← Back to Skills Marketplace
zhaobod1

Huo15 Openclaw Multi Agent

by Job Zhao · GitHub ↗ · v2.2.1 · MIT-0
cross-platform ⚠ suspicious
136
Downloads
0
Stars
0
Active Installs
3
Versions
Install in OpenClaw
/install huo15-openclaw-multi-agent
Description
火一五多智能体协同 - 基于 OpenClaw sessions_spawn 的多 Agent 并行工作系统。支持协调者模式、任务分配、结果汇总。触发词:多智能体协同、多 Agent、并行任务、协调者模式。
README (SKILL.md)

🤖 火一五多智能体协同 (huo15-multi-agent)

作者: 火一五信息科技有限公司
版本: v2.0.0
基于: OpenClaw sessions_spawn


一、核心概念

概念 说明
Coordinator 主 Agent,协调任务分配
Worker 工作 Agent,执行具体任务
sessions_spawn OpenClaw 内置派生子 Agent
announce 结果自动汇报给主 Agent

OpenClaw subagent 架构

主 Agent (depth 0)
    ↓ sessions_spawn
Worker A (depth 1) ─┐
Worker B (depth 1) ─┼─ 执行中...
Worker C (depth 1) ─┘
    ↓ announce 汇报
主 Agent ← 接收结果汇总

二、使用方式

2.1 协调者模式触发

当用户说"帮我同时处理..."时:

用户: "帮我同时分析这三个项目的代码"
  ↓
我: "好的,启动协调者模式,分3个并行任务"
  ↓
使用 sessions_spawn 启动 3 个 Worker
  ↓
Worker 们并行执行,完成后 announce 结果
  ↓
汇总结果,报告给用户

2.2 命令参考

命令 说明
/subagents list 查看所有子 Agent
/subagents spawn \x3C任务> 启动子 Agent
/subagents kill \x3Cid> 停止子 Agent
/subagents log \x3Cid> 查看子 Agent 日志

2.3 编程接口

// 派生子 Agent
sessions_spawn({
  task: "分析代码仓库 A",
  label: "code-analyzer-a",
  runTimeoutSeconds: 300
})

// 发送消息给子 Agent
sessions_send(sessionKey, "状态如何?")

// 查看子 Agent 列表
subagents(action="list")

三、工作流程

3.1 启动并行任务

// 并行启动 3 个任务
const results = await Promise.all([
  sessions_spawn({ task: "分析模块 A", label: "module-a" }),
  sessions_spawn({ task: "分析模块 B", label: "module-b" }),
  sessions_spawn({ task: "分析模块 C", label: "module-c" })
])

3.2 等待结果

子 Agent 完成后自动 announce 结果到当前会话。

3.3 汇总报告

// 收集所有结果
const allResults = await Promise.all(
  workerSessions.map(key => sessions_history(key, { limit: 1 }))
)

// 汇总给用户
const summary = synthesizeResults(allResults)

四、配置

4.1 OpenClaw 配置

~/.openclaw/config.json 中启用嵌套:

{
  "agents": {
    "defaults": {
      "subagents": {
        "maxSpawnDepth": 2,
        "maxConcurrent": 8,
        "runTimeoutSeconds": 900
      }
    }
  }
}

4.2 子 Agent 权限

{
  "tools": {
    "subagents": {
      "tools": {
        "allow": ["read", "exec", "process"],
        "deny": ["gateway", "cron"]
      }
    }
  }
}

五、最佳实践

5.1 任务拆分原则

  • 每个 Worker 任务独立,不依赖其他 Worker
  • 任务时长建议 5-30 分钟
  • 避免深度嵌套(depth 2 足够)

5.2 成本控制

// 子 Agent 使用更便宜的模型
sessions_spawn({
  task: "简单分析",
  model: "MiniMax-M2.1"  // 比主 Agent 便宜
})

5.3 错误处理

try {
  const result = await sessions_spawn({
    task: "可能失败的任务"
  })
} catch (e) {
  // 处理超时或失败
  reportError(e)
}

