Half-cascade Gemini Live model instructions

Hi everyone, I have been testing “half-cascade” gemini live model and the documentation says it internally generates text as output and make a TTS step. Then, are there any recommendations for the system prompt in order the entire audio-audio pipeline works properly?