gpt-image-edit

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

• Potentially malicious external URL detected (high risk: 0.80). The skill sends user-provided HTTPS image URLs to the remote RunComfy model at runtime (e.g., "https://.../portrait.jpg"), and those fetched images can contain embedded/textual prompt-instructions that directly influence the model's outputs (image-based prompt injection) as noted in the Security & Privacy section.