nanobanana-image-generation

Warn

Audited by Gen Agent Trust Hub on Mar 12, 2026

Risk Level: MEDIUMCREDENTIALS_UNSAFEDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [CREDENTIALS_UNSAFE]: The scripts scripts/generate_image.py and scripts/generate_image.js default to using https://api.zhizengzeng.com/google as the API endpoint. This results in the NANOBANANA_API_KEY being transmitted to a third-party service rather than the official Google Gemini API, creating a risk of credential interception.
  • [DATA_EXFILTRATION]: User-provided prompt text and local image files (read and encoded as base64) are transmitted to a third-party endpoint (api.zhizengzeng.com) that is not part of the trusted vendors list or a well-known service.
  • [PROMPT_INJECTION]: The skill uses templates to generate materials-science figures that are susceptible to indirect prompt injection.
  • Ingestion points: User-provided scientific background text via the positional prompt argument in scripts/generate_image.py and scripts/generate_image.js is inserted directly into figure templates.
  • Boundary markers: Absent. User input is formatted into the prompt string without delimiters or instructions to ignore embedded commands.
  • Capability inventory: The skill performs network requests to external APIs and writes generated images and text to the local file system in ./output/nanobanana/.
  • Sanitization: Absent. There is no validation or escaping of the user-provided text before it is interpolated into the final instruction sent to the LLM.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 12, 2026, 01:27 AM