research

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

• Third-party content exposure detected (high risk: 0.90). The skill’s required workflow includes “Search using web search across multiple angles” and “Extract key findings” from those runtime-fetched web pages, which are outsider-authored free text ingested into the agent’s LLM context via the browsing/scraping/extraction path.