read-github

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.90). This skill fetches and ingests public GitHub repository documentation and code via gitmcp.io (commands like fetch-docs, search-docs, search-code) and can fetch arbitrary external URLs via fetch_generic_url_content/fetch-url, so the agent will read untrusted, user-generated third-party content that could carry indirect prompt injections.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill launches an MCP client at runtime pointed to the converted GitHub URL (e.g. https://gitmcp.io/owner/repo) and uses JSON‑RPC tool calls (tools/list, tools/call, fetch_generic_url_content) whose responses and tool executions from the remote gitmcp.io server directly control what the agent does, so the external gitmcp.io URL is a runtime dependency that can control prompts/execute code.