The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a connection handling module (connections.py) and an evaluation harness (evaluation.py) that allow for the execution of local MCP servers via standard I/O commands provided by the user. This is a functional requirement for testing developmental servers.
[INDIRECT_PROMPT_INJECTION]: The evaluation script processes qa_pair entries from XML files, which are then interpolated into an agent prompt. While this presents an indirect prompt injection surface, the risk is inherent to the tool's purpose of testing LLM tool-use capabilities.
Ingestion points: scripts/evaluation.py (via parse_evaluation_file reading XML input)
Boundary markers: The prompt uses specific XML tags (<summary>, <feedback>, <response>) to structure the LLM's output, providing a measure of structural isolation.
Capability inventory: The evaluator agent can invoke any tool exposed by the connected MCP server using the connection.call_tool method.
Sanitization: Input questions from the XML are processed as plain text without specialized sanitization.
[EXTERNAL_DOWNLOADS]: The documentation guides the user to fetch the latest protocol specifications and SDK details from modelcontextprotocol.io and its official GitHub repository. These are recognized sources for the protocol being developed.

mcp-builder