The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill generates temporary GDScript files (e.g., in add_node.py, collision_generator.py, ui_assembler.py) by interpolating CLI arguments directly into code templates. These scripts are then executed via the Godot engine's CLI (-s flag). A malicious actor could provide input designed to execute arbitrary code within the Godot environment, which has system-level access through Godot's internal APIs.
[COMMAND_EXECUTION]: The GodotBase class in scripts/base.py and the stop_project.py script utilize subprocess.run to execute shell commands. While used for legitimate build tasks, this capability combined with the dynamic script generation creates a path for unauthorized system command execution.
[PROMPT_INJECTION]: The skill is highly vulnerable to indirect prompt injection. Because it processes external data (such as filenames and resource types) by embedding them directly into executable scripts, a malicious file or data source could be used to compromise the agent's execution environment.

godot-builder