godot-genre-sandbox
Warn
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The load_world function in scripts/sandbox_world_serializer.gd implements a pattern that uses load() and instantiate() on file paths retrieved from a JSON data structure (entry.res). This allows for the dynamic loading of arbitrary scenes and scripts based on the contents of the user://sandbox_world.json file.
- [INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface by ingesting untrusted data from the clipboard and local storage. * Ingestion points: The BlueprintManager in SKILL.md reads from DisplayServer.clipboard_get(), and scripts/sandbox_world_serializer.gd reads from a local JSON file. * Boundary markers: None are present to distinguish between trusted and untrusted scene paths. * Capability inventory: Includes load(), instantiate(), add_child(), and FileAccess.WRITE. * Sanitization: There is no evidence of path validation, allow-listing, or sanitization before the resource paths are loaded into the engine context.
Audit Metadata