---
title: "养虾日记2 SelfImproving"
type: source
tags: []
date: 2026-04-02
---

## Source File
- [[raw/微信公众号/养虾日记2：让Agent更懂你：OpenClaw + Self-Improving 复盘实战案例分享.md]]

## Summary
- 核心主题：OpenClaw Agent自改进机制和双层记忆架构
- 问题域：AI Agent记忆缺失、重复犯错、流程漏洞
- 方法/机制：短期记忆(memory文件)+长期记忆(LanceDB)+self-improving复盘机制
- 结论/价值：错误只犯一次，第二次就知道怎么做对——self-improving让agent在错误中持续进化

## Key Claims
- 双层记忆架构：每日文件管上下文，向量数据库管知识，self-improving管成长
- self-improving固定格式记录：Summary/Details/Suggested Action/MetaData
- Pattern-Key重复是系统性问题的信号，Recurrence-Count是重要指标
- 23:00定时复盘流程：读取memory→记录learnings→检查重复→同步到向量库→Telegram摘要

## Key Concepts
- [[SelfImproving]]：结构化经验记录与复盘机制
- [[双层记忆架构]]：短期+长期+自改进三层
- [[PatternKey]]：问题模式识别与追踪
- [[每日复盘]]：定时任务驱动持续改进

## Key Entities
- [[memory-lancedb-pro]]：基于LanceDB的向量记忆系统

## Connections
- [[OpenClaw]] ← enhanced_by ← [[SelfImproving]]
- [[双层记忆架构]] ← composes ← [[SelfImproving]]

## Contradictions
- 无冲突