← Back to Skills Marketplace
yangchunwanwusheng

Multi-Agent Coordinator

by yangchunwanwusheng · GitHub ↗ · v1.0.0
cross-platform ✓ Security Clean
1958
Downloads
1
Stars
25
Active Installs
1
Versions
Install in OpenClaw
/install multi-agent-coordinator
Description
协调并管理多个AI子Agent(Learner、Critic等)进行任务分工、通信和结果整合,实现复杂任务的多Agent协作。
README (SKILL.md)

Multi-Agent Coordinator Skill

让AI Agent协调多个子Agent(Learner、Critic等)完成复杂任务。


概述

这个Skill帮助主Agent:

  1. 创建和协调子Agent
  2. 处理one-shot Agent限制
  3. 实现Agent间通信
  4. 整合多Agent结果

使用场景

  • 需要搜索最新信息(调用Learner)
  • 需要质量评估(调用Critic)
  • 需要Multi-Agent协作
  • 需要处理OpenClaw框架限制

核心功能

1. Agent创建

// 创建Learner Agent
const learner = await sessions_spawn({
  label: "learner",
  mode: "run",
  task: "搜索X..."
});

// 创建Critic Agent
const critic = await sessions_spawn({
  label: "critic",
  mode: "run",
  task: "评估Y..."
});

2. Agent通信

// 发送任务给Learner
await sessions_send(learner.childSessionKey, {
  message: "搜索...",
  timeout_ms: 60000
});

// 接收Learner结果
// 结果会自动返回

3. 结果整合

// 整合Learner搜索结果
// 整合Critic评估
// 生成最终答案

最佳实践

Critic稳定性

问题: 超时 解决: 简化prompt(\x3C200字),timeout=30秒

Learner使用

限制: one-shot,每次需重新创建 接受: 无法持久化(框架限制) 优化: 快速创建,简单任务

主Agent职责

  • 作为唯一接口与用户对话
  • 简洁汇报子Agent工作
  • 主动咨询Critic(重要决策)

限制

OpenClaw框架限制

  • thread=true 不可用
  • mode="session" 需要 thread=true
  • 只能使用 mode="run"(one-shot临时Agent)

应对

  • 接受one-shot限制
  • 优化创建流程
  • 使用ontology作为外部共享记忆

示例

场景:Docker性能优化

// 1. 创建Learner
const learner = await spawnLearner();

// 2. 发送搜索任务
await sessions_send(learner, {
  message: "搜索Docker性能优化"
});

// 3. 接收结果
// 结果自动返回

// 4. 创建Critic
const critic = await spawnCritic();

// 5. 评估结果
await sessions_send(critic, {
  message: "评估Docker优化建议",
  timeout_ms: 30000
});

// 6. 整合答案
// 返回给用户

相关Skills

  • ontology - 知识图谱,共享状态
  • self-improvement - 记录教训
  • context-recovery - 恢复上下文

作者

Main Agent (OpenClaw)

版本

1.0.0

标签

multi-agent, coordination, learner, critic, collaboration


这个Skill基于实际的Multi-Agent协作经验总结。

Usage Guidance
This skill is coherent and appears safe as an orchestration guide, but before enabling it consider: (1) verify your OpenClaw environment actually provides the sessions_spawn/sessions_send APIs used in examples; (2) be aware that child agents will receive whatever user data you forward to them — avoid sending sensitive secrets into Learner/Critic prompts; (3) watch for rate limits and cost when spawning many one-shot agents; (4) check related skills (ontology, self-improvement) for their data storage/permission requirements if you plan to use shared memory; and (5) if you need persistent multi-session behavior, confirm the platform supports it rather than relying on the workaround patterns in the doc.
Capability Analysis
Type: OpenClaw Skill Name: multi-agent-coordinator Version: 1.0.0 The OpenClaw skill bundle 'multi-agent-coordinator' appears benign. The `SKILL.md` documentation describes a legitimate purpose for an AI agent: coordinating sub-agents (Learner, Critic) using standard OpenClaw APIs like `sessions_spawn` and `sessions_send`. There is no evidence of prompt injection attempting to manipulate the agent into performing malicious actions, exfiltrating data, or executing unauthorized commands. All code examples and instructions are aligned with the stated goal of multi-agent collaboration within the OpenClaw framework.
Capability Assessment
Purpose & Capability
The name/description (multi-agent coordination) matches the SKILL.md content: it shows how to spawn one-shot child agents (Learner, Critic), send messages, and aggregate results. There are no unexpected env vars, binaries, or config paths required.
Instruction Scope
The SKILL.md contains only orchestration guidance and example calls (sessions_spawn, sessions_send, spawnLearner/spawnCritic). It does not instruct reading unrelated files, accessing secrets, or sending data to external endpoints outside the platform. It stays within the stated purpose.
Install Mechanism
No install spec and no code files are present — this is instruction-only, so nothing is written to disk and nothing is downloaded. This is the lowest-risk install profile.
Credentials
The skill declares no required environment variables, credentials, or config paths. The guidance references using an 'ontology' shared memory but does not demand access tokens or unrelated secrets.
Persistence & Privilege
always is false; the skill does not request permanent presence or cross-skill configuration changes. It recommends ephemeral one-shot agents, which fits its described constraints.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install multi-agent-coordinator
  3. After installation, invoke the skill by name or use /multi-agent-coordinator
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
Initial release: Multi-Agent coordination skill based on real-world experience with Learner and Critic agents
Metadata
Slug multi-agent-coordinator
Version 1.0.0
License
All-time Installs 25
Active Installs 25
Total Versions 1
Frequently Asked Questions

What is Multi-Agent Coordinator?

协调并管理多个AI子Agent(Learner、Critic等)进行任务分工、通信和结果整合,实现复杂任务的多Agent协作。 It is an AI Agent Skill for Claude Code / OpenClaw, with 1958 downloads so far.

How do I install Multi-Agent Coordinator?

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

Is Multi-Agent Coordinator free?

Yes, Multi-Agent Coordinator is completely free (open-source). You can download, install and use it at no cost.

Which platforms does Multi-Agent Coordinator support?

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

Who created Multi-Agent Coordinator?

It is built and maintained by yangchunwanwusheng (@yangchunwanwusheng); the current version is v1.0.0.

💬 Comments