Human-in-the-loop

Some actions shouldn't happen without a person in the loop. An agent might be about to read a user's mailbox, spend credits, or write to a system, or it might reach a point where only the user can decide how to proceed. Human-in-the-loop capabilities let an agent pause mid-run, surface what it wants to do (or ask), and only continue once the user responds.

Serenity* AI Hub pauses the run and returns a pending action that the client answers on the next turn:

Skill Approvals

Gate sensitive skills behind an approval prompt, so a run pauses for the user to approve or reject each call before it runs.