agent-team

Agent Team Collaboration Framework

Task Objectives

This Skill is used for: unified management and flexible combination of multiple agent roles to form task-oriented agent teams
Capabilities include: agent registration, dynamic team formation, collaboration workflows, role extension, shared memory, real-time communication
Trigger conditions: when multiple agents need to collaborate to complete a task, or when an agent combination for a specific scenario needs to be managed

Core Concept: A Busy AI Team Company

Agents Are Roles

Each agent is an independent role with:

Role positioning: clear responsibilities and areas of expertise
Capability characteristics: professional knowledge and personality traits
Collaboration interfaces: ways of collaborating with other agents
Output specifications: standardized output formats

Agent Team: A Busy One-Person AI Team Company

The team is composed of multiple agents, just like a real work team:

Clear division of roles

Working agents(execution layer):

Automation engineering agent: implements technical solutions, generates code, deploys systems
Frontend/backend engineers: write code, implement features
Data analysts: process data, generate reports
Operations experts: execute operational strategies, optimize processes

Commanding agents(management layer):

Moderator: guides the discussion process, controls the pace
Project manager: formulates plans, assigns tasks, tracks progress
Strategic analysis agent: formulates strategy, plans direction
Product manager: defines the product, plans the roadmap

Fault-finding agents(review layer):

Reviewer: evaluates quality, identifies problems, provides suggestions
Technical architect: reviews architecture, points out risks
Market analyst: reviews proposals, points out market issues
Financial advisor: reviews costs, points out financial risks

Real-time communication is visible

The discussion process between agents is fully visible:

✅ Display the speaking process in real time
✅ Record each agent's viewpoints
✅ Show debates and arguments
✅ Show consensus and disagreements
✅ Record the decision-making process

Just like in a real meeting room, you can see the entire process of each person speaking, discussing, and debating.

Shared Database Memory

All agents share the same database memory:

✅ Unified contextual information
✅ Shared project status
✅ Consistent decision history
✅ Unified knowledge base
✅ No incoherent statements

Ensure information consistency among agents and avoid contextual confusion.

Team Characteristics

Dynamic combination: select suitable agents according to task requirements
Flexible Collaboration: Agents can invoke and complement each other
Extensible: Add new agent roles at any time
Scenario-based: Preset team configurations for common scenarios
Real-time Communication: Fully displays the discussion process between agents
Shared Memory: Unified database memory to ensure context consistency

Agent Categories

1. Meeting Decision-making Agents

Suitable for scenarios that require multi-perspective analysis, debate, and decision-making.

Core Roles：

Technical Architect: System architecture, technology selection
DevOps Engineer: Deployment and operations, stability
Frontend/Backend Engineer: User experience/business logic
Product Manager: Product planning, user requirements
Market Analyst: Market research, competitive analysis
Financial Advisor: Cost control, financial analysis
Project Manager: Schedule management, resource coordination

See:references/meeting-agents.md

2. OPC System Building Agents

Suitable for the complete process of a one-person company from "doing things → making products → building systems".

Core Roles：

Strategic Analysis Agent: Market scanning, requirement analysis, opportunity identification
Product Architecture Agent: Product design, modular packaging, technology selection
Automation Engineering Agent: Process design, technical implementation, system integration

See:references/opc-agents.md

3. General Collaboration Agents

Suitable for cross-scenario general collaboration needs.

Core Roles：

Moderator: Guide the discussion process, summarize consensus
Recorder: Organize discussion content, generate meeting minutes
Reviewer: Evaluate solution quality, provide improvement suggestions
Coordinator: Resolve conflicts, promote collaboration

See:references/general-agents.md

Quick Start

Method 1: Use Preset Team Configurations

Choose a suitable preset team according to common task scenarios:

