nano-banana-2
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided images as references, creating a surface for indirect prompt injection where instructions hidden within images could influence the model.\n
- Ingestion points: Reference image paths provided via the
--input-imageflag inscripts/generate_image.py.\n - Boundary markers: Input images and prompts are passed as a unified context to the API without delimiters or directives to ignore embedded instructions.\n
- Capability inventory: The skill has network access for API calls and the ability to write generated image files to the filesystem.\n
- Sanitization: No content-based validation or sanitization is performed on input images before they are passed to the model.
Audit Metadata