The Agent Skills Directory

[COMMAND_EXECUTION]: The Python scripts (ut_run.py, ut_live.py, ut_env.py) use subprocess.Popen and subprocess.run to launch the Unity Editor, the Unity CLI binary, and system utilities like ps (for project lock detection), cp (for APFS project cloning), and xcodebuild. This is the intended mechanism for wrapping Unity's command-line interface.
[DYNAMIC_EXECUTION]: The C# AgentBridge.cs component uses reflection (MethodInfo.Invoke) to execute static methods by name when requested through a file-based inbox in the project's Library/ folder. Additionally, ut_live.py implements an eval feature by generating C# source files at runtime, which are then compiled and executed by the Unity Editor.
[EXTERNAL_DOWNLOADS]: The documentation and scripts refer to the official Unity CLI and its installer hosted on public-cdn.cloud.unity3d.com. These downloads target a well-known service associated with the Unity ecosystem.
[PERSISTENCE]: The AgentBridge.cs class is marked with [InitializeOnLoad], which ensures that the RPC bridge automatically starts whenever the Unity Editor loads the project. This allows the agent to maintain a connection to a GUI editor session.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external data from Unity log files, NUnit test results (XML), and asset audit facts (JSON). It employs a specific boundary marker (AGENT_RESULT) to distinguish structured job output from standard engine log noise, reducing the risk of schema confusion.

scenario-unity-expert