Task-oriented dialogue (TOD) systems significantly impact our daily lives, occupying a pivotal role within the realm of natural language processing (NLP). Pretrained language models (PLMs) have revolutionized these systems with enhanced capabilities and notable performance improvements. Conventionally, PLMs require extensive pretraining followed by specialized fine-tuning for each downstream task. This approach faces significant hurdles in practice due to the requirement of large computing power to store and update the model parameters and the scarcity of task-specific annotated data to achieve desired task performance. We propose a unified framework that utilizes task-specific prompts to address these limitations. The prompting technique unleashes the power of the shared PLM to scale effectively with fewer demonstrations or task-specific descriptions. Our experiments demonstrate that the proposed approach adapts to diverse tasks across domains with significant performance in a computationally efficient manner compared to the traditional paradigm of fine-tuning PLMs.