name: github-intel
description: This skill should be used when the user asks to "discover trending repos", "find GitHub repositories", "research AI coding tools", "crawl GitHub for best practices", "extract code patterns from repos", "build knowledge from GitHub", "find Claude Code extensions", "discover MCP servers", or mentions finding valuable open source code, extracting architecture patterns, or building compound documentation from repositories.
version: 1.0.0

GitHub Intelligence Skill

Automated discovery and analysis of GitHub repositories to extract valuable code, architecture patterns, best practices, and compound documentation. Uses subagents for parallel research and knowledge extraction.

Overview

This skill provides a complete pipeline for:
1. Discovery - Find trending repos matching keywords (Claude Code, Codex, MCP, agents, etc.)
2. TODO Generation - Create structured exploration tasks for each repo
3. Exploration - Clone and analyze repos with specialized subagents
4. Extraction - Capture valuable patterns to knowledge store
5. Indexing - Semantic vector search via QMD integration

Quick Start

# Discover trending repos
./crawler/run.sh discover

# Check discovery status
./crawler/run.sh status

# Explore a specific repo
./crawler/run.sh explore <repo-name>

# Extract knowledge
./crawler/extract.sh add <category> <name>

# Search indexed repos (requires QMD)
qmd vsearch "multi-agent orchestration" -c ai-coding-repos

Discovery Pipeline

Step 1: Configure Search

Edit crawler/config.sh to customize:

# Minimum stars threshold
MIN_STARS=50

# Days since last update
DAYS_AGO=90

# Keywords to search
KEYWORDS=(
    "claude-code"
    "codex cli"
    "mcp server"
    "ai coding assistant"
    "llm code generation"
)

Step 2: Run Discovery

./crawler/run.sh discover

Outputs:
- Individual TODO files in todos/repos/
- Index file at todos/repos/INDEX.md
- Sorted by stars, deduplicated

Step 3: Index for Semantic Search

# Create QMD collection
qmd collection add todos/repos --name ai-coding-repos --mask "*.md"

# Generate vector embeddings
qmd embed

Subagent Orchestration

Exploration Subagent

Triggered via ./crawler/run.sh explore <repo-name>:

1. Clones repository to .clones/
2. Analyzes codebase structure
3. Identifies key files and patterns
4. Updates TODO with findings

Use the exploration prompt template:

cat crawler/prompts/explore.md

Review Subagent

For deep code analysis, spawn a review subagent:

Launch Explore agent to analyze the codebase at .clones/<repo-name>

Focus on:
1. Architecture patterns worth extracting
2. Reusable utilities and helpers
3. Novel approaches to common problems
4. Code quality and best practices

Compound-Docs Subagent

For knowledge extraction:

Launch agent to create compound documentation for <pattern-name>

Extract from .clones/<repo-name>:
1. Core concept explanation
2. Implementation details
3. Code examples
4. Usage patterns

Knowledge Store Structure

knowledge/
├── architecture/     # System design patterns
├── patterns/         # Code patterns and idioms
├── utilities/        # Reusable helper code
├── frameworks/       # Framework-specific knowledge
└── INDEX.md          # Master index

Adding Knowledge

# Interactive extraction
./crawler/extract.sh add architecture "plugin-system"
./crawler/extract.sh add patterns "hook-interceptor"
./crawler/extract.sh add utilities "regex-cache"

Knowledge Entry Format

Each entry in knowledge/<category>/<name>.md:

# Pattern Name

**Source:** repo-name
**Category:** architecture|patterns|utilities|frameworks

## Overview
Brief description of the pattern.

## Implementation
Code examples and details.

## Usage
When and how to apply this pattern.

## References
Links to source files.

Workflow Examples

Example 1: Discover Claude Code Extensions

# Set focused keywords
export CRAWLER_KEYWORDS="claude-code skill hook mcp-server"

# Run discovery
./crawler/run.sh discover

# Check results
./crawler/run.sh status

Example 2: Deep Dive into a Repo

# Explore the repo
./crawler/run.sh explore anthropics-claude-code

# Launch exploration agent
# (In Claude Code session)
> Explore the codebase at .clones/anthropics-claude-code
> Focus on plugin architecture and hook system

Example 3: Extract and Document Pattern

# After exploring, extract pattern
./crawler/extract.sh add patterns "progressive-disclosure"

# Edit the generated file
# Add implementation details from exploration

Example 4: Semantic Search

# Search for specific concepts
qmd vsearch "multi-agent workflow" -c ai-coding-repos -n 10

# Combined search with reranking
qmd query "how to implement hooks" -c ai-coding-repos

Prompt Templates

Exploration Prompt

Located at crawler/prompts/explore.md:
- Initial codebase analysis
- Structure discovery
- Key file identification
- Pattern recognition

Review Prompt

Located at crawler/prompts/review.md:
- Deep code analysis
- Quality assessment
- Best practice identification
- Improvement suggestions

Compound-Docs Prompt

Located at crawler/prompts/compound.md:
- Knowledge extraction format
- Documentation structure
- Example generation
- Cross-referencing

Integration with Other Skills

With /commit

After extracting knowledge:

git add knowledge/
/commit

With Code Review

Spawn parallel review agents:

Launch 3 review agents in parallel to analyze:
1. .clones/repo-a - focus on architecture
2. .clones/repo-b - focus on patterns
3. .clones/repo-c - focus on utilities

With QMD Search

# Find related patterns
qmd query "your search" -c ai-coding-repos --files

# Get full document
qmd get qmd://ai-coding-repos/repo-name.md

Scripts Reference

Configuration Options

Additional Resources

Reference Files

• references/workflow.md - Detailed workflow guide
• references/subagents.md - Subagent patterns
• references/qmd-integration.md - QMD search setup

Example Files

• examples/discovery-output.json - Sample discovery results
• examples/knowledge-entry.md - Template for knowledge entries

Scripts

• scripts/batch-explore.sh - Batch exploration utility
• scripts/sync-qmd.sh - QMD index synchronization