Quick Team

v1.1.3

用自然语言快速创建 AI 团队成员的目录结构和配置文件。自动生成 SOUL/IDENTITY/TOOLS 等模板,引导完成 openclaw.json 配置。

0· 153·0 current·0 all-time
byWenlong Zhao@zhao-zwl

Install

OpenClaw Prompt Flow

Install with OpenClaw

Best for remote or guided setup. Copy the exact prompt, then paste it into OpenClaw for zhao-zwl/quick-team.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "Quick Team" (zhao-zwl/quick-team) from ClawHub.
Skill page: https://clawhub.ai/zhao-zwl/quick-team
Keep the work scoped to this skill only.
After install, inspect the skill metadata and help me finish setup.
Use only the metadata you can verify from ClawHub; do not invent missing requirements.
Ask before making any broader environment changes.

Command Line

CLI Commands

Use the direct CLI path if you want to install manually and keep every step visible.

OpenClaw CLI

Bare skill slug

openclaw skills install quick-team

ClawHub CLI

Package manager switcher

npx clawhub@latest install quick-team
Security Scan
Capability signals
Requires sensitive credentials
These labels describe what authority the skill may exercise. They are separate from suspicious or malicious moderation verdicts.
VirusTotalVirusTotal
Benign
View report →
OpenClawOpenClaw
Benign
high confidence
Purpose & Capability
Name/description match the behavior: the SKILL.md instructs creating SOUL/IDENTITY/TOOLS/HEARTBEAT/MEMORY files, adding an entry to agents.list in ~/.qclaw/openclaw.json, and optionally updating agents.defaults.subagents.allowAgents so the controller can spawn the new agent. These operations are expected for a 'quick-team' agent-creation helper.
Instruction Scope
Instructions are narrowly scoped to creating files under the controller workspace and updating openclaw.json, and to restarting the OpenClaw Gateway to activate changes. The skill documents showing a diff and asking for user confirmation before writing or restarting, which limits silent changes. It calls a sessions_spawn-style validation step (a platform API call) — expected for verifying activation. No instructions to send data to external endpoints or to read unrelated system files are present.
Install Mechanism
This is an instruction-only skill with no install spec, no code files to execute, and no external downloads — minimal install risk.
Credentials
The skill requests no environment variables or credentials. Templates mention storing environment-specific configuration or API keys for tools, but the skill does not require any secrets itself. This is proportionate, but users should review any filled-in TOOLS.md for embedded API keys before saving.
Persistence & Privilege
The skill edits runtime configuration (~/.qclaw/openclaw.json) and may change spawn policies (agents.defaults.subagents.allowAgents) and restart the gateway — actions with real operational impact. The metadata documents these permissions and mandates showing diffs and getting explicit user confirmation. always:false and no forced persistence are appropriate; nevertheless, changing allowAgents (especially to ["*"]) expands privileges and should be avoided or tightly restricted.
Assessment
This skill appears coherent with its goal and contains built-in mitigations (showing diffs and asking confirmation). Before installing/using: 1) Back up ~/.qclaw/openclaw.json so you can revert; 2) Carefully review the diff the skill presents and decline any change that adds allowAgents: ["*"] or otherwise broadens spawn permissions; 3) Schedule the gateway restart for a maintenance window — it interrupts running sessions; 4) Inspect generated TOOLS.md or other files for accidentally inserted API keys/secrets before committing them; 5) If you want tighter safety, test on a non-production controller workspace first. If you want me to look over a proposed diff or the generated files before you confirm, paste them and I can review.

Like a lobster shell, security has layers — review code before you run it.

latestvk97bqbymyhqnv5v5nszk6rz9q585fvk5
153downloads
0stars
7versions
Updated 3d ago
v1.1.3
MIT-0

⚠️ 权限提醒:本 skill 会修改 openclaw.json 和重启 Gateway。所有配置变更会先展示 diff 供用户确认,不会静默写入。

Quick Team —— 快速创建团队成员

触发: 用户说「创建一个成员」「帮我新建一个角色」「添加团队成员」「搞个新agent」等。

核心理念: 一个主控,零门槛,三分钟一支团队。


零、快速开始

用户:帮我创建一个负责校对审核的成员,叫火眼
Agent:
    [执行创建流程 → 3分钟完成]
    ✅ 团队新成员:火眼
    workspace: ~/.qclaw/workspace-agent-ba01c6a8/huoyan/
    SOUL.md / IDENTITY.md / TOOLS.md / HEARTBEAT.md / MEMORY.md 已创建

一、创建流程(6步)

步骤1:确认需求

让用户明确回答以下问题(缺一不可):

问题示例作用
角色名称"火眼"、"小策"决定目录名和 Agent ID
核心职责"校对审核"、"策略分析"决定 SOUL.md 的职责描述
性格风格"毒舌"、"温和"、"专业冷静"决定语气和人设
禁止做的事"不横向联系"、"不自作主张"决定禁止事项
汇报方式"任务完成后直接汇报"决定汇报规则

