train-voice

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (medium risk: 0.30). The skill contains a clear, end-to-end voice-cloning workflow that targets real people (celebrities and public figures), including automatic YouTube discovery/download, transcription, manifest preparation, fine-tuning of a TTS model, and explicit speaker-embedding injection — enabling impersonation/deepfake generation (dual-use / suspicious steering).

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

• Third-party content exposure detected (medium risk: 0.30). In the required workflow’s “design” path, /interview runs with outsider-provided free-text answers from the interview wizard (or fallback CLI) via --file questions and then persists/uses the resulting /tmp/..._response.json when processing the interview responses.