The Agent Skills Directory

COMMAND_EXECUTION (MEDIUM): The MCPConnectionStdio class in scripts/connections.py is designed to execute system commands via the mcp.client.stdio transport. While this is a standard feature of the Model Context Protocol, it provides a direct path for arbitrary command execution if an attacker can influence the command or args parameters passed to the create_connection factory.
EXTERNAL_DOWNLOADS (LOW): The MCPConnectionSSE and MCPConnectionHTTP classes in scripts/connections.py facilitate connections to remote URLs. This capability can be used for server-side request forgery (SSRF) or data exfiltration if the agent is manipulated into connecting to an attacker-controlled endpoint.
INDIRECT_PROMPT_INJECTION (SAFE): The file scripts/example_evaluation.xml contains structured QA data. While this data is processed by the agent, it consists of standard mathematical and scientific questions and does not currently contain malicious injection markers. However, the use of external data inputs always presents a minor surface for indirect injection.

mcp-builder