The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions and scripts fetch documentation and SDK information from official Model Context Protocol repositories and websites, such as modelcontextprotocol.io and github.com/modelcontextprotocol. These are recognized as trusted, well-known sources for this technology.\n- [COMMAND_EXECUTION]: The scripts/evaluation.py and scripts/connections.py files allow developers to execute their own MCP servers for testing purposes. This is accomplished by running user-provided shell commands via the mcp library's stdio_client. This functionality is essential for the skill's purpose as an evaluation harness.\n- [DATA_EXPOSURE]: The evaluation script uses environment variables (e.g., ANTHROPIC_API_KEY) to manage credentials for LLM access, which aligns with security best practices for avoiding hardcoded secrets.\n- [INDIRECT_PROMPT_INJECTION]: The skill provides a framework where an LLM processes user-defined questions from an XML file to test an MCP server. This creates a data ingestion surface, but it is a standard part of the testing workflow and includes clear boundary prompts to guide the LLM's behavior.\n
Ingestion points: The scripts/evaluation.py script reads test questions from a user-supplied XML file.\n
Boundary markers: The script employs a structured EVALUATION_PROMPT using XML tags to isolate task instructions and define output formats.\n
Capability inventory: The harness can launch local processes and call MCP tools on the connected server.\n
Sanitization: XML content is parsed using standard library utilities (xml.etree.ElementTree).

mcp-builder