The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides functions idapython_snippet(code) and idapython_file(path) that directly execute arbitrary Python code within the IDA Pro process context.
Evidence: SKILL.md defines SQL functions that map directly to Python execution in the IDA runtime.
[COMMAND_EXECUTION]: By enabling Python execution, the agent can use standard Python libraries (e.g., os, subprocess) to execute arbitrary shell commands on the host operating system.
Evidence: Once PRAGMA idasql.enable_idapython = 1 is set, any Python snippet can spawn processes or modify files outside the scope of IDA Pro.
[DATA_EXFILTRATION]: Python execution combined with the requests library (mentioned in examples) or standard socket libraries allows for the exfiltration of sensitive analysis data, database contents, or local files to remote servers.
Evidence: The example in SKILL.md demonstrates using requests.post to send data to an external (though local) endpoint.
[DYNAMIC_EXECUTION]: The skill is specifically designed to create and execute code at runtime through SQL query interpolation.
Evidence: The 'Bridge pattern' example shows Python producing JSON which is then processed by SQL, creating a complex feedback loop where generated data can influence subsequent code execution.

idapython