Scenario 1: Technical Architecture Decision Meeting

推荐团队：
- 指挥的：主持人、项目经理
- 挑毛病的：技术架构师、DevOps工程师、评审员
- 干活的：前端工程师、后端工程师

调用方式："请用技术决策团队评估[某技术方案]"
实时沟通：你会看到主持人引导讨论，各智能体发言、辩论、形成共识的全过程

Scenario 2: OPC System Building

推荐团队：
- 指挥的：战略分析智能体（战略）、产品架构智能体（设计）
- 干活的：自动化工程智能体（实现）
- 挑毛病的：评审员（评估）

调用方式："请用OPC团队构建[某领域]的系统"
实时沟通：你会看到三步流程中的每个环节，各智能体的输出和反馈

Scenario 3: Product Pricing Strategy Meeting

推荐团队：
- 指挥的：主持人、产品经理
- 挑毛病的：市场分析师、财务顾问、评审员
- 干活的：销售总监（市场执行）

调用方式："请用产品团队制定[某产品]的定价策略"
实时沟通：你会看到市场分析、成本核算、定价讨论的全过程

Scenario 4: Complete Project Execution

推荐团队：
- 指挥的：项目经理、战略分析智能体
- 挑毛病的：技术架构师、市场分析师、财务顾问、评审员
- 干活的：自动化工程智能体、前端工程师、后端工程师

调用方式："请执行[某项目]的完整流程"
实时沟通：你会看到从需求分析到技术实现的全过程，所有智能体的沟通和协作

See:references/collaboration-templates.md

Method 2: Custom Team Composition

According to the specific task, manually select the agents needed, just like building a real project team:

View Available Agents: Use references/agent-registry.md to browse all agents
Select Agents: Select agents according to task requirements
- Directing ones: Select 1-2 management-type agents
- Fault-finding ones: Select 1-3 review-type agents
- Doing-the-work ones: Select 1-3 execution-type agents
Define the Collaboration Process: Clarify the order and method of collaboration between agents
Execute the Task: Agents collaborate to complete the task according to the collaboration process
Observe in Real Time: View the real-time communication process between agents

Method 3: Real-time Communication Mode

Enable real-time communication mode to fully display the discussion process between agents:

【实时沟通模式】

> 主持人：各位，我们今天讨论的主题是"评估微服务架构的可行性"。请技术架构师先从技术角度发表观点。

> 技术架构师：从架构角度看，微服务架构可以提升系统的可扩展性和可维护性。但是也存在一些问题，比如分布式事务处理、服务间通信复杂度增加。我的建议是先评估业务是否真的需要这种架构。

> DevOps工程师：从运维角度，我担心的是部署复杂度和运维成本。每个服务都需要独立部署和监控，管理成本会显著增加。建议采用容器化部署来降低复杂度。

> 前端工程师：从用户体验角度，微服务架构对前端影响不大。但是要注意API的接口一致性，避免不同服务返回格式不统一的问题。

> 主持人：技术架构师提到了"先评估业务需求"，能否具体说明评估标准？

> 技术架构师：评估标准包括：1）业务规模是否需要横向扩展；2）团队规模是否能支撑多个服务的开发；3）是否有足够的运维能力。

> 评审员：基于以上讨论，我评估这个方案的可行性。技术上可行，但运维成本较高。建议先在小范围试点，验证可行性后再推广。

> 主持人：经过讨论，我们达成了以下共识：1）采用微服务架构；2）先试点核心服务；3）使用容器化部署；4）小范围验证后再推广。

> 记录员：我已记录下完整的讨论过程和决策内容。

Enable real-time communication mode, and you will see the complete discussion process, just like in a real meeting room!

Agent Collaboration Process

Standard Collaboration Modes

Serial Collaboration: Agents process sequentially in order

智能体A → 输出 → 智能体B → 输出 → 智能体C → 最终结果

Applicable to: OPC system construction, phased decision-making

Parallel Collaboration: Multiple agents process simultaneously, then aggregate

智能体A ──┐
           ├→ 汇总整合 → 最终结果
智能体B ──┘

Applicable to: Multi-perspective analysis, competitive comparison

Loop Collaboration: Agents repeatedly discuss and iterate

智能体A ⟷ 智能体B ⟷ 智能体C
        ↓
   收敛到共识

Applicable to: Meeting decision-making, solution optimization

Hybrid Collaboration: Combines the above multiple modes

[智能体A、B并行] → 汇总 → 智能C → 智体D循环讨论 → 最终结果

Applicable to: Complex tasks, multi-stage processes

Collaboration Principles

Clear role boundaries: Each agent has a clearly defined scope of responsibility
Standardized output format: Agents use a unified data format
Clear collaboration protocol: Defines how agents invoke and respond
Feedback loop mechanism: Supports mutual feedback and iteration among agents
Fallback handling plan: When an agent fails, there is a backup plan
Shared database memory: All agents share the same database, ensuring context consistency

