The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/evaluation.py script and scripts/connections.py module facilitate executing local shell commands via the stdio transport. This allows the harness to launch and manage the MCP server process being tested (e.g., using -c python -a server.py as arguments). This is the intended primary purpose of the evaluation script and is necessary for testing local implementations.
[EXTERNAL_DOWNLOADS]: The skill guides and the SKILL.md file suggest fetching protocol specifications and SDK documentation from official sources, including modelcontextprotocol.io and github.com/modelcontextprotocol. These are recognized as trusted, well-known organizations for the Model Context Protocol.
[PROMPT_INJECTION]: The scripts/evaluation.py script exhibits a surface for indirect prompt injection by processing external data.
Ingestion points: The script reads test questions from a user-supplied XML file provided via the eval_file positional argument.
Boundary markers: Absent. The questions are interpolated directly into the message list sent to the LLM without specific protective delimiters.
Capability inventory: The agent in the evaluation loop can execute any tool exposed by the connected MCP server (specified via the transport CLI arguments).
Sanitization: None. The content of the XML file is passed directly to the model. This risk is inherent to the tool's primary function of evaluating agent performance against arbitrary test sets.

mcp-builder