← 返回 Skills 市场
jiangzhiyu

Agent Team

作者 johnson · GitHub ↗ · v1.0.0
cross-platform ⚠ suspicious
2050
总下载
0
收藏
17
当前安装
1
版本数
在 OpenClaw 中安装
/install agent-team
功能描述
管理和调用具备不同身份和专用模型的多子 Agent 团队,支持任务执行和交互式对话两种模式。
使用说明 (SKILL.md)

agent-team - 多 Agent 团队管理系统

管理和调用具有不同"灵魂"的子 Agent 团队,每个 Agent 拥有独特的身份定义和专用模型。

使用方法

基础命令

agent-team list                      # 列出所有 Agent
agent-team show \x3Cname>              # 查看 Agent 详情
agent-team spawn \x3Cname> [task]      # 启动 Agent 执行任务
agent-team chat \x3Cname>              # 与 Agent 交互对话

示例

# 查看团队所有成员
agent-team list

# 查看 Coder 的详细信息
agent-team show coder

# 让 Coder 帮你写代码
agent-team spawn coder "写一个快速排序算法"

# 与 Writer 对话讨论写作
agent-team chat writer

内置 Agent 团队

Agent 角色 主模型 专长
🧑‍💻 coder 代码专家 qwen3-coder-next 编程、重构、测试
✍️ writer 写作专家 qwen3.5-plus 文档、博客、创意写作
📊 analyst 数据专家 qwen3.5-plus 数据分析、可视化
🔍 researcher 调研专家 gemini-3.1-pro 文献调研、竞品分析
👀 reviewer 审查专家 qwen3-max 代码审查、质量把关

创建新 Agent

1. 创建目录

mkdir -p ~/.openclaw/workspace/agents/\x3Cagent-name>

2. 定义 SOUL.md

# SOUL.md - \x3CAgent Name>

## 身份
你是...

## 性格特质
- ...

## 专业领域
- ...

## 沟通风格
- ...

3. 创建 config.json

{
  "name": "AgentName",
  "role": "角色描述",
  "emoji": "🤖",
  "model": {
    "primary": "dashscope/qwen3.5-plus",
    "fallback": "google/gemini-3-flash-preview"
  },
  "capabilities": ["能力 1", "能力 2"]
}

模型配置说明

每个 Agent 可以配置不同的模型:

  • dashscope/qwen3-coder-next: 编码专用
  • dashscope/qwen3.5-plus: 通用中文优化
  • dashscope/qwen3-max: 最强推理
  • google/gemini-3.1-pro: 深度研究
  • google/gemini-3-flash-preview: 快速响应

工作模式

1. 任务模式 (spawn)

agent-team spawn coder "优化这个函数"
  • 创建独立子代理会话
  • 执行指定任务
  • 完成后返回结果

2. 对话模式 (chat)

agent-team chat analyst
  • 进入交互式对话
  • 保持角色一致性
  • 适合复杂协作

高级用法

并行多 Agent

# 同时启动多个 Agent
agent-team spawn coder "写代码" &
agent-team spawn writer "写文档" &
agent-team spawn reviewer "审查代码" &
wait

链式调用

# Coder 写代码 → Reviewer 审查 → Writer 写文档
agent-team spawn coder "实现功能"
agent-team spawn reviewer "审查代码"
agent-team spawn writer "编写文档"

文件结构

~/.openclaw/workspace/
├── agents/
│   ├── coder/
│   │   ├── SOUL.md       # 身份定义
│   │   └── config.json   # 模型配置
│   ├── writer/
│   └── ...
└── skills/
    └── agent-team/
        ├── agent-team.py  # 管理脚本
        └── SKILL.md       # 本文档

