The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to fetch protocol documentation and SDK information from official Model Context Protocol domains and GitHub repositories.
Evidence: Fetches from modelcontextprotocol.io and raw.githubusercontent.com/modelcontextprotocol/* for context gathering.
[COMMAND_EXECUTION]: The evaluation harness (scripts/evaluation.py) and connection utility (scripts/connections.py) allow for the execution of local MCP servers as subprocesses.
Evidence: Uses stdio_client to launch servers via user-provided commands and arguments for testing purposes.
[INDIRECT_PROMPT_INJECTION]: The evaluation script processes external XML files containing questions that are interpolated into the agent's system prompt during the testing loop.
Ingestion points: scripts/evaluation.py reads task questions from user-provided XML files.
Boundary markers: Questions are structured within <question> XML tags but lack additional prompt-level delimiters or instructions to ignore embedded commands.
Capability inventory: The agent has access to the full toolset of the MCP server being evaluated, which may include file system or network operations.
Sanitization: The script does not perform sanitization or escaping of the natural language content within the evaluation questions.

mcp-builder