← 返回 Skills 市场
huamu668

OpenClaw Multi-Agent System

作者 huamu668 · GitHub ↗ · v1.0.0
cross-platform ✓ 安全检测通过
297
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install huamu668-openclaw-multiagent
功能描述
Configure multi-agent TG group system with shared Workspace + MemOS memory. Use when user wants to set up multiple AI agents in a Telegram group, with specia...
使用说明 (SKILL.md)

OpenClaw 多 Agent TG 群组系统配置指南

架构说明

Workspace:共享模式

所有 Agent 共享同一个 workspace(.openclaw/workspace)。

  • 主 Agent 的文件在 workspace 根目录(SOUL.md, AGENTS.md 等)
  • 每个子 Agent 的专属文件在 workspace/agents/{agent_id}/ 子目录
  • 共享上下文在 workspace/shared-context/ — 所有 Agent 都可读取
  • 协作通过文件完成:一个 Agent 写文件,另一个 Agent 读文件

记忆:MemOS Cloud

  • MemOS Cloud 插件已安装并启用,挂载在 OpenClaw 实例级别
  • 所有 Agent 自动共享同一个记忆池
  • 不需要创建 memory/ 目录或 YYYY-MM-DD.md 日志文件

目录结构

workspace/
├── SOUL.md                    # 主 Agent 的灵魂
├── IDENTITY.md                # 主 Agent 身份卡
├── AGENTS.md                  # 主 Agent 行为规则
├── USER.md                    # 用户信息(所有 Agent 共享读取)
├── HEARTBEAT.md               # 主 Agent 心跳任务
├── shared-context/            # 跨 Agent 共享层
│   ├── FEEDBACK-LOG.md        # 通用反馈/修正记录
│   └── SIGNALS.md             # 当前关注的趋势/信号
└── agents/
    ├── {agent_id}/            # 子 Agent 专属目录
    │   ├── SOUL.md            # 子 Agent 灵魂
    │   ├── IDENTITY.md        # 子 Agent 身份卡
    │   └── AGENTS.md          # 子 Agent 行为规则
    └── {另一个agent_id}/
        └── ...

配置前准备

需要用户提供:

  1. TG 群组 ID(负数,如 -1002345678901)
  2. 用户的 TG 用户 ID(如 5701780765)
  3. 主 Bot Token(已配置或新提供)
  4. 子 Agent 列表(YAML 格式)

⚠️ 手动前置操作(必须)

1. BotFather 设置

每个子 Bot → /setprivacyDisable(否则 Bot 无法读取群消息)

2. 拉 Bot 进群

所有 Bot(主 + 子)必须先被添加到目标 TG 群组

3. 获取 Bot Token

每个子 Bot 需要从 @BotFather 获取独立的 Bot Token

4. 获取群组 ID

  • 转发群消息给 @raw_data_bot
  • 或查看群组消息的 chat.id 字段

执行步骤

Step 1:备份

cp ~/.openclaw/openclaw.json ~/.openclaw/openclaw.json.bak.$(date +%Y%m%d%H%M%S)

Step 2:创建目录结构

# 共享上下文目录
mkdir -p ~/.openclaw/workspace/shared-context

# 每个子 Agent 的专属目录
mkdir -p ~/.openclaw/workspace/agents/{agent_id}

# 每个子 Agent 的 OpenClaw 内部目录
mkdir -p ~/.openclaw/agents/{agent_id}/agent

Step 3:创建共享上下文文件

写入 shared-context/FEEDBACK-LOG.mdshared-context/SIGNALS.md

Step 4-6:为每个子 Agent 创建

  • SOUL.md
  • IDENTITY.md
  • AGENTS.md

Step 7:修改 openclaw.json

关键修改:

  1. channels.telegram → 改为 accounts 多账号模式
  2. agents → 添加子 Agent 配置
  3. bindings → 添加 agent 与 channel 的绑定
  4. tools → 确保 agentToAgent 和 sessions.visibility 配置

Step 8-11:验证、重启、验证上线、汇报

配置示例

Telegram Accounts 配置

"telegram": {
  "enabled": true,
  "dmPolicy": "pairing",
  "groupPolicy": "allowlist",
  "streaming": "partial",
  "accounts": {
    "default": {
      "botToken": "主BotToken",
      "dmPolicy": "pairing",
      "groupPolicy": "allowlist",
      "streaming": "partial",
      "groups": {
        "-100xxxx": { "requireMention": false }
      },
      "groupAllowFrom": ["用户ID"]
    },
    "agent_id": {
      "name": "Agent名称",
      "enabled": true,
      "botToken": "子BotToken",
      "dmPolicy": "allowlist",
      "allowFrom": ["用户ID"],
      "groupPolicy": "allowlist",
      "groupAllowFrom": ["用户ID"],
      "streaming": "off",
      "commands": {
        "native": false,
        "nativeSkills": false
      },
      "groups": {
        "-100xxxx": { "requireMention": true }
      }
    }
  }
}

