在长期任务中，, 决策相关状态通常分散在不断扩展的轨迹, 上，而行动代理必须将其浮现出来并采取行动。随着轨迹的增长, 任务要求, 环境事实, 先前的尝试, 诊断, 和开放的子目标可以被埋在上下文窗口中或超出它, 在需要时无法影响决策。我们将此故障模式称为"行为状态衰减"。我们将记忆作为一种主动干预机制而不是被动检索来研究。一个单独的记忆代理与一个未修改的动作代理,一起运行，根据最近的轨迹更新结构化记忆库，并决定是注入基于记忆的提醒还是保持沉默。该模块可与前沿行动代理和现有代理线束即插即用。在 Terminal-Bench 2.0 和 $\tau^2$-Bench, 中，它提高了较弱和较强作用代理, 的通过率@1，在 Terminal-Bench 上获得了 +8.3 pp，在 $\tau^2$-Bench 上获得了 +6.8 pp。消融表明，选择性干预优于被动银行敞口, 始终在线注入, 仅顾问指导, 和一般检索。作为实现开放权重内存策略, 的早期步骤，我们使用 SFT 和 GRPO, 在 SETA 上训练 Qwen3.5-27B，提高验证奖励并实现部分转移到 Terminal-Bench。
In long-horizon tasks, decision-relevant state is often scattered across an expanding trajectory, while the action agent must surface it and act. As trajectories grow, task requirements, environment facts, prior attempts, diagnoses, and open subgoals can be buried in the context window or pushed beyond it, failing to influence decisions when needed. We call this failure mode "behavioral state decay". We study memory as an active intervention mechanism rather than passive retrieval. A separate memory agent runs alongside an unmodified action agent, updating a structured memory bank from the recent trajectory and deciding whether to inject a memory-grounded reminder or remain silent. The module is plug-and-play with frontier action agents and existing agent harnesses. Across Terminal-Bench 2.0 and $\tau^2$-Bench, it improves pass@1 for both weaker and stronger action agents, with gains of +8.3 pp on Terminal-Bench and +6.8 pp on $\tau^2$-Bench. Ablations show that selective intervention outperforms passive bank exposure, always-on injection, advisor-only guidance, and general retrieval. As an early step toward open-weight memory policies, we train Qwen3.5-27B on SETA using SFT and GRPO, improving validation reward and achieving partial transfer to Terminal-Bench.
科目: 人工智能 (cs.AI); 计算和语言 (cs.CL)
Subjects: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)