read-github

Savant verdict: Unsafe

Savant's security policy blocks importing this skill.

Live evaluation

Not evaluated yet. Workspaces can request a live evaluation.

Security assessment

5 patterns found

• Agent Snooping: List available MCP toolshigh · scripts/gitmcp.py:169 · Skill accesses MCP server configuration files (mcp.json). MCP configs contain server URLs, authentication tokens, and tool definitions — reading them allows the skill to discover and potentially abuse other tool integrations.
• MCP Least Privilege: Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: shell.medium · SKILL.md:1 · Without declared permissions the skill's intent is opaque and cannot be validated.
• Dangerous Code Execution: subprocess.Popen( ["npx", "-y", "mcp-remote", mcp_url], stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True, bufsize=1medium · scripts/gitmcp.py:76 · subprocess module calls execute external commands. Without careful input validation, this enables command injection.
• Dangerous Code Execution: subprocess.Popen( ["npx", "-y", "mcp-remote", mcp_url], stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True, bufsize=1medium · scripts/gitmcp.py:107 · subprocess module calls execute external commands. Without careful input validation, this enables command injection.
• Excessive Agency: Call any toolmedium · scripts/gitmcp.py:161 · Skill grants unrestricted tool access without appropriate constraints. An agent with unfettered tool access can perform arbitrary actions including file modification, network requests, and code execution.

Structure

• Ships 1 executable scriptscripts/gitmcp.py. Review what they do before enabling the skill for agents with tool access.
• No license declaredConfirm you may reuse this skill before importing it into your repository.
• The source hub's security checks flagged this skillGen Agent Trust Hub: warn (This skill accesses GitHub repositories via an external service called gitmcp.io. It downloads and executes a package from the NPM registry at runtime to establish this connection. Furthermore, it is vulnerable to indirect prompt injection, as it processes untrusted documentation and code directly from GitHub repositories without sanitization.); Socket: warn (2 alerts: gptAnomaly); Snyk: warn (Risk: MEDIUM · 2 issues); Runlayer: warn (1/2 files flagged)

SKILL.md

---
name: read-github
description: |
  Read and search GitHub repository documentation via gitmcp.io MCP service.

  **WHEN TO USE:**
  - User provides a GitHub URL
  - User mentions a specific repo in owner/repo format
  - User asks "what does this repo do?", "read the docs for X repo", or similar
  - User wants to search code or docs within a repo
---

# 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