backlinks-anchors

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

• Third-party content exposure detected (high risk: 0.85). The required workflow calls a public/third-party Frevana endpoint at runtime and then reads the returned HTTP response body as text (curl -o "$RESPONSE_FILE" → python3 ... read_text(...) → cat "$RESULT_FILE"), so outsider-authored free text (the provider’s JSON) is ingested into the agent’s LLM context via stdout.