news-tracker

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

• Third-party content exposure detected (high risk: 1.00). The skill's required workflow (Step 3: "执行 2-3 次 search_web") explicitly fetches and ingests public web news results (and uses the "search_web returned full redirect URL") from open third‑party sites like Reuters/BBC/TechCrunch, which the agent reads, summarizes, and uses to answer questions—exposing it to untrusted third‑party content that could carry indirect prompt injections.