multimodal-llm

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

• Third-party content exposure detected (high risk: 0.80). The skill explicitly requires and shows using publicly accessible image URLs and other external resources (e.g., rules/video-multi-shot.md: "Reference images must be publicly accessible HTTPS URLs"; rules/vision-image-analysis.md and video-generation-patterns.md examples use image_url and external URLs), so the agent ingests untrusted, user/third-party-hosted content that it is expected to interpret and which can materially influence generation/workflow.