The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to fetch documentation from official sources including 'modelcontextprotocol.io' and 'github.com/modelcontextprotocol'. These are well-known technical repositories and documentation sites used neutrally to provide the agent with the latest protocol specifications.
[COMMAND_EXECUTION]: The skill includes Python scripts (evaluation.py and connections.py) designed to interact with and test MCP servers. These scripts use the official mcp library to establish connections via standard input/output (stdio), which involves executing the server commands provided by the user. This is a primary and documented feature of the MCP development workflow.
[DATA_EXPOSURE]: The evaluation scripts require an Anthropic API key to function. The documentation correctly instructs users to provide this via environment variables (ANTHROPIC_API_KEY), following industry best practices for secret management. No hardcoded credentials or access to sensitive local files were found.
[INDIRECT_PROMPT_INJECTION]: The evaluation.py script implements an evaluation harness that processes user-provided XML test cases. To mitigate potential instruction override from the test data, the script utilizes a structured system prompt with clear XML boundary markers (<summary>, <feedback>, <response>) to encapsulate the evaluator's output.

mcp-builder