media-higgsfield-product
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied text and images to assemble image generation prompts. This represents a potential surface for indirect prompt injection. Ingestion points: User text requests and uploaded product images processed via media_id. Boundary markers: The skill enforces strict 10-mode prompt structures and categorizes inputs before assembly. Capability inventory: Uses model exploration and image generation tools mediated via the media-higgsfield-core framework. Sanitization: Implements a mandatory Approval Gate (Step 5) where the final constructed prompt and estimated cost must be explicitly confirmed by the user before execution.
- [EXTERNAL_DOWNLOADS]: The skill instructions reference logic and mode definitions derived from the official Higgsfield AI GitHub repository. This reference to a well-known service provider is used for documentation and alignment with the underlying model's capabilities.
Audit Metadata