The Agent Skills Directory

[SAFE]: The skill provides documentation and tools for a legitimate developer workflow (MCP server creation and evaluation).
[EXTERNAL_DOWNLOADS]: The skill suggests fetching documentation from the official Model Context Protocol GitHub repositories (specifically modelcontextprotocol/python-sdk and modelcontextprotocol/typescript-sdk). These are well-known and official sources for the protocol.
[COMMAND_EXECUTION]: The evaluation.py script is a testing harness that executes a user-specified command to launch an MCP server for evaluation. This is standard functionality for local development and testing tools.
[INDIRECT_PROMPT_INJECTION]: The evaluation script processes external XML data and tool outputs from the server being tested. It uses clear boundary markers in its system prompt (e.g., <summary>, <feedback>, <response> tags) to maintain control over the LLM's output structure during testing.

mcp-builder