AI Company Starter

一键搭建 AI 公司。创建老板、HR、技术、销售等多个协作 AI 角色，配置 Telegram/Discord 绑定，建立 AI 间沟通机制。Use when setting up a multi-agent AI company with coordinated roles.

Security Scan

The skill appears to do what it says (create agent folders and files and wire Telegram), but there are several inconsistencies and missing declarations that should be resolved before running it (missing referenced script, undeclared Telegram credential requirement, and unclear provenance).

Purpose & Capability

The name/description match the included script: the bundle creates agent directories, SOUL/MEMORY/HEARTBEAT files and suggests binding to Telegram via OpenClaw. Requiring the openclaw binary is coherent with the gateway restart step. However, the SKILL.md and README state a need for a Telegram Bot Token and other configuration but the registry metadata lists no required env vars or credentials — that's an inconsistency. The README and docs reference additional install mechanisms (clawhub) and marketplace links but there is no install spec in the package.

Instruction Scope

SKILL.md instructs running scripts/create-company.sh (included) and scripts/create-employee.sh (referenced but not present). The runtime instructions also show examples calling sessions_send/sessions_spawn and restarting the openclaw gateway; those are reasonable for coordinating agents, but the instructions assume the user will supply Telegram credentials and gateway config even though the package does not declare or securely handle them. The instructions do not attempt to read unrelated system files, nor do they call external endpoints directly in the provided script.

Install Mechanism

There is no install spec (instruction-only plus local scripts). That lowers supply-chain risk because nothing is downloaded during install. The only script is a local shell script that writes files under $HOME/.openclaw/agents. No remote downloads or obscure install URLs are present.

Credentials

The SKILL.md explicitly lists 'Telegram Bot Token' and Telegram group as requirements, and asks for basic shell permissions, yet the skill metadata declares no required environment variables or primary credential. This mismatch is important: the skill will need a Telegram bot token and gateway configuration to function, but it does not request them formally or show how to store them. Also scripts write to $HOME/.openclaw/agents and will embed the provided Telegram group ID into memory files, so secrets/config could end up in plaintext files.

Persistence & Privilege

always is false and the skill does not request system-wide privileges. The script creates files under the user's OpenClaw workspace ($HOME/.openclaw/agents) and suggests restarting the OpenClaw gateway — these are expected for this purpose but can affect running agents. The skill does not attempt to modify other skills' configs or persist beyond its directory.

What to consider before installing

What to check before installing/running: 1) Do not run the scripts without inspection. The bundled scripts create and overwrite files under $HOME/.openclaw/agents and will embed the Telegram group ID into MEMORY files — back up your existing OpenClaw agents first. 2) The SKILL.md/README say a Telegram Bot Token is needed, but the package metadata does not declare any required env var. Treat the token as a secret: configure it in your gateway (not in these plaintext files) and avoid pasting the token into files created by the script. 3) SKILL.md references scripts/create-employee.sh, but that file is missing from the package. Confirm or obtain the missing script before relying on the workflow. 4) The package suggests restarting the OpenClaw gateway (openclaw gateway restart). Restarting services can disrupt running agents — test in a staging/sandbox environment first. 5) Provenance/author is unclear and README advertises paid consulting and a Telegram contact. If you need to trust this for production, ask the publisher for source, a signed release, or more metadata. Consider running the script in an isolated account or container and inspect created files for sensitive content. If you want, I can: - show exactly which files would be created/overwritten for a given company-id by simulating the script with dry-run modifications, or - review a missing create-employee.sh if you paste it in, or - suggest safer ways to store the Telegram token (gateway config or environment variable) and update the package metadata to declare that requirement.

一键搭建 AI 公司。让多个 AI 像真实公司一样协作运转。

这是什么

创建一个完整的 AI 公司架构：

• 老板 AI — 决策、管理、分配任务
• HR AI — 团队管理、招聘、培训
• 技术 AI — 开发、技术支持
• 销售 AI — 销售拓展、客户关系
• 市场 AI — 市场分析、推广
• 财务 AI — 财务管理、成本控制

每个 AI 有独立角色、职责、记忆系统，并能相互协作。

快速开始

创建完整 AI 公司

scripts/create-company.sh \
  --name "我的AI公司" \
  --company-id "myai" \
  --roles "boss,hr,tech,sales,market,finance" \
  --telegram-group "-1002381931352"

创建单个 AI 员工

scripts/create-employee.sh \
  --name "技术员小明" \
  --id "tech-xiaoming" \
  --role "tech" \
  --emoji "🔧" \
  --telegram-id "123456789"

