The Agent Skills Directory

[COMMAND_EXECUTION]: The evaluation script scripts/evaluation.py and the connection handler scripts/connections.py use the mcp library to spawn subprocesses for launching local MCP servers based on user-supplied commands and arguments. This is standard functionality for testing local integrations and is triggered by explicit user input.- [EXTERNAL_DOWNLOADS]: The documentation in SKILL.md references official MCP resources at modelcontextprotocol.io and its associated GitHub repositories. These are well-known, authoritative sources for the protocol, and referencing them for documentation retrieval is standard practice.- [PROMPT_INJECTION]: The evaluation harness presents a surface for indirect prompt injection through its ingestion of external data. Ingestion points: Test questions are loaded from user-provided XML files, and the agent processes results from tools provided by the MCP server being tested. Boundary markers: The script uses a system prompt with XML tags to structure the model's output, but does not employ strict delimiters for third-party tool content. Capability inventory: The tool can execute local shell commands to start servers and perform network requests to the Anthropic API. Sanitization: External content is passed to the LLM without specific sanitization or escaping.

mcp-builder