docs-seeker

Documentation Discovery

Find technical documentation through: llms.txt → Repomix → Research

Decision Flow

1. Try context7.com llms.txt
   ├─ Found → Process URLs with Explorer agents
   └─ Not found ↓
2. Find GitHub repository
   ├─ Found → Use Repomix
   └─ Not found ↓
3. Deploy Researcher agents

Method 1: llms.txt (Priority)

context7.com Patterns

Examples:

github.com/vercel/next.js     → context7.com/vercel/next.js/llms.txt
docs.imgix.com                → context7.com/websites/imgix/llms.txt
shadcn/ui + date picker       → context7.com/shadcn-ui/ui/llms.txt?topic=date

Fallback: WebSearch

"[library] llms.txt site:[docs domain]"

Processing URLs

Launch all agents in single message (parallel, not sequential).

Method 2: Repomix

When llms.txt unavailable, analyze GitHub repository directly.

Quick Commands

# Remote (recommended - no clone needed)
npx repomix --remote https://github.com/owner/repo

# Docs only
npx repomix --remote [url] --include "docs/**,*.md"

# Source only
npx repomix --remote [url] --include "src/**" --exclude "**/*.test.*"

# Specific version
npx repomix --remote [url] --remote-branch v2.0.0

# With security check
npx repomix --remote [url] --security-check

Key Options

Language-Specific Patterns

# JavaScript/TypeScript
--include "src/**,lib/**,*.md" --exclude "node_modules/**,dist/**,*.test.*"

# Python
--include "src/**,*.py,*.md" --exclude "__pycache__/**,venv/**"

# Go
--include "**/*.go,go.mod,*.md" --exclude "vendor/**,*_test.go"

# PHP/Laravel
--include "app/**,config/**,routes/**,*.md" --exclude "vendor/**,storage/**"

Error Handling

Method 3: Research (Fallback)

When no structured docs exist, deploy 3-4 Researcher agents:

1. Official sources: Package registry, website, releases
2. Tutorials: Blog posts, guides
3. Community: Stack Overflow, GitHub discussions
4. API/Reference: Auto-generated docs, code examples

Output Format

# [Library] Documentation

## Source
- Method: llms.txt / Repomix / Research
- URL: [source]
- Version: [version]
- Date: [date]

## Content
[Organized by topic, not by agent]

## Limitations
[Missing sections, version caveats]

Best Practices

1. context7.com first - fastest path to documentation
2. Use ?topic= when applicable - reduces noise
3. Parallel agents - never sequential for 3+ URLs
4. Verify official sources - check domain, version, date
5. Aggregate by topic - synthesize, don't concatenate
6. Report methodology - tell user how you found info
7. Handle versions explicitly - don't assume latest
8. Fail fast - 30s timeout, then try next method