research-refine

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

• Third-party content exposure detected (high risk: 1.00). The skill's Phase 1 Step 1.1 explicitly instructs the agent to "search recent top-venue/arXiv work online" when local material is insufficient, meaning it will fetch and read untrusted public web content (arXiv/top-venue papers) that can materially influence method choices and subsequent actions.