research-compendium

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

• Third-party content exposure detected (medium risk: 0.30). In the runtime workflow described for research-compendium (Phase 1/2), the agent performs gbrain query/search and then “archive every primary source” via web search/fetch tools, extracting and saving full text from outsider-authored web sources into research/<topic-slug>/sources/ (and later using archived corpus in gbrain lsd), creating an indirect prompt-injection surface from attacker-submitted content.