audio-transcriber

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

• Third-party content exposure detected (medium risk: 0.30). In scripts/transcribe.py, runtime LLM prompt construction in handle_prompt_workflow(...)/process_with_llm(...) injects the entire transcript_text (derived from user-provided audio content) and then feeds the resulting free text plus user-provided --prompt into the external claude / gh copilot suggest CLIs, meaning outsider-authored text from the audio is ingested by the LLM.