets, trust 8) or /websites/tanstack_query_v5 (4508 snippets)
Zustand: /pmndrs/zustand (462 snippets, trust 9.6)

Backend Frameworks

Hono: /honojs/website (740 snippets) or /llmstxt/hono_dev_llms_txt (1768 snippets)
oRPC: /unnoq/orpc (531 snippets, trust 9.4)
Better Auth: /better-auth/better-auth (1301 snippets, trust 7.6)

Databases & ORMs

PostgreSQL: /postgres/postgres (18 snippets, trust 8.4) or /porsager/postgres (178 snippets, JS client)
ClickHouse: /clickhouse/clickhouse (9889 snippets, trust 9.4)
Drizzle ORM: /drizzle-team/drizzle-orm (436 snippets, trust 7.6)

Build Tools & Runtimes

Vite: /vitejs/vite (480 snippets, trust 8.3)
Turborepo: /vercel/turborepo (822 snippets, trust 10)
Bun: /oven-sh/bun (2377 snippets, trust 9.4)
ESLint: /eslint/eslint (2853 snippets, trust 9.1)
TypeScript: /microsoft/typescript (15930 snippets, trust 9.9)

Testing Tools

Vitest: /vitest-dev/vitest (1183 snippets, trust 8.3)

Cloud & Infrastructure

Supabase: /supabase/supabase (4552 snippets, trust 10)
Cloudflare Workers: /cloudflare/workers-sdk (266 snippets, trust 9.3)
Google Cloud Platform: /googleapis/google-cloud-node (1610 snippets, trust 8.5)

Output Format Template

## Library: [Name] ([Context7 ID])

**Version**: [If specified]
**Topic**: [If focused]

### Key APIs/Features

#### [Feature Name]

**Purpose**: [What it does]
**Usage**:
\`\`\`javascript
[Code example]
\`\`\`
**Parameters**: [If relevant]
**Returns**: [If relevant]

### Implementation Examples

[Relevant code examples from docs]

### Best Practices

- [Practice 1 from official docs]
- [Practice 2 from official docs]

### Related Documentation

- [Other relevant sections to explore]

Error Handling

Library Not Found

If resolve-library-id returns no matches:

Try alternative names (e.g., "nodejs" vs "node")
Check if it's a sub-package (e.g., "@react/router")
Suggest using web search for newer/niche libraries

Insufficient Documentation

If documentation is sparse:

Try broader topic searches
Increase token limit
Consider multiple queries for different aspects
Fall back to web search for community resources

When to Use This Skill vs Training Data or Web Search

ALWAYS use this skill FIRST for:

ANY library or framework you're about to use (React, Vue, Next.js, Express, etc.)
Getting current API signatures and methods
Understanding the latest best practices
Checking if APIs have changed since your training cutoff
Verifying correct usage patterns before writing code
Claude Code documentation (/websites/claude_en_claude-code)

Why not rely on training data:

Your training data may be months or years old
Libraries update frequently with breaking changes
New features and patterns emerge constantly
Deprecated methods may still be in your training data
Best practices evolve - what was correct may now be anti-pattern

Use web search only when:

The library isn't indexed (try this skill first)
You need community tutorials and guides
Looking for real-world implementation stories
Troubleshooting specific error messages
Comparing multiple libraries

Integration Tips

With Codebase Analysis

Use to understand external dependencies
Clarify API usage in existing code
Verify version compatibility
Find migration paths for updates

With Implementation Planning

Get official patterns before coding
Understand API contracts
Find recommended approaches
Identify potential gotchas

Performance Optimization

Cache resolved library IDs mentally
Use focused topics to reduce tokens
Make multiple targeted queries vs one large query
Start with 5000 tokens, increase only if needed

Examples

Example 1: React Hooks Documentation

1. Direct retrieve (use verified ID): mcp__context7__get-library-docs(
     context7CompatibleLibraryID="/reactjs/react.dev",
     topic="hooks",
     tokens=7000
   )
2. Extract: useState, useEffect, useContext, custom hooks patterns

Example 2: TanStack Query Data Fetching

1. Direct retrieve (use verified ID): mcp__context7__get-library-docs(
     context7CompatibleLibraryID="/websites/tanstack_query_v5",
     topic="queries mutations",
     tokens=8000
   )
2. Extract: useQuery, useMutation, optimistic updates, caching strategies

Example 3: Drizzle ORM with PostgreSQL

1. Direct retrieve (use verified ID): mcp__context7__get-library-docs(
     context7CompatibleLibraryID="/drizzle-team/drizzle-orm",
     topic="postgres queries",
     tokens=6000
   )
2. Extract: Query builder, joins, transactions, migrations

Example 4: Hono API Development

1. Direct retrieve (use verified ID): mcp__context7__get-library-docs(
     context7CompatibleLibraryID="/llmstxt/hono_dev_llms_txt",
     topic="middleware routing",
     tokens=5000
   )
2. Extract: Route handling, middleware, context, error handling

Example 5: Claude Code Documentation

1. Direct retrieve (no resolution needed): mcp__context7__get-library-docs(
     context7CompatibleLibraryID="/websites/claude_en_claude-code",
     topic="skills agents mcp",
     tokens=8000
   )
2. Extract: Skill creation, agent patterns, MCP integration