The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data via game save resources (.tres) and dialogue files (.dialogue), creating a potential attack surface for indirect instructions.
Ingestion points: save_io.gd and gd_gameplay.py ingest and process external Godot resource files and dialogue cues.
Boundary markers: The skill uses explicit security wrappers like load_resource_checked to identify and block unauthorized scripts.
Capability inventory: The skill possesses file system write access (atomic saves, CSV logging) and can execute Godot-specific tasks through the platform's execution utility.
Sanitization: Active sanitization is performed via regex-based scanning in save_io.gd and gd_gameplay.py, which identifies embedded GDScript or CSharpScript within text resources before they are loaded by the engine.
[COMMAND_EXECUTION]: The gd_gameplay.py script and crowd_bench.gd facilitate running Godot benchmarks and project installations. These utilize gd_run to interface with the Godot binary for project setup and performance measurement, which is standard behavior for development-focused skills.
[EXTERNAL_DOWNLOADS]: The documentation references external Godot community resources and addons from well-known repositories on GitHub (e.g., Nathan Hoad's Dialogue Manager and Limbonaut's LimboAI). These references are provided for expert knowledge and manual integration by the developer, rather than automated execution of remote code.

godot-gameplay