The Agent Skills Directory

[DYNAMIC_EXECUTION]: The bridge server implementation in scripts/mx_bridge_server.py uses exec() and compile() to execute Python code received from the bridge client. While the code is read from a local directory, this pattern allows for arbitrary code execution within the Maya environment.
[COMMAND_EXECUTION]: The skill makes extensive use of system commands to drive external processes.
scripts/mx_run.py uses subprocess.Popen to launch the mayapy interpreter.
scripts/mx_review.py uses subprocess.check_call to execute ffmpeg for video encoding and imconvert for image processing.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external Autodesk Maya scene files (.ma, .mb) which may contain embedded malicious logic in 'script nodes'. While scripts/mx_validate.py includes checks for known malicious script nodes (e.g., vaccine_gene, breed_gene), the inherent capability of the agent to parse and execute scene-embedded data remains a potential attack surface.
[REMOTE_CODE_EXECUTION]: In scripts/mx_bridge_server.py, the start() function opens a Maya commandPort on 127.0.0.1. As noted in the script's own documentation, this port has no authentication, allowing any process on the local machine to send and execute arbitrary Python or MEL commands with the permissions of the Maya user.

scenario-maya-expert