nano-banana-pro
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by processing untrusted data from external files and user input to generate or edit images.
- Ingestion points: User-provided text prompts and image files are ingested as command-line arguments and file paths in
generate_image.py,edit_image.py,compose_images.py,generate_with_references.py, andmulti_turn_chat.py. - Boundary markers: Absent; user input is interpolated directly into the API request payloads without delimiters or instructions to ignore embedded commands.
- Capability inventory: The skill possesses network access to the Google Gemini API (
generativelanguage.googleapis.com) and file system read/write access to manage input and output image/text files. - Sanitization: Absent; the skill does not perform validation, filtering, or escaping on the user-provided text instructions before sending them to the model.
Audit Metadata