deep-research

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

• Third-party content exposure detected (medium risk: 0.30). In SKILL.md Step 3b–3c the agent uses web_search to obtain URLs and then web_extract to ingest full outsider-authored page text from selected search results (based on the user’s research question), making the runtime workflow exposed to indirect prompt-injection content present in those pages.