prior-art-search

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

• Third-party content exposure detected (high risk: 0.90). This skill's required workflow (Step 2 and Step 3) instructs the agent to WebSearch/WebFetch public patent and literature sites (e.g., patents.google.com, worldwide.espacenet.com, Google Scholar, arXiv, Semantic Scholar) and to extract and act on their text to assess relevance and freedom-to-operate, so untrusted third-party content is ingested and can materially influence decisions.