caption-polish

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

• Third-party content exposure detected (medium risk: 0.65). The skill ingests audio/captions.json and audio/voice.json (auto-transcription outputs) and uses them as LLM-readable text to rewrite/polish captions; since these caption/transcript contents originate from outside the operating user (the spoken audio/transcription source), this is an outsider-authored free-text injection surface via the runtime caption/transcript text.