The Agent Skills Directory

[COMMAND_EXECUTION]: The script located at scripts/evaluation.py allows for the execution of arbitrary shell commands via the stdio transport configuration. It accepts user-defined commands and arguments to spawn subprocesses, which is the intended method for testing MCP servers but provides a vector for system-level execution.
[EXTERNAL_DOWNLOADS]: The instructions guide the agent to fetch protocol documentation and SDK README files from modelcontextprotocol.io and GitHub. These references target well-known and official repositories for the Model Context Protocol ecosystem and are considered safe.
[PROMPT_INJECTION]: The evaluation process in scripts/evaluation.py is susceptible to indirect prompt injection from the server being tested.
Ingestion points: Data returned by the tools of the MCP server is read and included directly in the agent's prompt history.
Boundary markers: The script does not utilize explicit delimiters or 'ignore instructions' warnings to isolate tool output from the agent's core instructions.
Capability inventory: The evaluation script possesses capabilities to execute subprocesses and initiate network connections via SSE or HTTP.
Sanitization: Tool results are converted to strings and appended to the message context without filtering or sanitization.

mcp-builder