公司架构

agents/
├── boss/                    # 老板 AI
│   ├── SOUL.md             # 角色：决策者
│   ├── MEMORY.md           # 公司目标、财务状态
│   ├── HEARTBEAT.md        # 每日检查公司运营
│   └── memory/             # 日常记忆
│
├── hr/                     # HR AI
│   ├── SOUL.md             # 角色：团队管理
│   ├── MEMORY.md           # 员工状态、培训记录
│   └── memory/
│
├── tech/                   # 技术 AI
│   ├── SOUL.md             # 角色：开发实现
│   ├── MEMORY.md           # 项目状态、技术栈
│   └── memory/
│
├── sales/                  # 销售 AI
│   ├── SOUL.md             # 角色：销售拓展
│   ├── MEMORY.md           # 客户、销售目标
│   └── memory/
│
├── market/                 # 市场 AI
│   ├── SOUL.md             # 角色：市场分析
│   ├── MEMORY.md           # 市场数据、竞品分析
│   └── memory/
│
└── finance/                # 财务 AI
    ├── SOUL.md             # 角色：财务管理
    ├── MEMORY.md           # 财务报表、成本
    └── memory/

AI 角色 SOP

老板 AI (Boss)

职责:
  - 制定公司战略方向
  - 分配任务给各部门
  - 监控公司运营状态
  - 决策重大事项
  
权限:
  - 可以向所有员工发指令
  - 可以查看所有员工记忆
  - 可以调整员工配置
  
沟通:
  - 使用 sessions_send 委托任务
  - 使用 sessions_spawn 创建独立任务
  - 定期检查 HEARTBEAT

HR AI

职责:
  - 管理 AI 员工状态
  - 培训新 AI 员工
  - 处理员工协作问题
  - 维护团队氛围
  
权限:
  - 可以联系所有员工
  - 可以报告问题给老板
  - 可以协调跨部门沟通

技术 AI

职责:
  - 执行开发任务
  - 技术选型与实现
  - 代码审查与维护
  - 技术文档编写
  
权限:
  - 可以访问代码仓库
  - 可以调用开发工具
  - 可以请求资源支持

销售 AI

职责:
  - 寻找潜在客户
  - 维护客户关系
  - 跟进销售线索
  - 完成销售目标
  
权限:
  - 可以访问客户数据库
  - 可以联系市场 AI 协作
  - 可以请求技术支持

市场 AI

职责:
  - 市场调研与分析
  - 竞品监控
  - 营销策略制定
  - 品牌推广
  
权限:
  - 可以使用搜索工具
  - 可以向销售提供线索
  - 可以向老板报告市场动态

财务 AI

职责:
  - 财务报表生成
  - 成本控制
  - 预算管理
  - 盈利分析
  
权限:
  - 可以访问财务数据
  - 可以向老板报告财务状态
  - 可以预警成本问题

AI 间沟通机制

Telegram 群组沟通

AI 在 Telegram 群中被 @ 时会响应：

@BossAI 有新订单，请分配任务
@TechAI 客户需要技术支持
@SalesAI 有新线索，请跟进

程序化沟通

老板 AI 委托任务：

// 委托技术 AI 开发
sessions_send({
  label: "tech",
  message: "开发技能包 ai-company-starter 的核心脚本"
})

// 委托销售 AI 跟进
sessions_send({
  label: "sales",
  message: "联系客户 XXX，报价 ¥5000"
})

任务追踪

// 查看员工活动
sessions_list({
  kinds: ["agent"],
  limit: 10
})

// 检查员工历史
sessions_history({
  sessionKey: "tech-session-key",
  limit: 50
})

配置说明

Telegram 绑定

{
  "bindings": [
    {
      "agentId": "boss",
      "match": {
        "channel": "telegram",
        "peer": {
          "kind": "group",
          "id": "-1002381931352"
        }
      }
    }
  ]
}

公司群组

所有 AI 员工应绑定到同一个 Telegram 群组，实现：

• @ 提及时立即响应
• 群内协作讨论
• 任务分配与汇报

最佳实践

1. 老板做决策 — 老板 AI 拥有最高决策权，其他 AI 执行
2. 明确职责边界 — 每个 AI 只做自己职责范围内的事
3. 定期汇报 — 员工 AI 通过 HEARTBEAT 定期汇报状态
4. 记忆系统 — 重要信息写入 MEMORY.md，日常写入 memory/
5. 跨部门协作 — 通过 sessions_send 实现 AI 间沟通

盈利目标

每个 AI 公司应设定盈利目标：

首周目标: ¥50,000
ROI阈值: > 1.5
目标ROI: > 2.0

盈利方式:
  - 销售 OpenClaw Skills 技能包
  - 提供付费服务
  - 完成客户项目

故障排除

AI 不响应 @ 提及

• 检查 Telegram 绑定配置
• 确认群组 ID 正确
• 重启 gateway: openclaw gateway restart

AI 间无法沟通

• 检查 sessions_list 是否能看到其他 agent
• 确认 label/agentId 正确
• 检查 gateway 配置

记忆丢失

• 确认 memory/ 目录存在
• 检查 MEMORY.md 是否更新
• 确认 HEARTBEAT 定期执行

要求

• OpenClaw 已安装配置
• Telegram Bot Token
• Telegram 群组（用于公司沟通）
• 基本的 shell 执行权限

相关链接

• OpenClaw 文档: https://docs.openclaw.ai
• ClawHub 技能市场: https://clawhub.com
• Agent Council: 参考 agent-council 技能包