3d-asset-generation
Pass
Audited by Gen Agent Trust Hub on Nov 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in the form of text descriptions and concept images, which are then used as inputs for external 3D generation tools.
- Ingestion points: User-supplied object descriptions in
atlas_3dand concept images infal_3d(Hunyuan/SAM). - Boundary markers: The skill lacks explicit boundary markers or instructions to ignore potential hidden commands within the user-provided asset descriptions or image metadata.
- Capability inventory: The skill utilizes tools for network API calls (
atlas_3d,fal_3d), file system operations (downloading artifacts), and complex scene manipulation viablender_world. - Sanitization: While the instructions provide a 'prompt contract' and 'mesh QA' process, these are focused on asset quality rather than sanitizing inputs to prevent the generation model from being manipulated by adversarial instructions embedded in the input data.
Audit Metadata