协作

如何在 CrewAI 团队中让 Agent 彼此协作、委派任务并高效沟通。

概览

CrewAI 中的协作机制使 Agent 能够像团队一样工作，通过委派任务和相互提问来利用彼此的专业能力。当设置 allow_delegation=True 时，Agent 会自动获得强大的协作工具。

快速开始：启用协作

  1. from crewai import Agent, Crew, Task
  3. # Enable collaboration for agents
  4. researcher = Agent(
  5.     role="Research Specialist",
  6.     goal="Conduct thorough research on any topic",
  7.     backstory="Expert researcher with access to various sources",
  8.     allow_delegation=True,  # 🔑 Key setting for collaboration
  9.     verbose=True
  10. )
  12. writer = Agent(
  13.     role="Content Writer", 
  14.     goal="Create engaging content based on research",
  15.     backstory="Skilled writer who transforms research into compelling content",
  16.     allow_delegation=True,  # 🔑 Enables asking questions to other agents
  17.     verbose=True
  18. )
  20. # Agents can now collaborate automatically
  21. crew = Crew(
  22.     agents=[researcher, writer],
  23.     tasks=[...],
  24.     verbose=True
  25. )

Agent 协作如何工作

当设置 allow_delegation=True 时，CrewAI 会自动为 Agent 提供两个强大的工具：

1. Delegate Work Tool

允许 Agent 将任务分配给具备特定专长的队友。

  1. # Agent automatically gets this tool:
  2. # Delegate work to coworker(task: str, context: str, coworker: str)

2. Ask Question Tool

允许 Agent 向同事提出具体问题，以获取所需信息。

  1. # Agent automatically gets this tool:
  2. # Ask question to coworker(question: str, context: str, coworker: str)

协作实战

下面是一个完整示例，展示 Agent 如何协作完成内容创作任务：

  1. from crewai import Agent, Crew, Task, Process
  3. # Create collaborative agents
  4. researcher = Agent(
  5.     role="Research Specialist",
  6.     goal="Find accurate, up-to-date information on any topic",
  7.     backstory="""You're a meticulous researcher with expertise in finding 
  8.     reliable sources and fact-checking information across various domains.""",
  9.     allow_delegation=True,
  10.     verbose=True
  11. )
  13. writer = Agent(
  14.     role="Content Writer",
  15.     goal="Create engaging, well-structured content",
  16.     backstory="""You're a skilled content writer who excels at transforming 
  17.     research into compelling, readable content for different audiences.""",
  18.     allow_delegation=True,
  19.     verbose=True
  20. )
  22. editor = Agent(
  23.     role="Content Editor",
  24.     goal="Ensure content quality and consistency",
  25.     backstory="""You're an experienced editor with an eye for detail, 
  26.     ensuring content meets high standards for clarity and accuracy.""",
  27.     allow_delegation=True,
  28.     verbose=True
  29. )
  31. # Create a task that encourages collaboration
  32. article_task = Task(
  33.     description="""Write a comprehensive 1000-word article about 'The Future of AI in Healthcare'.
  35.     The article should include:
  36.     - Current AI applications in healthcare
  37.     - Emerging trends and technologies  
  38.     - Potential challenges and ethical considerations
  39.     - Expert predictions for the next 5 years
  41.     Collaborate with your teammates to ensure accuracy and quality.""",
  42.     expected_output="A well-researched, engaging 1000-word article with proper structure and citations",
  43.     agent=writer  # Writer leads, but can delegate research to researcher
  44. )
  46. # Create collaborative crew
  47. crew = Crew(
  48.     agents=[researcher, writer, editor],
  49.     tasks=[article_task],
  50.     process=Process.sequential,
  51.     verbose=True
  52. )
  54. result = crew.kickoff()

协作模式

模式 1：研究 → 写作 → 编辑

  1. research_task = Task(
  2.     description="Research the latest developments in quantum computing",
  3.     expected_output="Comprehensive research summary with key findings and sources",
  4.     agent=researcher
  5. )
  7. writing_task = Task(
  8.     description="Write an article based on the research findings",
  9.     expected_output="Engaging 800-word article about quantum computing",
  10.     agent=writer,
  11.     context=[research_task]  # Gets research output as context
  12. )
  14. editing_task = Task(
  15.     description="Edit and polish the article for publication",
  16.     expected_output="Publication-ready article with improved clarity and flow",
  17.     agent=editor,
  18.     context=[writing_task]  # Gets article draft as context
  19. )

模式 2：单任务协作

  1. collaborative_task = Task(
  2.     description="""Create a marketing strategy for a new AI product.
  4.     Writer: Focus on messaging and content strategy
  5.     Researcher: Provide market analysis and competitor insights
  7.     Work together to create a comprehensive strategy.""",
  8.     expected_output="Complete marketing strategy with research backing",
  9.     agent=writer  # Lead agent, but can delegate to researcher
  10. )

