The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/connections.py facilitates the execution of arbitrary shell commands to initialize MCP servers using the stdio transport method. This is a functional requirement for local server testing but represents a powerful capability.
[PROMPT_INJECTION]: The scripts/evaluation.py script is vulnerable to indirect prompt injection (Category 8). It reads untrusted questions from an XML file and provides them to the agent without sanitization or boundary markers. This allows a crafted evaluation file to potentially manipulate the agent's behavior during the testing phase. Evidence: 1. Ingestion point: parse_evaluation_file function in scripts/evaluation.py reads data from the user-provided XML file. 2. Boundary markers: Absent. The question text is directly inserted into the agent's message context. 3. Capability inventory: The testing agent has access to all tools exposed by the MCP server under evaluation. 4. Sanitization: Absent. The raw question content is not filtered or escaped.

mcp