tutor-research

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

• Third-party content exposure detected (high risk: 0.75). WebSearch/WebFetch로 런타임에 공용 웹 페이지(외부 작성자) 본문을 읽어 LLM이 종합본에 인용·요약하는 경로가 포함되어 있어, outsider free text가 컨텍스트에 들어갈 가능성이 있습니다.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.80). This skill explicitly performs runtime WebFetch of WebSearch result URLs (i.e., external web URLs returned by WebSearch) and injects the fetched content into the agent's context to verify sources and build outputs, so those external URLs can directly control prompts at runtime.