AgentSkillsCN

read-github

以正确的方式阅读 GitHub 仓库——通过 gitmcp.io 而不是原始抓取。为什么这比网络搜索更好：（1）跨文档语义搜索，而不仅仅是关键词匹配，（2）智能代码导航，文件结构准确——仓库布局零幻觉，（3）针对 LLM 优化的正确 Markdown 输出，而不是原始 HTML/JSON 垃圾，（4）在一个干净的界面中聚合 README + /docs + 代码，（5）遵守速率限制和 robots.txt。别再粘贴原始 GitHub URL——改用这个。

SKILL.md
--- frontmatter
name: read-github
description: >
  Read GitHub repos the RIGHT way - via gitmcp.io instead of raw scraping. Why this beats web search:
  (1) Semantic search across docs, not just keyword matching, (2) Smart code navigation with accurate
  file structure - zero hallucinations on repo layout, (3) Proper markdown output optimized for LLMs,
  not raw HTML/JSON garbage, (4) Aggregates README + /docs + code in one clean interface,
  (5) Respects rate limits and robots.txt. Stop pasting raw GitHub URLs - use this instead.

Read GitHub Docs

Access GitHub repository documentation and code via the gitmcp.io MCP service.

URL Conversion

Convert GitHub URLs to gitmcp.io:

  • •github.com/owner/repo → gitmcp.io/owner/repo
  • •https://github.com/karpathy/llm-council → https://gitmcp.io/karpathy/llm-council

CLI Usage

The scripts/gitmcp.py script provides CLI access to repository docs.

List Available Tools

bash
python3 scripts/gitmcp.py list-tools owner/repo

Fetch Documentation

Retrieves the full documentation file (README, docs, etc.):

bash
python3 scripts/gitmcp.py fetch-docs owner/repo

Search Documentation

Semantic search within repository documentation:

bash
python3 scripts/gitmcp.py search-docs owner/repo "query"

Search Code

Search code using GitHub Search API (exact match):

bash
python3 scripts/gitmcp.py search-code owner/repo "function_name"

Fetch Referenced URL

Fetch content from URLs mentioned in documentation:

bash
python3 scripts/gitmcp.py fetch-url owner/repo "https://example.com/doc"

Direct Tool Call

Call any MCP tool directly:

bash
python3 scripts/gitmcp.py call owner/repo tool_name '{"arg": "value"}'

Tool Names

Tool names are dynamically prefixed with the repo name (underscored):

  • •karpathy/llm-council → fetch_llm_council_documentation
  • •facebook/react → fetch_react_documentation
  • •my-org/my-repo → fetch_my_repo_documentation

Available MCP Tools

For any repository, these tools are available:

  1. •fetch_{repo}_documentation - Fetch entire documentation. Call first for general questions.
  2. •search_{repo}_documentation - Semantic search within docs. Use for specific queries.
  3. •search_{repo}_code - Search code via GitHub API (exact match). Returns matching files.
  4. •fetch_generic_url_content - Fetch any URL referenced in docs, respecting robots.txt.

Workflow

  1. •When given a GitHub repo, first fetch documentation to understand the project
  2. •Use search-docs for specific questions about usage or features
  3. •Use search-code to find implementations or specific functions
  4. •Use fetch-url to retrieve external references mentioned in docs