conversationsevaluationsagents
Diagnose a conversation with the assistant, and a dashboard-chat privacy fix
Ask the assistant why an agent answered the way it did in a real conversation, and get a change set and a regression case out of the answer.
New
- Diagnose a conversation with the assistant — point it at a recent conversation, or let it find one, and it reads the full trace, tells you why the agent answered the way it did, proposes a change set to fix it, and can save a regression case so it never happens again. See ask the assistant to diagnose it.
Fixed
- The assistant now knows when you approve or discard a change set it proposed. Previously it could keep describing an applied change as still pending, or propose a discarded one again on your next message.
- A dashboard user's private chat with the assistant could be read or acted on by another workspace member who had its conversation ID. This is now limited to the person it belongs to.