如果用户说不清楚: 主动给选项,让用户选,不要停下来等。


步骤2:生成 Agent ID

规则: 用中文拼音首字母,不能重复。

名称ID
火眼huoyan
小策xiaoce
小编xiaobian
小创xiaochuang

如果重复: 自动加数字后缀,如 xiaoce-2


步骤3:创建文件结构

在主控 workspace 下创建子代理目录:

workspace/
├── huoyan/
│   ├── SOUL.md        ← 必须:人格定义
│   ├── IDENTITY.md    ← 必须:身份标识
│   ├── TOOLS.md       ← 必须:工具备忘
│   ├── HEARTBEAT.md   ← 必须:心跳配置
│   └── MEMORY.md      ← 必须:长期记忆

必须文件说明:

文件必须用途
SOUL.md角色定位 + 职责 + 禁止事项 + 汇报规则
IDENTITY.md名称/Emoji/氛围描述
TOOLS.md常用工具备忘
HEARTBEAT.md周期性任务配置
MEMORY.md长期记忆文件

辅助文件说明:

文件必须用途
AGENTS.md主控才有,记录所有成员配置
USER.md⚠️主控才需要,子代理一般不需要
BOOTSTRAP.md仅首次启动用,创建成员时不需要

步骤4:填充模板

templates/ 目录复制对应模板,根据用户需求填写内容。

重点:SOUL.md 必须包含以下章节(不可删减):

# SOUL.md

## 我是谁
{一句话角色定位}

## 我的职责
1. {具体职责1}
2. {具体职责2}
3. {具体职责3}

## 禁止事项
- {明确不做的事}

## 汇报规则
- {汇报格式和要求}

## 铁律
- {不可逾越的红线}

步骤5:配置 openclaw.json

先展示变更,用户确认后再写入。

需要修改的内容:

  1. agents.list 中追加新成员:
{
  "id": "huoyan",
  "name": "火眼",
  "workspace": "/Users/xxx/.qclaw/workspace-agent-主控ID/huoyan"
}
  1. 关于 allowAgents:如果主控需要 spawn 子代理,需设置 agents.defaults.subagents.allowAgents

推荐最小白名单(仅允许需要的成员):

"allowAgents": ["huoyan"]

通配符 ["*"] 会允许 spawn 任意 agent,存在权限扩大风险,不推荐除非用户明确要求。

操作流程:

  • 展示完整 diff(新增了什么、改了什么)
  • 用户确认后再写入文件
  • 如果用户不确认,跳过此步骤,提示用户稍后手动配置

步骤6:重启 Gateway 并验证

⚠️ 重启 Gateway 会中断当前所有会话和正在运行的任务。请用户确认后再执行。

# 重启前提醒用户
openclaw gateway restart

重启后测试激活:

sessions_spawn({
  agentId: "huoyan",
  cwd: "/Users/xxx/.qclaw/workspace-agent-主控ID/huoyan",
  mode: "run",
  task: "介绍一下你自己"
})

验证通过的标准:

  • ✅ 回复体现正确的角色人格
  • ✅ 没有读错 SOUL.md(比如读成了其他成员的)
  • ✅ 文件路径正确

二、子代理标准目录结构

workspace/
├── 主控workspace/
│   ├── SOUL.md
│   ├── IDENTITY.md
│   ├── AGENTS.md          ← 记录所有成员配置
│   ├── MEMORY.md
│   ├── TOOLS.md
│   ├── HEARTBEAT.md
│   │
│   ├── 子代理A/            ← 每个子代理一个独立目录
│   │   ├── SOUL.md
│   │   ├── IDENTITY.md
│   │   ├── TOOLS.md
│   │   ├── HEARTBEAT.md
│   │   └── MEMORY.md
│   │
│   ├── 子代理B/
│   │   └── ...
│   │
│   └── memory/            ← 主控的日常记忆
│       └── YYYY-MM-DD.md
│
└── skills/                ← 主控的 skills
    └── ...

三、模板说明

模板文件用途备注
SOUL.md角色人格定义核心,必须完整填写
IDENTITY.md身份标识名称/Emoji/氛围
TOOLS.md工具备忘常用命令和配置
HEARTBEAT.md周期性任务保持空可节省API
MEMORY.md长期记忆持续积累信息

examples/assistant/ 目录中有完整示例,可参考格式。


四、常见问题

Q:用户只给了一个名字怎么办?

A: 先根据名字推断角色方向,然后主动给选项:

用户:帮我建一个叫小明
Agent:好的,小明大概负责什么方向?
  A. 执行类——完成具体任务(校对、整理、测试)
  B. 审核类——检查、评分、给意见
  C. 创意类——头脑风暴、策划、提案
  D. 其他(请描述)

Q:spawn 后读错了 SOUL.md?

A: 检查是否同时传了 agentId + cwd。两个参数缺一不可。

Q:新建的成员不汇报?

A: 在 SOUL.md 中明确写「任务完成后立即汇报,不等催」。并在首次派任务时提醒。


五、去AI味铁律

所有新创建的子代理 SOUL.md,必须包含以下检查:

检查项AI爱写人应该写
去AI味"仿佛/像/似乎/好像"用具体动作/感受替代
去格式列表/"首先其次"自然段落
去注释//<!-- -->删除所有注释

Comments

Loading comments...