Multi-Agent Memory Optimizer

多Agent记忆管理系统 - 开放协作的知识库解决方案 支持私有+公共双层记忆空间，自动生成每日总结，跨Agent知识检索

Security Scan

The skill is broadly consistent with a local multi‑agent memory manager, but documentation and shipped defaults conflict in ways that could lead to unintended automatic publishing of private memories — review configs before use.

Purpose & Capability

The name/description (multi-agent memory manager with private+public tiers) matches the included code: all modules operate on local filesystem paths (workspace/memory, private/public) and implement summarization, searching, uploading (to a local public directory), tiering, etc. No cloud credentials, remote endpoints, or unrelated binaries are requested. Minor discrepancy: README/SKILL.md mention optional environment variables (e.g., CHROME_PATH, MX_APIKEY) that are not required or referenced by the code — this is an unneeded note but not evidence of exfiltration.

Instruction Scope

Runtime instructions and the code read local OpenClaw memory files (MEMORY.md, memory/YYYY-MM-DD.md) and then generate and can publish summaries to memory/public. That behavior is expected for this purpose. However the documentation repeatedly claims 'upload requires manual confirmation' and 'privacy protection', while the actual included config/default.json sets upload.require_upload_confirm (and related upload defaults) in a way that can enable automatic publishing (the file shipped in the package sets auto_publish/require_upload_confirm values that contradict the safety claims). If deployed with the shipped defaults (or if config is changed), private content could be auto-published without the user noticing. The summarizer also supports automatic operation (crontab guidance) which increases the risk if confirmation is disabled.

Install Mechanism

There is no external install script or remote download — the skill is delivered as local Python scripts and JSON config. No brew/npm/remote archive fetches are present. This lowers supply-chain risk; nothing in the package retrieves or executes external code at runtime.

Credentials

The skill declares no required environment variables and the code does not read cloud credentials. That's proportional to its stated local filesystem purpose. The README/SKILL.md mention optional variables (CHROME_PATH, MX_APIKEY) as examples for other integrations, but those are not used in the shipped code. The main concern is configuration (config/default.json) controlling publish behavior — these are local config flags, not secrets, but they materially affect privacy.

Persistence & Privilege

The skill does write files: it creates agent config files under config/agents, writes summaries into the workspace memory directories, and cmd_config persists changes to config/default.json. It does not set always: true, does not modify other skills, and does not require elevated system privileges. Writing global default.json is expected for a CLI tool but is something to be aware of (it changes global behavior for all agents using that workspace).

What to consider before installing

This package is mostly coherent with a local memory manager, but pay close attention to configuration before running. The shipped config can enable automatic summarize/publish behavior contrary to the README's safety claims — which could cause private memory files (MEMORY.md or memory/YYYY-MM-DD.md) to be uploaded to the 'public' folder without an interactive confirmation. Before installing or running: 1) Inspect config/default.json in the package and set upload.require_upload_confirm = true and upload.auto_publish = false (and keep upload.backup_private = true). 2) Run summarize and upload manually at first to verify output, do not enable crontab auto jobs until you are confident. 3) Run the tool in a safe test workspace (copy ~/.openclaw/workspace to a sandbox) to observe behavior. 4) If you rely on strict privacy, audit any generated public files before exposing them and consider disabling auto_summarize and any automatic cron tasks. If you want higher assurance, review the full source locally (it is included) and search for any network calls — none were found in the provided code, but changing config or extensions could add them later.

让每个 AI Agent 拥有独立记忆，同时共享公共知识库

🌟 核心价值

• 隐私保护：每个 Agent 独立记忆空间，互不干扰
• 知识传承：重要经验发布到公共空间，其他 Agent 可检索使用
• 自动化：每日自动生成总结（通过 crontab），减少人工负担
• 开箱即用：完整工具链，5 分钟快速部署

---

📦 安装

# 1. 进入 skills 目录
cd ~/.npm-global/lib/node_modules/@qingchencloud/openclaw-zh/skills/

# 2. 确保此技能目录存在
#    memory-optimizer-base/

# 3. 初始化你的 Agent
./memory-optimizer-base/memory_optimizer.py init --agent <your_agent_id>

---

🎯 核心工作流

1. OpenClaw 日常会话 → 记录到 memory/YYYY-MM-DD.md
2. 每日 23:00 自动 summarize → 生成 medium-term/YYYY-MM-DD.md
3. 人工确认内容 → 执行 upload → 发布到 public/<agent>/<date>/
4. 任何 Agent 可 search-public → 获取其他 Agent 的经验

---

🔨 命令速查

# 初始化
./memory_optimizer.py init --agent xiaotian

# 手动生成总结（测试用）
./memory_optimizer.py summarize --agent xiaotian --date 2026-04-06

# 上传到公共空间
./memory_optimizer.py upload --agent xiaotian --date 2026-04-06 --title "事件标题"

# 搜索公共知识
./memory_optimizer.py search-public "关键词"

# 查看私有记忆
./memory_optimizer.py private list --agent xiaotian

# 分析系统状态
./memory_optimizer.py analyze --agent xiaotian

# 配置管理
./memory_optimizer.py config --set memory.sync_enabled=true

---

⚙️ 配置

编辑 config/default.json：

{
  "memory": {
    "base_path": "~/.openclaw/workspace-xiaotian",
    "private_root": "memory/private",
    "public_root": "memory/public",
    "medium_term_retention_days": 1
  },
  "summarizer": {
    "template": "...",  // 自定义输出格式
  },
  "upload": {
    "require_upload_confirm": true  // 发布前是否需要人工确认
  }
}

---

📂 文件结构

memory-optimizer-base/
├── memory_optimizer.py   # 主程序
├── .gitignore            # 隐私保护
├── README.md             # 详细文档
├── SKILL.md              # 本文档
├── config/
│   ├── default.json      # 默认配置
│   └── agents/           # 各 Agent 配置（自动生成）
└── lib/                  # 核心模块
    ├── analyzer.py
    ├── summarizer.py
    ├── uploader.py
    ├── retriever.py
    ├── optimizer.py
    ├── tierer.py
    └── sync.py

---

🔐 安全与隐私

• ✅ 所有真实记忆文件（memory/、.openclaw/）已在 .gitignore 中排除
• ✅ 默认上传需人工确认，避免误发
• ✅ 私有空间严格隔离，公共空间仅包含主动分享的内容
• ⚠️ 公共空间内容永久可见，发布前请检查

---

📖 完整文档

详见 README.md

---

🤝 贡献

欢迎 Issue 和 PR！

---

版本: 0.2.0 | 许可: MIT-0