nano-banana-2-product-detail-page

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). The runtime workflow reads outsider-authored free text when a user passes --prompt (and optional --system) into cmd_image() → client.generate_image() and sends it as the prompt field to POST .../openapi/v1/generation/image, so the LLM consumes arbitrary user text.