Continual learning has garnered significant interest due to its practicality in enabling deep models to incrementally incorporate new tasks of different classes without forgetting in a rapidly evolving world. The prompt-based methods, due to their ability to effective instruct pre-trained model to different tasks with few learnable prompt pool, have been the prevailing approaches on this line. However, prompt pool-based methods constrain the coarse information within group-level prompts, thereby not fully leveraging the more detailed information present in individual samples themselves. To address this, we propose an adaptive decoupled prompting method for class incremental learning. Specifically, we design an adaptive prompt generator to generate the specific prompt for each image of each task, so as to obtain the knowledge at the instance level. Moreover, we claim that there exists relevant information among different tasks, thus we further decompose the prompt to capture the knowledge shared across multiple tasks. Experimental evaluations on four datasets demonstrate the effectiveness of the proposed Dual-AP(Adaptive Decoupled Prompting for Class Incremental Learning) in comparison to the related class-incremental learning methods.