The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill exposes the idapython_snippet() SQL function, which enables the execution of arbitrary Python code strings directly within the IDA Pro process. This creates a powerful sink for dynamic code execution that has full access to the IDA SDK and the host system's Python environment.
[COMMAND_EXECUTION]: The idapython_file() function allows the agent to execute any Python script located on the filesystem accessible to the IDA process. This can be used to run unauthorized scripts if an attacker manages to place a file on the local disk.
[INDIRECT_PROMPT_INJECTION]: The skill establishes a high-privilege execution surface for the AI agent. If the agent processes untrusted data (such as strings, function names, or comments extracted from a binary) and interpolates that data into a Python snippet, it could lead to the execution of malicious code embedded in the analyzed binary.
[DATA_EXFILTRATION]: Because the IDAPython environment has access to the complete IDA database (including disassembly, cross-references, and strings) and standard Python network libraries (e.g., requests, socket), this skill can be used to exfiltrate sensitive intellectual property from the files being analyzed.

idapython