Shared Database Memory

Memory Mechanism

All agents share the same database memory, just like a real team's shared knowledge base:

Unified context information：

{
  "project_context": {
    "project_name": "AI脚本生成器",
    "current_stage": "技术架构设计",
    "participants": ["技术架构师", "DevOps工程师", "前端工程师"],
    "decisions_made": [
      "采用微服务架构",
      "使用Vue 3 + FastAPI技术栈"
    ]
  }
}

Shared knowledge base：

{
  "knowledge_base": {
    "technical_standards": {
      "api_format": "RESTful",
      "database": "PostgreSQL",
      "deployment": "Docker"
    },
    "business_goals": {
      "primary_goal": "1分钟生成3个脚本",
      "secondary_goals": ["支持多平台", "降低成本"]
    }
  }
}

Decision history：

{
  "decision_history": [
    {
      "timestamp": "2024-01-01T10:00:00Z",
      "decision": "采用微服务架构",
      "reason": "提升可扩展性",
      "participants": ["技术架构师", "DevOps工程师"],
      "voting": {"support": 3, "oppose": 1, "abstain": 0}
    }
  ]
}

Advantages of Memory

Context consistency: All agents see the same information, and there will be no incoherent statements
Avoid repetition: Content already discussed does not need to be discussed again
Easy to track: You can view the decision history and discussion process
Supports backtracking: You can view previous knowledge and decisions at any time
Team collaboration: All agents work as if in the same office

Memory Access Example

Agents can access shared memory at any time to ensure communication continuity:

> 技术架构师：基于之前的讨论（访问共享记忆），我们已经决定采用微服务架构。现在我来详细设计服务拆分方案。

> 前端工程师：我记得之前的决策（访问共享记忆）是要使用Vue 3。那我来设计前端组件结构。

> 评审员：查看决策历史（访问共享记忆），之前的技术选型是否还有优化空间？

> 主持人：基于当前的项目状态（访问共享记忆），我们正在进行技术架构设计阶段，接下来进入产品架构设计阶段。

Extending with New Agents

Steps to Add a New Role

Define the role: Clarify the agent's positioning, responsibilities, and capabilities
Write the template: Use assets/agent-templates/ templates in
Register the agent: In references/agent-registry.md add
Design the collaboration interface: Define how it collaborates with other agents
Test and verify: Verify whether the agent can collaborate normally

See details:assets/agent-templates/new-agent-template.md

Resource Index

Agent Registry

references/agent-registry.md
- Complete list of all available agents
- Categories, positioning, and capability descriptions of agents
- When to read: When you need to select agents to form a team

Detailed Agent Definitions

references/meeting-agents.md
- Detailed definitions of meeting decision-making agents
- When to read: When a meeting decision-making scenario is needed
references/opc-agents.md
- Detailed definition of OPC system construction agents
- When to read: when an OPC system needs to be constructed
references/general-agents.md
- Detailed definition of general collaboration agents
- When to read: when general collaboration capabilities are needed

Collaboration Templates

references/collaboration-templates.md
- Preset team configurations for common scenarios
- Collaboration process design templates
- When to read: when a collaboration scenario needs to be started quickly

Agent Templates

assets/agent-templates/new-agent-template.md
- Standard template for adding a new agent
- When to use: when a new agent role needs to be extended

Operating Steps

Standard Process

Task Analysis
- Clarify task objectives and constraints
- Identify the professional domains and perspectives required by the task
- Assess the complexity of the task and its phase divisions
Team Formation
- Consult references/agent-registry.md
- Select appropriate agent roles (3-6 is preferable)
- Or use a preset team configuration
Collaboration Design
- Define the collaboration mode among agents (serial/parallel/cyclic/hybrid)
- Design the collaboration process and output format
- Determine the consensus convergence method
Task Execution
- Agents collaborate to complete the task according to the collaboration process
- Record intermediate outputs and the discussion process
- Support iteration and adjustment
Result Output
- Summarize the outputs of each agent
- Generate a unified task result
- Record the collaboration process and key decisions

