The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The documentation references official Model Context Protocol specifications and SDKs hosted on GitHub (e.g., github.com/modelcontextprotocol). These are well-known technology resources and are documented neutrally for implementation reference.\n- [COMMAND_EXECUTION]: The evaluation script in scripts/evaluation.py allows launching local processes to connect to MCP servers via standard input/output. This is the standard operational mode for local MCP integrations and is explicitly controlled by user-provided command-line arguments.\n- [PROMPT_INJECTION]: The skill provides an evaluation harness that ingests questions from user-provided XML files and includes them in prompts sent to an LLM. Ingestion occurs in scripts/evaluation.py via the parse_evaluation_file function. Boundary markers are present in the EVALUATION_PROMPT which uses XML tags to structure the agent output. The agent possesses tool-calling capabilities through the connection.call_tool method. No explicit sanitization or filtering is performed on the ingested question content. This surface is intended for validating server logic in a development environment.\n- [SAFE]: Analysis of the Python scripts (evaluation.py, connections.py) and documentation reveals no signs of malicious intent, credential theft, or unauthorized network activity. The code follows established patterns for building AI-integrated tools.

mcp-builder