agent-organizer
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
ChineseAgent Organizer
Agent 编排器
Purpose
用途
Provides expertise in multi-agent system architecture, coordination patterns, and autonomous workflow design. Handles agent decomposition, communication protocols, and collaboration strategies for complex AI systems.
提供多智能体系统架构、协调模式和自主工作流设计方面的专业能力。处理复杂AI系统中的智能体拆分、通信协议和协作策略问题。
When to Use
适用场景
- Designing multi-agent architectures or agent teams
- Implementing agent-to-agent communication protocols
- Building hierarchical or swarm-based agent systems
- Orchestrating autonomous workflows across agents
- Debugging agent coordination failures
- Scaling agent systems for production
- Designing agent memory sharing strategies
- 设计多智能体架构或智能体团队
- 实现智能体间通信协议
- 构建层级式或群体式智能体系统
- 跨智能体编排自主工作流
- 调试智能体协调故障
- 为生产环境扩展智能体系统
- 设计智能体内存共享策略
Quick Start
快速入门
Invoke this skill when:
- Designing multi-agent architectures or agent teams
- Implementing agent-to-agent communication protocols
- Building hierarchical or swarm-based agent systems
- Orchestrating autonomous workflows across agents
- Scaling agent systems for production
Do NOT invoke when:
- Building single-agent LLM applications (use ai-engineer)
- Optimizing prompts for individual agents (use prompt-engineer)
- Managing agent context windows (use context-manager)
- Handling agent failures and recovery (use error-coordinator)
在以下场景调用此技能：
- 设计多智能体架构或智能体团队
- 实现智能体间通信协议
- 构建层级式或群体式智能体系统
- 跨智能体编排自主工作流
- 为生产环境扩展智能体系统
请勿在以下场景调用：
- 构建单智能体LLM应用（请使用ai-engineer）
- 优化单个智能体的提示词（请使用prompt-engineer）
- 管理智能体上下文窗口（请使用context-manager）
- 处理智能体故障与恢复（请使用error-coordinator）
Decision Framework
决策框架
Agent System Design:
├── Single task, no coordination → Single agent
├── Parallel independent tasks → Worker pool pattern
├── Sequential dependent tasks → Pipeline pattern
├── Complex interdependent tasks
│ ├── Clear hierarchy → Hierarchical orchestration
│ ├── Peer collaboration → Swarm/consensus pattern
│ └── Dynamic roles → Adaptive agent mesh
└── Human-in-the-loop → Supervisor patternAgent System Design:
├── Single task, no coordination → Single agent
├── Parallel independent tasks → Worker pool pattern
├── Sequential dependent tasks → Pipeline pattern
├── Complex interdependent tasks
│ ├── Clear hierarchy → Hierarchical orchestration
│ ├── Peer collaboration → Swarm/consensus pattern
│ └── Dynamic roles → Adaptive agent mesh
└── Human-in-the-loop → Supervisor patternCore Workflows
核心工作流
1. Agent Team Design
1. 智能体团队设计
- Decompose problem into agent responsibilities
- Define agent capabilities and interfaces
- Design communication topology (hub, mesh, hierarchy)
- Implement coordination protocol
- Add monitoring and observability
- Test failure scenarios
- 将问题拆解为各智能体的职责
- 定义智能体的能力与接口
- 设计通信拓扑结构（中心式、网状、层级式）
- 实现协调协议
- 添加监控与可观测性
- 测试故障场景
2. Agent Communication Setup
2. 智能体通信设置
- Choose message format (structured, natural language, hybrid)
- Define message routing strategy
- Implement handoff protocols
- Add retry and timeout handling
- Log all inter-agent messages
- 选择消息格式（结构化、自然语言、混合式）
- 定义消息路由策略
- 实现交接协议
- 添加重试与超时处理
- 记录所有智能体间的消息
3. Scaling Agent Systems
3. 智能体系统扩展
- Profile bottlenecks in current architecture
- Identify parallelization opportunities
- Implement load balancing across agents
- Add agent pooling for burst capacity
- Monitor resource utilization per agent
- 分析当前架构的瓶颈
- 识别并行化机会
- 在智能体间实现负载均衡
- 添加智能体池以应对突发流量
- 监控每个智能体的资源使用情况
Best Practices
最佳实践
- Keep agent responsibilities single-purpose and well-defined
- Use explicit handoff protocols between agents
- Implement circuit breakers for failing agents
- Log all inter-agent communication for debugging
- Design for graceful degradation when agents fail
- Version agent interfaces for backward compatibility
- 确保每个智能体的职责单一且定义明确
- 在智能体间使用明确的交接协议
- 为故障智能体实现断路器机制
- 记录所有智能体间的通信以便调试
- 设计智能体故障时的优雅降级方案
- 为智能体接口添加版本控制以保证向后兼容性
Anti-Patterns
反模式
| Anti-Pattern | Problem | Correct Approach |
|---|---|---|
| God agent | Single agent doing everything | Decompose into specialized agents |
| Chatty agents | Excessive inter-agent messages | Batch communications, async where possible |
| Tight coupling | Agents depend on internal state | Use contracts and interfaces |
| No supervision | Agents run without oversight | Add supervisor or human-in-loop |
| Shared mutable state | Race conditions and conflicts | Use message passing or event sourcing |
| 反模式 | 问题 | 正确做法 |
|---|---|---|
| 上帝智能体 | 单个智能体处理所有事务 | 拆分为多个专业智能体 |
| 话痨智能体 | 智能体间消息过多 | 批量通信，尽可能使用异步方式 |
| 紧耦合 | 智能体依赖内部状态 | 使用契约与接口 |
| 无监督 | 智能体无监督运行 | 添加监督者或人机协同模式 |
| 共享可变状态 | 出现竞态条件与冲突 | 使用消息传递或事件溯源 |