The Agent Skills Directory

[COMMAND_EXECUTION]: The evaluation harness and connection modules facilitate the execution of local shell commands via the stdio transport to launch MCP servers for testing purposes. Evidence: Found in scripts/connections.py and scripts/evaluation.py through the use of stdio_client and command-line arguments -c (command) and -a (args).
[PROMPT_INJECTION]: The scripts/evaluation.py script ingests user-provided XML files containing test questions and interpolates them directly into the LLM prompt, creating a surface for indirect prompt injection. Ingestion points: scripts/evaluation.py reads data using parse_evaluation_file from a user-specified XML file. Boundary markers: No explicit delimiters or instructions to ignore embedded commands are used when interpolating the question into the agent prompt. Capability inventory: The agent loop in scripts/evaluation.py can invoke any tool exposed by the connected MCP server via connection.call_tool. Sanitization: There is no evidence of input validation, escaping, or sanitization of the question content before it is added to the prompt context.
[EXTERNAL_DOWNLOADS]: The skill instructions guide the user to download documentation and SDK references from GitHub repositories associated with the Model Context Protocol. Evidence: SKILL.md contains instructions to fetch README files and protocol specifications from https://raw.githubusercontent.com/modelcontextprotocol/*.

mcp-builder