open-memory-system

三层记忆系统 — Working/Short-Term/Long-Term Memory 管理。支持自动偏好记忆、实体记忆、事件记录、L2自动提炼、Hook自动触发、Cron定期整理。用于记忆用户偏好、跨会话积累知识、自动化记忆管理。

Security Scan

The skill is internally consistent with a local, file-based three-tier memory manager: it reads/writes files under the agent workspace and deploys hooks/crons to manage memory; nothing in the bundle requests remote credentials or external network endpoints.

Purpose & Capability

Name/description promise a local memory system and the included scripts and hooks exclusively implement file-based memory operations (working/short-term/long-term), L2 distillation, and cron/hook wiring — these requirements align with the stated purpose.

Instruction Scope

Runtime instructions and hook code read and write files in the agent workspace (MEMORY.md, .learnings, memory/*). This is expected for a memory manager, but hooks will automatically read .learnings and MEMORY.md from the workspace when deployed, so you should ensure those directories/files do not contain secrets you don't want the skill to ingest.

Install Mechanism

No remote installers or downloads — the skill is instruction-plus-local-scripts only. The install steps are copy/unzip and creating directories, which is proportionate to the task.

Credentials

The skill declares no required credentials or env vars. Code uses optional environment variables (MEMORY_DIR, OPENCLAW_WORKSPACE) and defaults to a path under /root/.openclaw/workspace/memory; this is reasonable but the default absolute root path and reliance on workspace env vars should be reviewed in your environment. No network credentials are requested.

Persistence & Privilege

always:false (no forced always-on). Hooks are intended to be copied to ~/.openclaw/hooks/ and will run on agent events (session:end, agent:bootstrap) if deployed — autonomous invocation of hooks is expected behavior for this skill but you should be aware hooks will execute automatically once installed.

Assessment

This package appears to do what it says: manage local memory files and provide hooks/crons to auto-save/load and distill short-term items into long-term events. Before installing: (1) inspect the files you’ll allow the hooks to read (MEMORY.md, ~/.openclaw/workspace/.learnings) to ensure they contain no secrets; (2) consider changing the default MEMORY_DIR to a non-root path appropriate for your deployment; (3) avoid passing untrusted input into preference/entity keys (filenames are created directly from keys — this can allow path traversal or odd filenames); (4) review and control deployment of the hook directory (~/.openclaw/hooks/) since hooks will run automatically on agent events; and (5) verify whether any referenced helper (index.py) exists in your environment (cleanup tries to run it if present). Overall the skill is coherent and local-only, but follow these precautions before enabling hooks or crons.

基于 OpenViking + Microsoft Agent Memory 的三层记忆框架，适用于 AI Agent 的持久化记忆系统。

核心概念

层级	用途	存储
Working Memory	当前会话临时信息	working.json
Short-Term Memory	会话级别短期记忆	short-term/*.md
Long-Term Memory	偏好、实体、事件持久化	user/preferences/, user/entities/, user/events/

目录结构

memory/                          # 记忆主目录
├── working.json                # Working Memory
├── short-term/                 # Short-Term Memory (会话级别)
│   └── YYYY-MM-DD-HHMM.md
├── user/                       # Long-Term Memory (用户视角)
│   ├── preferences/           # 偏好记录
│   ├── entities/              # 实体记忆
│   └── events/                # 重要事件
└── agent/                     # Agent 自身记忆
    ├── persona/                # 人设/角色设定
    └── episodic/               # 经验教训

快速安装

Step 1: 解压 skill 到 workspace

cd ~/.openclaw/workspace/skills
unzip open-memory-system.zip -d open-memory-system

Step 2: 初始化记忆目录

export MEMORY_DIR=~/.openclaw/workspace/memory
mkdir -p $MEMORY_DIR/{user/{preferences,entities,events},agent/{persona,episodic},short-term}

Step 3: 部署 Hook

# auto-save-memory: 部署到 ~/.openclaw/hooks/
cp -r open-memory-system/scripts/auto-save-memory ~/.openclaw/hooks/

# load-memory-on-start 已在 ~/.openclaw/hooks/ 中预装，无需重复部署

Step 4: 创建 Cron（定时任务）

参考 crons/memory-crons.txt 创建每日定时任务。

CLI 命令

# 读取核心记忆
python3 scripts/memory.py

# 每日统计
python3 scripts/memory.py summary

# 清理过期
python3 scripts/memory.py cleanup

# L2 提炼（从 short-term → long-term events）
python3 scripts/distill_l2.py

# 记录偏好
python3 scripts/memory.py pref "沟通方式" "直接高效" "用户偏好"

# 记录事件
python3 scripts/memory.py event "项目启动" "Miloya 正式成立"

# 记录经验
python3 scripts/memory.py episode "第一次部署" "negative" "需要检查依赖版本"

自动化流程

用户对话
    ↓
[session:end Hook]  → auto-save-memory → 保存 .learnings → memory
    ↓
[memory-check cron] → 清理过期 + 会话摘要 → short-term
    ↓
[distill_l2 cron @20:00] → short-term 提炼 → user/events/ (L2)
    ↓
[session:start Hook] → load-memory-on-start → 加载 MEMORY.md + short-term

内置 Hook 说明

Hook	位置	触发	功能
load-memory-on-start	~/.openclaw/hooks/（预装）	agent:bootstrap	运行 memory.py read 加载记忆
auto-save-memory	~/.openclaw/hooks/（需部署）	session:end	保存 .learnings/ 到 events

自定义配置

修改 scripts/memory.py 开头的配置区：

MEMORY_DIR = Path(os.environ.get("MEMORY_DIR", "/root/.openclaw/workspace/memory"))
DEFAULT_EXPIRE_DAYS = 90