When the network fluctuates, the model often returns a 400 error code, which currently leads to an error "ACP error (ACP_TURN_FAILED): acpx exited with code 1 next: Retry, or use /acp cancel and send the message again.” This error caused the acp connection to be interrupted. Therefore, a retry mechanism needs to be added, similar to what is done in openclaw.
If anyone can achieve it, we will be very grateful for his contribution!