video-research

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

• Third-party content exposure detected (high risk: 0.85). The required workflow explicitly includes runtime web/social content discovery and “Content Gathering” that extracts “Notable Quotes” and “Key Facts” from outsider-authored sources (e.g., web pages, Reddit/HN/X posts, YouTube demos), which would be ingested as readable text into the agent’s LLM context.