Compress What You See, Not What You Say: Anchored Context Distillation for Latent-Observation Software Engineering Agents· 潜观测软件工程代理的锚定上下文蒸馏
Tool observations dominate the context of software-engineering agents, making long interaction histories costly to maintain. Existing context compression methods can discard information needed by later actions, while adapting agents to soft-token representations can compromise their original behavior. To reduce context while preserving action-critical information and agent behavior, we combine Latent Observations, Hard Actions (LOHA), a context layout that separates compressed history from text needed for exact reference, with Anchored Context Distillation (ACD), a training method that enables latent reading while constraining behavioral drift. LOHA compresses older tool observations into soft tokens while retaining the agent's own turns and the last K observations in text, providing compa
通过压缩工具观察而非文本，减少软件工程代理交互历史的维护成本。
- 核心方法
- 提出LOHA和ACD方法，LOHA分离压缩历史和需要精确引用的文本，ACD训练代理在压缩上下文中保持行为一致性。
- 适合谁读
- 软件工程研究者和工程师
- 要解决的问题
- 软件工程代理在长期交互中产生大量工具观察记录，现有压缩方法可能导致重要信息丢失或行为漂移。
- 关键实验
- 未提供
- 主要贡献
- LOHA和ACD有效减少交互历史存储成本，同时保持了代理执行关键动作所需的信息和原始行为。
- 意义与局限
- 该方法有助于提高软件工程代理的效率和性能，但具体应用场景和效果还需进一步实验验证。