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’s workflow, the agent uses firecrawl_search/firecrawl_scrape and exa web_search_exa/crawling_exa to fetch and deep-read full content from arbitrary web URLs discovered via web search based on user-provided sub-question keywords, exposing it to outsider-authored text.