We use Vapi Squad to build an AI receptionist that handles a complex workflow of incoming calls. Typically the issue we face is during hand offs: 1. the primary agent doesn’t hand off to the sub agent. 2. Even after explicit instructions it does sometimes mention that it is going to transfer to a specialist. In our primary agents prompt today, we have not mentioned the instructions on when to transfer - we let the LLM figure this out basis the description of the tools attached. We have done this because it significantly reduces the #2 issue I mentioned above. Question: 1. What are the best ways to solve this for a foolproof experience? 2. Should I mention hand offs instructions in the primary agent too?