experiment-design

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

• Third-party content exposure detected (high risk: 0.80). The skill instructs pulling and classifying recipients' email replies via "/positive-reply-scoring" (Step 6: "Pull metrics via /positive-reply-scoring" and "Positive replies (per arm, classified by Claude)"), which clearly ingests untrusted, user-generated third-party content and uses those classifications to decide experiment winners and next actions.