个性化 agent 面临记忆承诺决策：何时将交互信息写入持久记忆 vs. 仅当前上下文使用 vs. 要求澄清
方法
MCB 基准：140 场景（70 开发 + 70 测试），跨多个 LLM 家族评测记忆承诺策略
结果
不同家族模型在记忆承诺策略上差异显著；过度持久化和不足持久化都有具体的下游代价
与研究方向的关联
个性化对话 agent 的核心问题，直接影响长期交互的可靠性和用户体验
原文摘要
Persistent memory can personalize an LLM agent, but an incorrect durable update can silently distort future behavior. We study the memory-clarification boundary: whether interaction-derived information should be persisted, used only in the current context, re-verified, or clarified with the user. MCB contains 140 primary scenarios, split into 70 development and 70 held-out items, plus a separate 70-item contrast set. It evaluates both action labels and structured tool-call selection. Two non-aut