The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/evaluation.py and scripts/connections.py scripts facilitate the execution of shell commands through the Model Context Protocol's stdio transport mechanism. This is an intended feature for launching and interacting with MCP servers during development and testing.
[EXTERNAL_DOWNLOADS]: The documentation in SKILL.md directs the agent to fetch protocol specifications and SDK README files from well-known sources, including the modelcontextprotocol.io domain and the modelcontextprotocol organization's official GitHub repositories.
[INDIRECT_PROMPT_INJECTION]: The evaluation harness processes external data which presents a theoretical attack surface.
Ingestion points: The scripts/evaluation.py script reads evaluation tasks from XML files and processes output returned by tools during the agent loop.
Capability inventory: The harness can execute shell commands, make network calls, and interact with the Anthropic API.
Boundary markers: The EVALUATION_PROMPT uses structured XML-style tags to help the model distinguish between summary, feedback, and final response content.
Sanitization: The harness does not implement specific sanitization for the evaluation questions or tool outputs before they are processed by the agent.
[DYNAMIC_EXECUTION]: The skill enables the dynamic execution of local server implementations through command-line interaction, allowing the agent to launch and test binaries or scripts designated as MCP servers.

mcp-builder