The Agent Skills Directory

[COMMAND_EXECUTION]: The toolkit uses subprocess.Popen and subprocess.run across several scripts (ue_run.py, ue_env.py, ue_remote.py) to launch the Unreal Editor, execute commandlets, and run system diagnostics like sw_vers, sysctl, and lsof for environment preflight and process management.
[DYNAMIC_EXECUTION]: The skill employs dynamic execution patterns to facilitate agent tasks. ue_run.py stages temporary Python boot scripts and utilizes runpy.run_path to execute agent-provided jobs within the Unreal Engine environment. Additionally, ue_remote.py generates Python payloads in its call_code function to be sent and executed in a running editor via the Python Remote Execution API.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a significant ingestion surface for untrusted external data, including Unreal project files (.uproject, .ini), engine trace logs, and image assets. While the skill incorporates safety procedures such as scoped transactions and human-in-the-loop verification lists, the combination of processing untrusted project-level data with the skill's extensive file-write and process-control capabilities presents an inherent indirect prompt injection surface.

unreal-expert