The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides a utility module (scripts/connections.py) and an evaluation script (scripts/evaluation.py) that facilitate the execution of MCP servers as local subprocesses. This functionality is a core requirement for testing servers during development and relies on user-specified commands and arguments provided via the command-line interface.
[EXTERNAL_DOWNLOADS]: The instructions guide the agent to fetch documentation and specification files from the official modelcontextprotocol.io website and the official SDK repositories on GitHub. These are well-known and trusted sources for the protocol being implemented.
[INDIRECT_PROMPT_INJECTION]: The evaluation harness (scripts/evaluation.py) represents a surface for indirect prompt injection as it processes external data from XML files and the outputs of the tools provided by the MCP server being tested.
Ingestion points: The script parses XML evaluation files containing questions and answers and processes the output of any tools called by the agent during the test loop.
Boundary markers: The system prompt uses specific XML-like tags (<summary>, <feedback>, <response>) to help separate and structure the agent's reasoning from its final output.
Capability inventory: The harness has the capability to execute shell commands to start the MCP server, though this is restricted to the local environment and user-supplied parameters.
Sanitization: Input XML is parsed using standard libraries, and tool outputs are serialized to JSON or strings, providing basic data handling safety.

mcp-builder