research-expert

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

• Third-party content exposure detected (high risk: 0.85). The skill is a research agent that performs runtime web searches and fetches external pages (“Initial Broad Search”, “Targeted Deep Dives”, and “Fetch multiple pages simultaneously”), so outsider-authored free text from public web content can be ingested into the LLM context via the web browsing/search tool outputs.