reply-in-your-voice

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

• Third-party content exposure detected (medium risk: 0.30). The skill “Reply in Your Voice” takes two types of outsider-supplied free text at runtime—the user pastes “2-3 emails you’ve actually written before” and the message “you’re replying to”—and the agent then studies those samples and drafts a reply based on the pasted message.