The Agent Skills Directory

[COMMAND_EXECUTION]: The Python helper scripts/gd_architecture.py invokes the Godot binary in headless mode using subprocess.run. This is a controlled execution used to perform smoke tests on project exports.
[DYNAMIC_EXECUTION]: The skill utilizes Godot's load() and instantiate() methods in scripts like arch_audit.gd to dynamically audit scene structures. This is a standard and necessary function for project-level tooling.
[DYNAMIC_EXECUTION]: scripts/gd_architecture.py dynamically modifies sys.path to resolve dependencies from the required sibling skill, scenario-godot-expert.
[INDIRECT_PROMPT_INJECTION]: The skill processes Godot project files and resources, creating a surface for indirect prompt injection.
Ingestion points: arch_audit.gd and validate_data.gd load and process various Godot resource and scene files (.tscn, .tres, .gd) from the local project.
Boundary markers: Absent for static code and scene auditing processes.
Capability inventory: Includes project file manipulation via FileAccess/DirAccess, scene instantiation, and execution of the Godot engine via subprocess.run.
Sanitization: The skill proactively mitigates data-driven attacks by enforcing a JSON-based save system (save_service.gd) that uses JSON.to_native and JSON.from_native. This pattern specifically prevents the instantiation of malicious objects or embedded scripts from untrusted data files, addressing a known remote code execution vector in the Godot engine.

scenario-godot-architecture