Agents 配置

"agents": {
  "defaults": {
    "workspace": "~/.openclaw/workspace",
    "model": { "primary": "..." }
  },
  "list": [
    { "id": "main" },
    {
      "id": "agent_id",
      "name": "agent_id",
      "workspace": "~/.openclaw/workspace",
      "model": "..."
    }
  ]
}

Bindings 配置

"bindings": [
  {
    "agentId": "agent_id",
    "match": {
      "channel": "telegram",
      "accountId": "agent_id"
    }
  }
]

踩坑防护清单

  • 子 Bot 的 /setprivacy 必须设为 Disable
  • 子 Bot 的 commands.native 和 nativeSkills 必须为 false
  • 原顶层 botToken 必须删除,移入 accounts.default
  • JSON 修改后必须验证语法
  • 所有 Agent 的 workspace 指向同一个路径
  • 区分 workspace/agents/(工作文件)和 .openclaw/agents/(内部数据)
  • shared-context/ 文件遵循一写多读原则
  • MemOS 记忆已全局生效,不要创建 memory/ 目录

参考

安全使用建议
This guide is coherent and appears to do what it advertises, but take these precautions before proceeding: - Back up ~/.openclaw/openclaw.json (the guide suggests doing so) and validate JSON after edits. A bad config can break your instance. - Treat every Bot Token as a secret: store it securely (not pasted into public chat or a public repository). If a token is leaked, rotate it in BotFather immediately. - Disabling BotFather privacy (/setprivacy → Disable) allows bots to read all group messages — only do this if you understand and accept the privacy implications. - Forwarding a message to @raw_data_bot sends message content to that third party. If you do not trust that bot, obtain the chat ID via your own bot or other admin tooling instead. - Confirm where "MemOS Cloud" stores memory before enabling/shared use: if memory is cloud-hosted, sensitive conversation content may be uploaded off-host. - Test configuration in a private or test group first, and limit bot permissions to the minimum required. - The skill is only guidance (no code runs automatically). Follow the steps manually and verify each change before putting agents into a production group.
功能分析
Type: OpenClaw Skill Name: huamu668-openclaw-multiagent Version: 1.0.0 The skill bundle is a configuration utility designed to help users set up a multi-agent Telegram system within OpenClaw. It provides structured instructions in SKILL.md for the AI agent to back up the existing configuration, create necessary directory structures, and update the openclaw.json file with user-provided Telegram bot tokens and IDs. While the skill performs high-privilege actions like file system modifications and handling sensitive tokens, these behaviors are transparently documented and strictly aligned with the stated purpose of the skill, with no evidence of malicious intent or data exfiltration.
能力评估
Purpose & Capability
The name/description (multi-agent Telegram group with shared workspace and MemOS memory) matches the instructions: creating ~/.openclaw workspace, adding per-agent dirs, and editing openclaw.json to include multiple bot tokens and bindings. The only credentials the guide needs are Telegram bot tokens and user/group IDs, which are appropriate for this task. The guide assumes a MemOS Cloud plugin is present — that is an environmental assumption but not inconsistent.
Instruction Scope
Instructions are limited to local config changes (back up and edit ~/.openclaw/openclaw.json), creating workspace directories and text files, and operational steps for adding bots to the Telegram group. These actions are within scope. Two notes: (1) the guide suggests forwarding a message to @raw_data_bot to obtain chat.id — that sends group message data to a third-party bot and may leak content; (2) it requires disabling BotFather privacy so bots can read all group messages, which is a security/privacy tradeoff the user should consider.
Install Mechanism
No install spec or code files are present — this is instruction-only. Nothing is downloaded or written by the skill itself, so there is no installer risk from the skill package.
Credentials
The guide does not declare environment variables or request unrelated credentials. It instructs the user to place each Bot Token into openclaw.json (appropriate and proportional). No other secrets or unrelated service keys are requested.
Persistence & Privilege
Skill flags are default (always: false, agent invocation allowed). The guide instructs edits only to the user's ~/.openclaw configuration and workspace; this is expected for a local configuration guide and does not request elevated or system-wide privileges.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install huamu668-openclaw-multiagent
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /huamu668-openclaw-multiagent 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v1.0.0
Initial release: Multi-agent Telegram group system with shared workspace
元数据
Slug huamu668-openclaw-multiagent
版本 1.0.0
许可证
累计安装 0
当前安装数 0
历史版本数 1
常见问题

OpenClaw Multi-Agent System 是什么?

Configure multi-agent TG group system with shared Workspace + MemOS memory. Use when user wants to set up multiple AI agents in a Telegram group, with specia... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 297 次。

如何安装 OpenClaw Multi-Agent System?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install huamu668-openclaw-multiagent」即可一键安装,无需额外配置。

OpenClaw Multi-Agent System 是免费的吗?

是的,OpenClaw Multi-Agent System 完全免费(开源免费),可自由下载、安装和使用。

OpenClaw Multi-Agent System 支持哪些平台?

OpenClaw Multi-Agent System 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 OpenClaw Multi-Agent System?

由 huamu668(@huamu668)开发并维护,当前版本 v1.0.0。

💬 留言讨论