elevenlabs-voice-changer

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

• Third-party content exposure detected (medium risk: 0.30). The runtime workflow described for belt app run elevenlabs/voice-changer ingests the user-supplied --input JSON (including arbitrary audio URLs) and passes that content to the voice transformation CLI, so outsider-authored free text can reach the model via attacker-controlled inputs.

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

• Potentially malicious external URL detected (high risk: 0.90). This skill invokes the inference.sh "belt" CLI to run remote apps (e.g., belt app run elevenlabs/voice-changer) and points to the inference.sh service and an install script that fetches/executes remote code at https://inference.sh and https://raw.githubusercontent.com/inference-sh/skills/refs/heads/main/cli-install.md, so external content is fetched and executed at runtime.