<p>Continual graph learning addresses the challenges of temporal evolution in dynamic graph structures, enhancing a model’s ability to adapt to new information while retaining long-term knowledge. Although existing methods mitigate catastrophic forgetting, they often neglect latent semantic correlations and knowledge transfer across time, limiting forward transfer and long-term learning. To address these issues, we propose a prompt learning-based continual graph learning framework, PromptGNN, built upon Graph Neural Network architectures. It introduces learnable prompts to explicitly model global temporal context, guiding the model to better capture semantic relationships over time. We design a temporal-adaptive prompt mechanism to dynamically extract structural patterns at each time point, improving node representation expressiveness and generalization. Additionally, a cross-temporal shared prompt mechanism enables efficient reuse of critical knowledge across time, further reducing forgetting and enhancing adaptation to new data. Extensive experiments demonstrate the effectiveness of PromptGNN. It achieves leading classification accuracy on five benchmark datasets and significantly lower forgetting rates compared to prior methods, highlighting its advantages in knowledge retention and transfer.</p>