nano-banana-pro

Fail

Audited by Gen Agent Trust Hub on Feb 15, 2026

Risk Level: HIGHPROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
  • [PROMPT_INJECTION] (HIGH): High risk of Indirect Prompt Injection through untrusted external data.
  • Ingestion points: Untrusted image data enters the context via the --input-image argument in scripts/generate_image.py.
  • Boundary markers: Absent. The script does not use any delimiters or specific instructions to the model to ignore potential instructions embedded in the image content or metadata.
  • Capability inventory: The skill possesses file-read capabilities (PILImage.open) and file-write capabilities (image.save) within scripts/generate_image.py.
  • Sanitization: No sanitization, filtering, or validation is performed on the input image before it is processed by the model.
  • [CREDENTIALS_UNSAFE] (MEDIUM): Insecure API key handling.
  • The SKILL.md documentation and the scripts/generate_image.py script support passing API keys as command-line arguments (--api-key). This practice is unsafe as it can expose secrets in shell history, process monitors, and system logs.
  • [EXTERNAL_DOWNLOADS] (LOW): Use of external library dependencies.
  • The script requires google-genai and pillow. While these are from trusted sources (google and the established Python Imaging Library community), which downgrades the severity per [TRUST-SCOPE-RULE], they still represent an external dependency surface.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 15, 2026, 08:23 PM