In AI Agent scenarios, session management involves the storage, retrieval, and injection of conversation history; if each interaction requires loading large amounts of historical records and redundant skill modules, it significantly increases inference latency and memory usage.