nano-banana-pro

Pass

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied prompts and input images by passing them to the Gemini model. While inherent to image generation tools, this creates an attack surface for indirect prompt injection.
  • Ingestion points: Data enters via the --prompt and --input-image CLI arguments in scripts/generate_image.py.
  • Boundary markers: The script does not employ delimiters or explicit system instructions to isolate user input from model logic.
  • Capability inventory: The skill possesses network access to Google API endpoints and local file-writing capabilities through the Pillow library.
  • Sanitization: No validation or sanitization is performed on input prompts or the provided output filename.
  • [SAFE]: The skill uses the official google-genai library and targets trusted Google infrastructure. Credentials are managed via standard environment variables, and the code contains no obfuscation, unauthorized privilege escalation, or persistence attempts.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 17, 2026, 07:54 PM
Security Audit — agent-trust-hub — nano-banana-pro