image-gen-blockeden-gemini

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

• Third-party content exposure detected (high risk: 0.80). The skill's SKILL.md explicitly instructs downloading user-provided image URLs (e.g., "If the user gives a URL, download it first with curl -L -o /tmp/ref_image.jpg ''"), and the scripts (gemini_flash.py) then load and embed that arbitrary third‑party image as a reference that the model will interpret and use to drive generation, so untrusted external content is fetched and can materially influence behavior.