voice-mode

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

• Third-party content exposure detected (high risk: 0.90). Outsider-authored free text can enter the LLM via the runtime STT pipeline: microphone audio is transcribed by ElevenLabs (transcribe() uses /speech-to-text and returns resp.json().get("text")), then that user_text is embedded into the Claude prompt (ask_brain() builds prompt from the running transcript and passes it to claude -p ... --append-system-prompt).