The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill frequently generates temporary GDScript files by embedding command-line arguments into f-string templates. These scripts are then executed headlessly using Godot's -s flag. This pattern is central to several scripts, including add_node.py, collision_generator.py, config_compiler.py, create_scene.py, load_sprite.py, tilemap_generator.py, and ui_assembler.py.
[COMMAND_EXECUTION]: The skill uses subprocess.run in scripts/base.py to launch the Godot engine process. Additionally, scripts/stop_project.py executes system commands such as taskkill on Windows or os.kill on Unix-like systems to manage external process lifecycles based on user-supplied PIDs.
[INDIRECT_PROMPT_INJECTION]: The skill processes structured data that could serve as a vector for indirect instruction injection, particularly in workflows that generate scenes or configurations from external JSON or asset metadata.
Ingestion points: Multiple entry points including scripts/tilemap_generator.py (JSON data), scripts/ui_assembler.py (JSON layout), and scripts/config_compiler.py (JSON config).
Boundary markers: The skill does not use explicit boundary markers or sanitization logic when interpolating external strings into executable script templates.
Capability inventory: The skill has permissions to write to the project directory and execute system processes via the Godot binary.
Sanitization: No sanitization or escaping of input data was identified before its inclusion in dynamically generated code.

godot-builder