vfx

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

• Third-party content exposure detected (medium risk: 0.65). Yes: the workflow ingests the OUTSIDER-provided video content and audio via runtime URLs (e.g., state.clip_url from a user-supplied public URL or uploaded file) into LLM-readable text through analyze_media and transcribe_audio, then incorporates that readout into the agent’s subsequent LLM prompt (Step 4 → Step 5).