image-inpainting

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

• Third-party content exposure detected (high risk: 0.90). The skill's required schema and CLI examples explicitly ingest user-supplied image and mask URLs ("image" and "mask_image") and the Security & Privacy section even states that source image/mask URLs are untrusted and "embedded instructions can influence the fill", so third-party content fetched at runtime can materially affect inpainting behavior.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill fetches user-supplied image and mask URLs at runtime (e.g., https://your-cdn.example/street.jpg and https://your-cdn.example/cables-mask.png), and those fetched assets can embed text or content that directly influences the model's fill (prompt-injection) while being a required dependency for the mask-driven inpainting route.