github-kb

GitHub Knowledge Base

Unified skill for local repo management + social media hot topic tracking.

Configuration

KB_PATH: Default ~/github. Customize in project CLAUDE.md:

KB_PATH: /your/path

Commands

/github-kb search <keyword>

Search local knowledge base.

1. Read KB_PATH/CLAUDE.md index
2. Match keyword against names, descriptions, tags
3. Display matches with paths
4. Offer to explore repo details

/github-kb update

Sync index with filesystem.

1. Scan KB_PATH for .git directories
2. Extract: name, README first line, tech stack tags
3. Diff against existing CLAUDE.md
4. Show changes, ask confirmation, write update

Index format (keep under 100 lines):

# GitHub Knowledge Base

| Name | Description | Path | Tags |
|------|-------------|------|------|
| repo | One-line desc | ./repo | tag1, tag2 |

/github-kb discover <topic>

Find trending repos from multiple sources.

1. Gather (parallel):

   • Xiaohongshu: See references/xiaohongshu.md
   • Twitter/X: See references/twitter.md
   • Web: mcp__tavily__tavily-search query "<topic> github"

2. Extract GitHub URLs from results

3. Enrich via gh repo view owner/repo --json name,description,stargazerCount

4. Present candidates with source attribution

5. Clone user-selected repos: gh repo clone owner/repo

6. Update index automatically

/github-kb xiaohongshu <keyword>

Direct Xiaohongshu search. See references/xiaohongshu.md.

/github-kb twitter <url>

Fetch Twitter/X content. See references/twitter.md.

Tools Reference

Source	Tool/Method
Local repos	gh CLI, filesystem
Xiaohongshu	mcp__rednote__* tools
Twitter/X	Jina API via scripts/fetch_tweet.py
Web search	mcp__tavily__tavily-search

AI-Driven Query (DeepWiki Mode)

When solving problems, proactively search the knowledge base:

1. Auto-detect relevance: If user asks about a technology/tool, check KB first
2. Query before web search: Prefer local repos over external docs
3. Deep dive: Read repo's README, docs/, examples/ for answers

Trigger phrases (AI should recognize):

• "How does X work?"
• "Show me an example of X"
• "What's the best practice for X?"

Workflow:

User asks question → Check KB index → Found relevant repo?
  → Yes: Read repo docs, provide answer with citation
  → No: Use discover to find repos, or fall back to web search

Example:

User: How do I use MCP tools in Claude?
AI: [Checks KB, finds mcp-servers repo]
    Let me check your local knowledge base...
    Found: mcp-servers in ~/github/mcp-servers
    [Reads README.md and examples/]
    Based on your local repo, here's how to use MCP tools...

Example Flows

Discover new tool:

/github-kb discover Claude MCP
→ [Searches Xiaohongshu, Tavily]
→ Found 3 repos, select to clone
→ Index updated

Research on Xiaohongshu:

/github-kb xiaohongshu Python数据分析
→ [Returns top notes with summaries]

Fetch tweet:

/github-kb twitter https://x.com/user/status/123
→ [Returns full tweet content]