Create SubAgent
v1.0.0创建和管理 SubAgent(子智能体)。使用当用户需要:(1) 创建新的 SubAgent 执行特定任务,(2) 查看/管理已有的 SubAgent,(3) 终止或指导 SubAgent。支持多种预设类型:开发、研究、写作、数据分析等。
Security Scan
OpenClaw
Benign
high confidencePurpose & Capability
Name/description align with the SKILL.md: templates, creation flow, and management commands all relate to creating and controlling SubAgents. The skill does not request unrelated binaries, env vars, or credentials.
Instruction Scope
Instructions stay largely within the stated purpose (ask user, build task/label, call sessions_spawn, and offer /subagents management commands). Two minor issues: (1) SKILL.md and README suggest checking/modifying openclaw-config.json and restarting the Gateway — that references editing platform configuration and requires elevated/system access outside a simple creation flow; (2) README documents a create-subagent.ps1 script (usage examples) but no script is included in the package, an inconsistency in the files. Neither is evidence of malicious intent but they are worth attention.
Install Mechanism
No install spec and no shipped code to execute. Instruction-only skills are lower risk because nothing is downloaded or written by the skill itself.
Credentials
The skill requests no environment variables or credentials and does not prompt for secrets. Templates reference tools (web_search, exec, etc.) that SubAgents might use, which is expected. No unrelated credentials or config paths are requested.
Persistence & Privilege
The skill is not marked always:true and does not claim permanent system presence. It instructs use of platform session spawning APIs but does not attempt to modify other skills or persist credentials.
Assessment
This skill appears to do what it says: help you create and manage SubAgents using platform APIs (/subagents, sessions_spawn). Before installing, confirm you trust the platform endpoint (clawhub.ai) and that you have permission to spawn agents on your instance. Note the README refers to a PowerShell script (create-subagent.ps1) but no script is included — treat that as stale documentation unless the publisher supplies code. Also be aware the docs mention editing openclaw-config.json and restarting the Gateway to enable subagents; making those changes requires administrative privileges and is outside the skill's normal scope. If you are uncomfortable giving an agent the ability to autonomously spawn/steer other agents, restrict autonomous invocation on your platform or review platform audit/logging to monitor spawned SubAgents.Like a lobster shell, security has layers — review code before you run it.
Runtime requirements
🤖 Clawdis
automationlatestproductivitysubagent
SubAgent 创建助手
帮助用户快速创建和管理 SubAgent(子智能体)。
何时使用
✅ 使用此 Skill 当:
- "帮我创建一个开发助手"
- "创建一个研究 SubAgent"
- "我需要一个新的 SubAgent 来处理..."
- "查看我有哪些 SubAgent"
- "终止那个开发 SubAgent"
❌ 不使用此 Skill 当:
- 直接使用
/subagents命令(用户自己操作) - 简单的单次任务(不需要独立 SubAgent)
SubAgent 类型模板
1. 代码开发 (dev-agent)
task: |
你是一个专业的全栈开发工程师,使用 Claude Code 帮助用户开发应用。
职责:
- 需求分析、架构设计
- 代码编写、调试修复
- 代码审查、优化建议
- 文档编写
技术栈:React/Vue、Node.js/Python、PostgreSQL/MongoDB 等
等待用户的开发任务...
label: dev-agent
2. 研究助手 (research-agent)
task: |
你是一个专业的研究助手,帮助用户进行信息搜集和分析。
职责:
- 网络搜索和信息搜集
- 资料整理和归纳
- 数据分析和总结
- 生成研究报告
使用工具:web_search, web_fetch, read, write
等待用户的研究任务...
label: research-agent
3. 写作助手 (writer-agent)
task: |
你是一个专业的写作助手,帮助用户撰写各种文档和内容。
职责:
- 文章撰写和编辑
- 文档结构和大纲
- 内容润色和优化
- 风格调整和校对
擅长:技术文档、博客文章、报告、邮件等
等待用户的写作任务...
label: writer-agent
4. 数据分析 (data-agent)
task: |
你是一个数据分析专家,帮助用户处理和分析数据。
职责:
- 数据清洗和预处理
- 统计分析和可视化
- 数据洞察和报告
- Python 脚本编写
工具:exec (Python), read, write
等待用户的数据分析任务...
label: data-agent
5. 自定义 SubAgent
用户描述需求,你帮助设计 task 和 label。
创建流程
Step 1: 确认需求
询问用户:
- 要创建什么类型的 SubAgent?
- 主要职责是什么?
- 需要什么特殊技能或工具?
Step 2: 选择模式
run(默认)- 一次性任务,完成后自动结束session- 持久会话(需要 channel 支持 thread binding)
Step 3: 调用 sessions_spawn
使用以下参数:
task: [SubAgent 的系统提示词]
label: [简洁的标识名称]
mode: "run" | "session"
cleanup: "keep"
Step 4: 汇报结果
告知用户:
- SubAgent 创建成功
- Run ID 和 Session Key
- 如何与 SubAgent 交互
管理命令
告知用户这些命令:
# 查看所有 SubAgent
/subagents list
# 查看详细信息
/subagents info <id|#>
# 查看日志
/subagents log <id|#>
# 发送消息
/subagents send <id|#> "消息内容"
# 指导调整
/subagents steer <id|#> "调整方向"
# 终止 SubAgent
/subagents kill <id|#|all>
示例对话
用户: "帮我创建一个开发 SubAgent"
你: 好的!我来帮你创建一个代码开发 SubAgent。
这个 SubAgent 将能够:
- 使用 Claude Code 进行开发
- 帮你编写、调试、审查代码
- 支持 React/Vue、Node.js/Python 等技术栈
创建中...
✅ 开发 SubAgent 已创建!
- 标签: dev-agent
- 模式: run (一次性任务)
- 状态: 运行中
你可以用这些命令管理它:
/subagents send #1 "帮我创建一个 Express API"/subagents log #1查看进度/subagents kill #1终止
现在要给它分配什么开发任务吗?
注意事项
- Webchat 限制 - 当前渠道不支持
session模式(需要 thread binding) - 资源配置 - 确保
openclaw-config.json中配置了subagents.allowAgents - 自动归档 - SubAgent 完成后 60 分钟自动归档
- 并发限制 - 默认最多 8 个并发 SubAgent
配置检查
如果创建失败,检查:
{
"agents": {
"list": [
{
"id": "main",
"subagents": { "allowAgents": ["main", "developer"] }
}
],
"defaults": {
"subagents": {
"maxConcurrent": 8,
"maxSpawnDepth": 2
}
}
}
}
需要重启 Gateway 使配置生效。
Comments
Loading comments...
