The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The Python script scripts/gitmcp.py executes npx -y mcp-remote, which downloads the mcp-remote package from the NPM registry every time a tool is called if it is not already cached. This introduces a runtime dependency on an external package registry.
[REMOTE_CODE_EXECUTION]: The skill uses npx to run the mcp-remote client which connects to an external URL (gitmcp.io). This allows the execution of remote logic provided by the MCP service based on user-supplied repository identifiers.
[COMMAND_EXECUTION]: Shell commands are executed via subprocess.Popen in scripts/gitmcp.py to bridge the Python CLI to the Node.js npx utility. While the arguments are passed as a list, this pattern is used to invoke external processes with user-influenced arguments.
[INDIRECT_PROMPT_INJECTION]: The skill's primary purpose is to ingest documentation and code from GitHub repositories, which are untrusted external sources.
Ingestion points: Untrusted data enters the agent context through the fetch-docs, search-docs, and search-code commands in scripts/gitmcp.py.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present in the script's output or the skill's instructions.
Capability inventory: The skill possesses the ability to fetch arbitrary URLs mentioned in documentation and execute various tools on the remote MCP server.
Sanitization: The skill does not perform any validation, filtering, or escaping of the content retrieved from GitHub repositories before presenting it to the agent.
[DYNAMIC_EXECUTION]: The use of npx to fetch and run code at runtime constitutes dynamic execution, as the behavior of the skill depends on the version and content of the package fetched from the NPM registry at the moment of execution.

read-github