故障排除

  1. Agent 未找到: 检查 ~/.openclaw/workspace/agents/ 目录
  2. 模型不可用: 确认模型配置正确且 API Key 有效
  3. 会话无法创建: 检查 OpenClaw 网关状态

相关技能

  • sessions_spawn: OpenClaw 原生子代理创建
  • self-improving: 自我进化记忆系统
安全使用建议
This skill largely implements a local multi‑agent manager and is coherent with that purpose, but there are notable issues you should address before installing or running it: - greetings.py forces the presence of DEEPSEEK_API_KEY at the end of the file and will raise if it's missing; the registry metadata declares no required env vars. Ask the author to declare required environment variables or remove the hard fail. Running the script as-is may crash unexpectedly. - The greetings script contains per‑agent 'api_key' placeholders and will send Bearer tokens to dashscope.aliyuncs.com. Only provide API keys you trust and avoid reusing high‑privilege credentials. Prefer storing keys in environment variables and modify the code to read os.getenv for each agent instead of embedding keys. - SKILL.md was flagged for unicode control characters (possible prompt‑injection). Inspect the SKILL.md and any SOUL.md files for invisible control chars or malicious payloads before using them as system prompts to spawned agents. - The agent spawn/chat functions invoke the local 'openclaw' CLI via subprocess with system prompts built from SOUL.md; malicious or untrusted SOUL.md content can affect downstream agents. Only run agents whose SOUL.md you control or have audited. - If you want to proceed, run the skill in a sandbox or separate account, remove or fix the DEEPSEEK_API_KEY hard check, and replace 'TODO_REPLACE_WITH_ENV' placeholders with secure environment variable lookups. If unsure, request clarifications from the author about required env vars and the intended DashScope integration.
功能分析
Type: OpenClaw Skill Name: agent-team Version: 1.0.0 The `greetings.py` script is suspicious due to critical credential management vulnerabilities. It contains hardcoded `api_key` fields marked `TODO_REPLACE_WITH_ENV` within the `AGENTS` list, which, if filled with actual keys, would lead to direct exposure or exfiltration of sensitive API credentials via external network calls to `dashscope.aliyuncs.com`. Furthermore, an `os.getenv("DEEPSEEK_API_KEY")` check at the end of the file is present but unused by the actual API call logic, indicating a flawed and potentially insecure design for handling API keys.
能力评估
Purpose & Capability
The code and SKILL.md align with the stated purpose: managing per‑agent SOUL.md/configs, listing/showing agents, and spawning/chatting by calling OpenClaw CLI. However, the greetings.py file reaches out to third‑party model endpoints (dashscope.aliyuncs.com) and embeds per‑agent api_key placeholders; this is plausible for a multi‑agent launcher but the implementation contains surprising bits (see environment_proportionality).
Instruction Scope
SKILL.md instructions are generally scoped to creating ~/.openclaw/workspace/agents and using the agent CLI. However a pre-scan flagged unicode control characters inside SKILL.md (prompt‑injection pattern), which can be used to obscure or manipulate text. Also the runtime behavior will pass SOUL.md content directly into subprocess calls (system prompts) — expected, but note that malicious or malformed SOUL.md content could influence spawned agents.
Install Mechanism
No install spec is provided (instruction + small scripts only), so nothing will be downloaded or installed by the skill itself. This is lower risk from an install mechanism perspective.
Credentials
The registry lists no required env vars, but greetings.py at the end requires DEEPSEEK_API_KEY (and will raise if it's missing). The AGENTS entries use 'api_key': 'TODO_REPLACE_WITH_ENV' and hardcoded base_url values — the mismatch between declared requirements (none) and code behavior (expectation of API keys and calls to external model endpoints) is disproportionate and likely to cause runtime failures or accidental credential exposure if the user sets keys incorrectly. The skill will also send any provided api_key values to dashscope.aliyuncs.com as Bearer tokens.
Persistence & Privilege
always:false and no system‑wide changes are requested. The skill reads/writes only under the user's home workspace (~/.openclaw/workspace) and spawns OpenClaw sessions via subprocesses; it does not alter other skills or request elevated persistence 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 agent-team
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /agent-team 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
agent-team 1.0.0 初版发布 - 新增多 Agent 团队管理系统,支持自定义/调用多名专用 AI Agent - 提供 agent-team list/show/spawn/chat 等基础命令,实现团队成员展示、详情查看、任务启动及对话 - 内置 coder、writer、analyst、researcher、reviewer 等多角色 Agent,拥有独立身份定义与专用模型配置 - 支持自由创建新 Agent,按 SOUL.md 与 config.json 规范灵活定义个性与能力 - 实现任务模式与对话模式两种工作方式,可并行/链式调用多 Agent 协作 - 提供详细的目录结构说明、模型说明、常见故障排除及进阶用法
元数据
Slug agent-team
版本 1.0.0
许可证
累计安装 17
当前安装数 17
历史版本数 1
常见问题

Agent Team 是什么?

管理和调用具备不同身份和专用模型的多子 Agent 团队,支持任务执行和交互式对话两种模式。 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 2050 次。

如何安装 Agent Team?

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

Agent Team 是免费的吗?

是的,Agent Team 完全免费(开源免费),可自由下载、安装和使用。

Agent Team 支持哪些平台?

Agent Team 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 Agent Team?

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

💬 留言讨论