research

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

• Third-party content exposure detected (high risk: 0.85). The skill’s Phase 2 “Parallel Dispatch” explicitly uses “web search and document fetching” (public web content fetched at runtime) and then has agents synthesize that fetched free-form text into the final LLM context/document, creating an indirect prompt-injection path from outsider-authored web pages into the agent’s context.