recommender-systems

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

• Third-party content exposure detected (high risk: 0.85). The skill’s runtime LLM context is populated by the marketer insights layer, which explicitly performs web_search for “recommendation engine CTR uplift … benchmark …” and then uses the retrieved web text to generate the insights; this is public web content fetched at runtime (outsider-authored) that can be ingested into the agent’s LLM context.