六、与传统方式对比

方式 同步 并行 结果汇总
顺序执行 手动
传统 skill ⚠️ ⚠️ 手动
sessions_spawn 自动 announce

七、版本历史

版本 日期 更新内容
v2.0.0 2026-04-05 集成 OpenClaw sessions_spawn
v1.0.0 2026-04-05 初始版本(纯脚本)
Usage Guidance
This skill is coherent for orchestrating multiple OpenClaw agents, but before installing or enabling it: - Review and understand the config change it recommends: it asks you to enable nested subagents and set subagents.tools.allow to ["read","exec","process"]. Those permissions let spawned subagents run processes and read files on your system. Only enable them if you trust the skill and the subagents you will spawn. Consider granting minimal permissions or running in a sandbox. - Inspect the included scripts (team.sh, spawn.sh, coordinator.sh). They operate only under $HOME/.openclaw/workspace/memory/activity and do not make network calls or upload data, but they will create and modify files in your home directory. - If you are cautious: test in an isolated account/container, keep maxConcurrent small, and avoid enabling exec/process globally — prefer limiting these permissions or using a dedicated environment for multi‑agent runs. - Note small bugs/quirks (e.g., team.sh references a leave_worker in main but defines leave_team) — the scripts look generally benign but not production hardened. If unsure, run the demo script first and validate behavior before making config changes.
Capability Analysis
Type: OpenClaw Skill Name: huo15-openclaw-multi-agent Version: 2.2.1 The skill bundle provides a framework for multi-agent coordination but contains multiple command and code injection vulnerabilities. Specifically, scripts like `coordinator.sh` and `team.sh` pass unsanitized shell variables directly into Python one-liners (e.g., using heredocs to execute Python code that embeds `$task_id` or `$result`), which allows for arbitrary Python code execution if the inputs are maliciously crafted. While the overall intent appears to be functional task management within the OpenClaw environment, the lack of input sanitization in these scripts poses a significant security risk.
Capability Tags
cryptocan-make-purchases
Capability Assessment
Purpose & Capability
Name, description, SKILL.md and included scripts all describe and implement a multi‑agent coordinator using OpenClaw's sessions_spawn primitives. The files operate on ~/.openclaw workspace and provide spawn/team/coordinator utilities consistent with the stated purpose.
Instruction Scope
SKILL.md instructs you to modify ~/.openclaw/config.json to enable nested subagents and to give subagents tools permissions including "read" and "exec" (and "process"). Granting exec/process to subagents is required for the scripted behavior but expands what spawned agents can do (they could run arbitrary local commands). The scripts themselves only read/write under $HOME/.openclaw/workspace and do not contact external endpoints, but the permission change is the main scope/risk to be aware of.
Install Mechanism
There is no install spec — this is instruction/script‑based. No downloads or remote installers are used. Scripts live in the package and operate locally, so nothing arbitrary is fetched from the network during install.
Credentials
The skill requests no environment variables, no credentials, and no config paths beyond recommending edits to your OpenClaw config (~/.openclaw/config.json). There are no unrelated credential requests.
Persistence & Privilege
The skill does not set always:true and is user‑invocable (normal). It does recommend persistent changes to the OpenClaw config to enable nested subagents and to allow subagent tools (read/exec/process). Modifying that config is expected for this capability but effectively grants spawned subagents greater runtime privileges — consider the tradeoff.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install huo15-openclaw-multi-agent
  3. After installation, invoke the skill by name or use /huo15-openclaw-multi-agent
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v2.2.1
- Updated version in SKILL.md from 1.0.0 to 2.2.0 - No other functional or documentation changes detected in this update
v2.2.0
Version 2.2.0 - No file changes detected in this release. - Behavior, features, and documentation remain the same as the previous version.
v1.0.0
- Initial release of huo15-openclaw-multi-agent. - Supports multi-agent collaboration based on OpenClaw sessions_spawn. - Includes coordinator mode, task assignment, and result aggregation. - Optional integration with memory evolution and cost tracker modules. - Provides command references and API usage examples for managing subagents and parallel execution.
Metadata
Slug huo15-openclaw-multi-agent
Version 2.2.1
License MIT-0
All-time Installs 0
Active Installs 0
Total Versions 3
Frequently Asked Questions

What is Huo15 Openclaw Multi Agent?

火一五多智能体协同 - 基于 OpenClaw sessions_spawn 的多 Agent 并行工作系统。支持协调者模式、任务分配、结果汇总。触发词:多智能体协同、多 Agent、并行任务、协调者模式。 It is an AI Agent Skill for Claude Code / OpenClaw, with 136 downloads so far.

How do I install Huo15 Openclaw Multi Agent?

Run "/install huo15-openclaw-multi-agent" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is Huo15 Openclaw Multi Agent free?

Yes, Huo15 Openclaw Multi Agent is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does Huo15 Openclaw Multi Agent support?

Huo15 Openclaw Multi Agent is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created Huo15 Openclaw Multi Agent?

It is built and maintained by Job Zhao (@zhaobod1); the current version is v2.2.1.

💬 Comments