character-design-sheet

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

• Third-party content exposure detected (medium risk: 0.30). The required runtime workflow invokes the belt CLI to run remote inference (e.g., belt login and belt app run ... --input '{ "prompt": ... }'), so attacker-controlled free text can be provided by the user as the prompt and then sent to third-party inference as readable text.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill invokes the inference.sh service at runtime (https://inference.sh) via the belt CLI to send prompts and perform remote model inference, so this external URL is a required runtime dependency that controls prompt execution.