unity-image-generator

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

• Third-party content exposure detected (medium risk: 0.30). In scripts/critique_image.py, the runtime reads outsider-provided free text from --subject and --intent (and uses --reference image bytes as “CANON REFERENCE” inputs) and then sends the assembled instruction string to the Gemini vision model via client.models.generate_content, creating an indirect prompt-injection surface through those CLI fields.