58 lines
2.2 KiB
Markdown
58 lines
2.2 KiB
Markdown
---
|
||
title: "Shared Memory Architecture"
|
||
type: concept
|
||
tags: [OpenClaw, Agent, Architecture, Memory]
|
||
sources: [multi-agent-team]
|
||
last_updated: 2026-04-23
|
||
---
|
||
|
||
## Definition
|
||
|
||
**Shared Memory Architecture** 是多 Agent 系统中,所有 Agent 共用一套公共记忆文件(GOALS.md、DECISIONS.md、PROJECT_STATUS.md),确保团队目标、关键决策和项目状态对所有 Agent 可见的架构模式。
|
||
|
||
## 核心洞察
|
||
|
||
> "Shared memory + private context: The combination is critical — agents need common ground (goals, decisions) but also their own space to accumulate domain expertise"
|
||
|
||
共享记忆解决的问题:
|
||
- **知识孤岛**:营销洞察不会自动影响开发优先级
|
||
- **上下文丢失**:每个 Agent 独立会话无法继承团队决策
|
||
- **重复工作**:Agent A 的发现对 Agent B 不可见
|
||
|
||
## 共享文件结构
|
||
|
||
```
|
||
team/
|
||
├── GOALS.md # 当前 OKR 和优先级(所有 Agent 读取)
|
||
├── DECISIONS.md # 关键决策日志(追加写入)
|
||
├── PROJECT_STATUS.md # 当前项目状态(所有 Agent 更新)
|
||
├── agents/
|
||
│ ├── milo/ # Milo 的私有上下文和笔记
|
||
│ ├── josh/ # Josh 的私有上下文
|
||
│ ├── marketing/ # 营销 Agent 的研究笔记
|
||
│ └── dev/ # 开发 Agent 的技术笔记
|
||
```
|
||
|
||
## 共享记忆类型
|
||
|
||
| 文件 | 用途 | 写入规则 |
|
||
|------|------|----------|
|
||
| GOALS.md | 团队目标和 OKR | 定期更新(周/日) |
|
||
| DECISIONS.md | 关键决策记录 | 决策时追加 |
|
||
| PROJECT_STATUS.md | 项目当前状态 | 状态变更时更新 |
|
||
|
||
## 与 Private Context 的关系
|
||
|
||
**Shared Memory** + **Private Context** 组合是关键:
|
||
- **Shared Memory**:共同基础(目标、决策、项目状态)
|
||
- **Private Context**:独立空间积累领域专业知识
|
||
- 两者缺一不可:只有共享记忆导致缺乏深度,只有私有上下文导致知识孤岛
|
||
|
||
## Related Concepts
|
||
|
||
- [[Agent-Memory]] — OpenClaw 的长期记忆机制
|
||
- [[Private Context]] — Agent 私有上下文
|
||
- [[OpenClaw]] — 提供 workspace 目录体系支持
|
||
- [[Agent Specialization]] — 专业化 Agent 依赖共享记忆协调
|
||
|