分层协作

对于复杂项目，可以使用带有 manager Agent 的分层流程：

  1. from crewai import Agent, Crew, Task, Process
  3. # Manager agent coordinates the team
  4. manager = Agent(
  5.     role="Project Manager",
  6.     goal="Coordinate team efforts and ensure project success",
  7.     backstory="Experienced project manager skilled at delegation and quality control",
  8.     allow_delegation=True,
  9.     verbose=True
  10. )
  12. # Specialist agents
  13. researcher = Agent(
  14.     role="Researcher",
  15.     goal="Provide accurate research and analysis",
  16.     backstory="Expert researcher with deep analytical skills",
  17.     allow_delegation=False,  # Specialists focus on their expertise
  18.     verbose=True
  19. )
  21. writer = Agent(
  22.     role="Writer", 
  23.     goal="Create compelling content",
  24.     backstory="Skilled writer who creates engaging content",
  25.     allow_delegation=False,
  26.     verbose=True
  27. )
  29. # Manager-led task
  30. project_task = Task(
  31.     description="Create a comprehensive market analysis report with recommendations",
  32.     expected_output="Executive summary, detailed analysis, and strategic recommendations",
  33.     agent=manager  # Manager will delegate to specialists
  34. )
  36. # Hierarchical crew
  37. crew = Crew(
  38.     agents=[manager, researcher, writer],
  39.     tasks=[project_task],
  40.     process=Process.hierarchical,  # Manager coordinates everything
  41.     manager_llm="gpt-4o",  # Specify LLM for manager
  42.     verbose=True
  43. )

协作最佳实践

1. 清晰定义角色

  1. # ✅ Good: Specific, complementary roles
  2. researcher = Agent(role="Market Research Analyst", ...)
  3. writer = Agent(role="Technical Content Writer", ...)
  5. # ❌ Avoid: Overlapping or vague roles  
  6. agent1 = Agent(role="General Assistant", ...)
  7. agent2 = Agent(role="Helper", ...)

2. 有策略地启用委派

  1. # ✅ Enable delegation for coordinators and generalists
  2. lead_agent = Agent(
  3.     role="Content Lead",
  4.     allow_delegation=True,  # Can delegate to specialists
  5.     ...
  6. )
  8. # ✅ Disable for focused specialists (optional)
  9. specialist_agent = Agent(
  10.     role="Data Analyst", 
  11.     allow_delegation=False,  # Focuses on core expertise
  12.     ...
  13. )

3. 共享上下文

  1. # ✅ Use context parameter for task dependencies
  2. writing_task = Task(
  3.     description="Write article based on research",
  4.     agent=writer,
  5.     context=[research_task],  # Shares research results
  6.     ...
  7. )

4. 清晰的任务描述

  1. # ✅ Specific, actionable descriptions
  2. Task(
  3.     description="""Research competitors in the AI chatbot space.
  4.     Focus on: pricing models, key features, target markets.
  5.     Provide data in a structured format.""",
  6.     ...
  7. )
  9. # ❌ Vague descriptions that don't guide collaboration
  10. Task(description="Do some research about chatbots", ...)

协作问题排查

问题：Agent 没有发生协作

现象： Agent 各自独立工作，没有发生任务委派

  1. # ✅ Solution: Ensure delegation is enabled
  2. agent = Agent(
  3.     role="...",
  4.     allow_delegation=True,  # This is required!
  5.     ...
  6. )

问题：来回沟通过多

现象： Agent 提问过于频繁，导致进度缓慢

  1. # ✅ Solution: Provide better context and specific roles
  2. Task(
  3.     description="""Write a technical blog post about machine learning.
  5.     Context: Target audience is software developers with basic ML knowledge.
  6.     Length: 1200 words
  7.     Include: code examples, practical applications, best practices
  9.     If you need specific technical details, delegate research to the researcher.""",
  10.     ...
  11. )

问题：委派循环

现象： Agent 之间来回反复委派，无法结束

  1. # ✅ Solution: Clear hierarchy and responsibilities
  2. manager = Agent(role="Manager", allow_delegation=True)
  3. specialist1 = Agent(role="Specialist A", allow_delegation=False)  # No re-delegation
  4. specialist2 = Agent(role="Specialist B", allow_delegation=False)

高级协作功能

自定义协作规则

  1. # Set specific collaboration guidelines in agent backstory
  2. agent = Agent(
  3.     role="Senior Developer",
  4.     backstory="""You lead development projects and coordinate with team members.
  6.     Collaboration guidelines:
  7.     - Delegate research tasks to the Research Analyst
  8.     - Ask the Designer for UI/UX guidance  
  9.     - Consult the QA Engineer for testing strategies
  10.     - Only escalate blocking issues to the Project Manager""",
  11.     allow_delegation=True
  12. )

监控协作过程

  1. def track_collaboration(output):
  2.     """Track collaboration patterns"""
  3.     if "Delegate work to coworker" in output.raw:
  4.         print("🤝 Delegation occurred")
  5.     if "Ask question to coworker" in output.raw:
  6.         print("❓ Question asked")
  8. crew = Crew(
  9.     agents=[...],
  10.     tasks=[...],
  11.     step_callback=track_collaboration,  # Monitor collaboration
  12.     verbose=True
  13. )

记忆与学习

启用 Agent 记忆，让它们记住过去的协作过程：

  1. agent = Agent(
  2.     role="Content Lead",
  3.     memory=True,  # Remembers past interactions
  4.     allow_delegation=True,
  5.     verbose=True
  6. )

启用记忆后，Agent 可以从过去的协作中学习，并随着时间推移不断优化其委派决策。

下一步

  • 尝试这些示例：先从基础协作示例开始
  • 实验不同角色：测试不同的 Agent 角色组合
  • 观察交互过程：使用 verbose=True 查看协作如何发生
  • 优化任务描述：清晰的任务会带来更好的协作效果
  • 逐步扩展：对于复杂项目，尝试分层流程

协作能够把单个 AI Agent 转变为强大的团队，使它们可以共同应对复杂、多维度的挑战。