AgentSkillsCN

docs-seeker

通过 llms.txt 标准、context7.com 和 repomix 查找相关文档,掌握库研究的最佳实践。

SKILL.md
--- frontmatter
name: docs-seeker
description: Find documentation using llms.txt standard, context7.com, and repomix. Best practices for library research.

Docs Seeker

Load this skill when you need to find documentation for libraries, frameworks, or APIs.

Documentation Sources (Priority Order)

1. Context7 (llms.txt aggregator) - TRY FIRST

code
# Use the context7 MCP tool
context7_resolve_library_id(libraryName="next.js")
context7_get_library_docs(libraryId="...", topic="app router")

Supported libraries: Next.js, React, shadcn/ui, Tailwind CSS, Better Auth, Drizzle ORM, and many more.

Why first: Aggregates llms.txt files from official sources, AI-optimized format.


2. Official Documentation Sites

If context7 doesn't have it:

code
# Use webfetch for official docs
webfetch(url="https://docs.example.com/api/reference")

Best sources:

  • GitHub README.md
  • Official docs site
  • API reference pages

3. Repomix (GitHub Repo Analysis)

For understanding library internals:

bash
# Package entire repo for analysis
npx repomix --remote https://github.com/owner/repo --output repo.xml

Use when:

  • Need to understand internal implementation
  • Docs are incomplete
  • Looking for usage examples in source

Search Strategy

For Framework/Library Usage

code
1. context7 → resolve library ID
2. context7 → get docs for specific topic
3. If incomplete → webfetch official docs
4. If still unclear → repomix the source

For API/SDK Integration

code
1. webfetch → API reference page
2. Look for: endpoints, auth, rate limits
3. Find code examples
4. Test with minimal example first

For Error Messages

code
1. grep_app → search GitHub for exact error
2. Look for: issues, discussions, PRs
3. Find the fix or workaround
4. Verify solution applies to your version

llms.txt Standard

Many sites now provide /llms.txt files optimized for AI:

code
# Check for llms.txt
webfetch(url="https://example.com/llms.txt")

Format: Markdown with structured sections for AI consumption.

Growing adoption: Next.js, Vercel, Anthropic, and more.


Best Practices

DO

  • ✅ Check context7 first (fastest, most reliable)
  • ✅ Get specific: "Next.js app router dynamic routes" not "Next.js"
  • ✅ Verify version compatibility
  • ✅ Look for code examples, not just prose
  • ✅ Test examples before using in production

DON'T

  • ❌ Rely on memory for API details
  • ❌ Use outdated Stack Overflow answers
  • ❌ Copy code without understanding it
  • ❌ Skip version checks (APIs change!)
  • ❌ Assume docs are complete

Quick Reference

NeedToolExample
Library docscontext7context7_get_library_docs("react", "hooks")
API referencewebfetchwebfetch("https://api.example.com/docs")
GitHub searchgrep_appgrep_app("error message")
Source analysisrepomixnpx repomix --remote url

Parallel Research Pattern

For comprehensive research, fire multiple agents:

code
task(agent="researcher", prompt="Find Next.js app router docs via context7")
task(agent="researcher", prompt="Find authentication patterns for Next.js")
task(agent="explore", prompt="Find how auth is implemented in this codebase")

Combine external docs with internal patterns for best results.