exa-search

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

• Third-party content exposure detected (high risk: 0.85). The Exa Search skill calls POST https://api.exa.ai/search and, when contents.text/contents.highlights/contents.summary are requested, Exa returns extracted page text/highlights/summaries from arbitrary web pages at runtime; that readable content is outsider-authored (public web) and is fed into the agent’s LLM context via the skill’s response fields like results[].text/results[].highlights/results[].summary (and potentially output.content).