<p>Conversational agents have emerged as promising tools for depression-diagnosis-oriented chats. However, the integration of task-oriented dialogues with empathetic chit-chat in diagnostic interactions presents significant challenges. Such unique requirements demand professional expertise and empathetic communication, pushing beyond the capabilities of traditional dialogue frameworks designed for single optimization objectives. To bridge this gap, we propose a novel ontology-driven framework specifically designed for depression diagnosis dialogues. Our approach unifies the precision of task-oriented interactions with the empathetic nuance of chit-chat, enabling a more balanced and effective diagnostic exchange. We apply this framework to D⁴, the only publicly available Chinese dialogue dataset focused on depression diagnosis. Extensive experimental results demonstrate notable improvements in all tasks, highlighting the potential of our approach to advance task-oriented dialogue systems in digital mental health applications.</p>