openai-vision

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

• Third-party content exposure detected (medium risk: 0.65). The skill’s runtime path ingests outsider-authored free text when the operating user supplies an HTTP image_url (or a downloaded/forwarded image converted to base64) whose image contains text (e.g., screenshots, documents, web pages) that the vision model performs OCR on and includes in the LLM context as extracted/recognized text; this is not limited to user-authored content.