Use of interrupt for human input

I am thinking of when to use interrupt vs restart from ‘supervisor’ agent in a hierarchical multi-agent design. so consider a supervisor agent along with a flight booking agent and hotel agent (each with its own set of tools). The user will ask to find flight, which the flight booking agent will use to find flights. Of course the user has to select one of them. I think interrupt might fit here. But then the user might ask no please find flights on another date etc My hunch is that in such a case the interrupt mechanism will fail. Am I right? Should I avoid using interrupt in this case where the tools might be needed to be invoked again? And the same case with the hotel agent (where user might ask please give me other options for X chain hotel etc.)

Thanks.

Use an interrupt for selection/clarification/slot-filling inside the ongoing plan; it will not fail if the user changes their mind. You can resume, update state, and invoke tools again as many times as needed. Use a restart (new run or replan from the supervisor) when the user fundamentally changes the goal or when you want to deliberately discard/refresh the plan and transient state.