p-image

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

• Third-party content exposure detected (high risk: 0.70). The SKILL.md examples for Image Editing and Multi-Image Compositing require passing external image URLs (e.g., "images": ["https://your-image.jpg"], "https://img1.jpg") that the service will fetch and process, so the app ingests arbitrary public/user-provided content which could influence model outputs and thus enable indirect injection.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.80). The skill invokes the external inference.sh platform at runtime (via the infsh CLI command "infsh app run pruna/p-image"), meaning https://inference.sh is a required runtime dependency that executes remote model code on which the skill depends.