The Agent Skills Directory

[COMMAND_EXECUTION]: The editor_execute_menu skill allows the agent to trigger arbitrary menu items within the Unity Editor. While the documentation notes that this requires exact paths and is subject to platform-level permission checks, it provides a broad interface for controlling application behavior.
[DYNAMIC_EXECUTION]: The editor_run_scenario skill supports an invoke step that calls component_invoke_method, enabling the execution of arbitrary code within the Unity runtime environment. The skill documentation explicitly flags this as high-risk and mentions that it requires specific platform bypass or allowlist configurations.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the local Unity environment, which could potentially contain malicious instructions embedded by an external actor in the project.
Ingestion points: Data enters the agent context through editor_get_context, editor_get_changes, editor_playmode_inspect, and console logs captured by editor_play_capture (specifically object names, component fields, and log messages).
Boundary markers: The documentation does not specify the use of delimiters or 'ignore' instructions when processing strings read from the Unity Editor.
Capability inventory: The skill provides access to significant side-effect capabilities, including menu execution, undo/redo operations, and arbitrary method invocation in the Unity runtime.
Sanitization: No evidence of input sanitization or validation for the data read from the Unity Editor is provided in the instruction file.

unity-editor