google-patents-search

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

• Third-party content exposure detected (high risk: 0.85). The required workflow calls curl to POST to a Frevana endpoint and then reads the returned response body (raw = response_path.read_text(...)) and prints it to stdout (cat "$RESULT_FILE"), so outsider-authored free text/JSON from the external API is ingested into the agent/LLM context.