Optional Branches

When the task is simple: directly use a single agent to complete the task
When time is urgent: use a preset team configuration to start quickly
When in-depth discussion is needed: adopt a cyclic collaboration mode with multiple rounds of iteration
When an agent fails: use a fallback plan, or replace it with a backup agent

Usage Examples

Example 1: Complete OPC System Construction

Task: "I want to build an OPC system in the AI short video field"

Team configuration: Strategic analysis agent + Product architecture agent + Automation engineering agent

Collaboration process：

1. 战略分析智能体
   - 分析AI短视频领域
   - 输出：TOP3产品化机会

2. 产品架构智能体（基于战略分析的输出）
   - 针对优先级最高的机会设计产品架构
   - 输出：模块化产品蓝图

3. 自动化工程智能体（基于产品架构的输出）
   - 实现核心模块
   - 输出：可运行代码和部署指南

Final deliverable: Complete technical solution and MVP code

Example 2: Technical Architecture Decision Meeting

Task: "Evaluate whether to adopt a microservices architecture"

Team configuration: Technical architect + DevOps engineer + Frontend engineer + Backend engineer + CTO

Collaboration process：

1. 开场发言：各智能体从各自专业角度陈述观点
2. 自由讨论：智能体相互质疑、补充、辩论
3. 深入辩论：针对争议焦点展开针对性讨论
4. 共识收敛：总结各方观点，探索折中方案
5. 决策生成：综合各方意见形成决策结论

Final deliverable: Complete meeting record including technical debate, cost analysis, and risk assessment

Example 3: Hybrid Collaboration Scenario

Task: "Evaluate the feasibility of an AI short video script generation system"

Team configuration: Strategic analysis agent + Technical architect + Product manager + Market analyst

Collaboration process：

1. 并行分析（战略分析智能体 + 市场分析师）
   - 战略分析：识别市场机会
   - 市场分析：竞品和用户需求

2. 汇总输入（产品经理）
   - 整合市场分析结果
   - 定义产品需求

3. 技术评估（技术架构师）
   - 评估技术可行性
   - 输出技术方案

4. 循环讨论（所有智能体）
   - 技术架构师质疑市场需求的合理性
   - 产品经理补充产品细节
   - 市场分析师反馈用户预期
   - 收敛到共识

Final deliverable: Complete feasibility report including market analysis, product design, and technical solution

Precautions

Moderate number of agents: 3-6 agents is appropriate; too many will lead to complex collaboration
Clear role boundaries: Each agent clearly defines its own responsibilities to avoid overlapping duties
Unified output format: Agents use a unified data format to facilitate information transfer
Support degradation handling: When an agent fails, have a backup plan to ensure the task continues
Iterative optimization: Based on collaboration results, continuously optimize agent definitions and collaboration processes
Fully utilize agent capabilities: Agents have analysis, creation, and reasoning capabilities; there is no need to write scripts for simple tasks

Applicable scenarios

Best suited for：

Complex tasks requiring collaboration across multiple professional domains
Scenarios requiring multi-perspective analysis and decision-making
Requires systematic construction of OPC scenarios
Requires flexibly forming teams to handle different tasks

Not very suitable for：

Tasks that can be completed by a simple single agent
Scenarios requiring real-time interaction and immediate response
Unstructured tasks that rely entirely on human judgment

About this skill

Task Objectives

Core Concept: A Busy AI Team Company

Agents Are Roles

Agent Team: A Busy One-Person AI Team Company

Clear division of roles

Real-time communication is visible

Shared Database Memory

Team Characteristics

Agent Categories

1. Meeting Decision-making Agents

2. OPC System Building Agents

3. General Collaboration Agents

Quick Start

Method 1: Use Preset Team Configurations

Method 2: Custom Team Composition

Method 3: Real-time Communication Mode

Agent Collaboration Process

Standard Collaboration Modes

Collaboration Principles

Shared Database Memory

Memory Mechanism

Advantages of Memory

Memory Access Example

Extending with New Agents

Steps to Add a New Role

Resource Index

Agent Registry

Detailed Agent Definitions

Collaboration Templates

Agent Templates

Operating Steps

Standard Process

Optional Branches

Usage Examples

Example 1: Complete OPC System Construction

Example 2: Technical Architecture Decision Meeting

Example 3: Hybrid Collaboration Scenario

Precautions

Applicable scenarios