How-to

智能体的删除策略:会话之间该保留什么、丢弃什么

Why we picked it

做智能体工作流的人多半给记忆做过加法,很少做过减法。这条推把删除当策略来讨论:旧计划、失败方案哪些留、哪些删、谁拍板。自检那个问题挺扎心,适合用来审一遍自己的 prompt 和 memory 设计。

aiDotEngineer 提出智能体上下文管理中被忽视的一环:删除策略(delete policy)。问题针对三类常见残留——旧计划、失败方案、临时 workaround——追问它们哪些该在会话之间存活、哪些该过期、由谁决定。

Body

这条推文的实质是把智能体记忆管理当作显式设计问题:上下文里塞满历史失败尝试会拖慢后续会话,而全部清空又会丢失有用经验。作者给出的自检问题是"你刻意不再带入下一会话的一件事是什么",指向按条目维护可执行清单,而非笼统的"清理上下文"。

对写智能体工作流的开发者,这可以和 compaction、外部记忆文件等做法配合:把已验证结论写入持久笔记,把失败路径标记为过期条目。

Original · AI Engineer

What's your agent's delete policy? Old plans. Failed approaches. Temporary workarounds. What survives between sessions, what expires, and who decides? What's one thing you deliberately stopped carrying into the next session?

Industry Coordinates & Source Verification

This report is categorized under AITOP's "How-to" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include agents, context management, memory, prompt engineering.

Source attribution: curated from AI Engineer (@aiDotEngineer).