The Agent Skills Directory

[COMMAND_EXECUTION]: The evaluation script scripts/evaluation.py and connection utility scripts/connections.py include functionality to launch local MCP servers using standard input/output (stdio). This involves executing shell commands and managing subprocesses, which is the primary intended purpose of the evaluation harness.
[EXTERNAL_DOWNLOADS]: The skill fetches official documentation and SDK information from modelcontextprotocol.io and the modelcontextprotocol organization's GitHub repositories. These are well-known, trusted sources for the protocol being implemented.
[DATA_EXFILTRATION]: The evaluation harness (scripts/evaluation.py) communicates with the Anthropic API to process tasks. It requires an API key, which is managed via environment variables according to standard security practices.
[PROMPT_INJECTION]: The skill processes task questions from user-provided XML files. While this creates a surface for indirect prompt injection if the evaluation questions are sourced from an untrusted third party, it is an inherent and expected feature of an evaluation tool.

mcp-builder