Hi, @krunal Make handoffs system-driven, not prompt-driven. Let the LLM gather intent and signals, but use deterministic rules (intent, confidence, slots) to force transfers. Trigger the switch silently and lock the primary agent once transfer starts. Yes, mention handoffs — but lightly. Don’t tell the primary agent when to transfer. Just make it transfer-aware (collect info, call tools when it’s not the right agent). The system should decide and execute. let's talk more in my DM