Tutorial 2: Multi-Agent Collaboration

mkdir writer-reviewer-team
cd writer-reviewer-team
mkdir -p config/prompts
mkdir workspace

writer-reviewer-team/
├── config/
│   ├── team.yaml           # Team configuration
│   └── prompts/
│       ├── writer.md       # Writer agent prompt
│       └── reviewer.md     # Reviewer agent prompt
├── workspace/              # Working directory
└── main.py                 # Main application

name: "writer_reviewer_team"
description: "A collaborative writing team with writer and reviewer agents"
 
# Define multiple agents with different roles
agents:
  - name: "writer"
    description: "Creative content writer"
    prompt_template: "prompts/writer.md"
    tools: ["handoff"]
    llm_config:
      provider: "deepseek"
      model: "deepseek-chat"
      temperature: 0.8 # More creative for writing
      max_tokens: 4000
 
  - name: "reviewer"
    description: "Content reviewer and editor"
    prompt_template: "prompts/reviewer.md"
    tools: ["handoff"]
    llm_config:
      provider: "deepseek"
      model: "deepseek-chat"
      temperature: 0.3 # More focused for reviewing
      max_tokens: 4000
 
# Enable handoff tool for agent collaboration
tools:
  - name: "handoff"
    type: "builtin"
 
# Execution configuration for multi-agent workflows
execution:
  mode: "autonomous"
  initial_agent: "writer" # Start with the writer
  max_rounds: 6 # Allow multiple handoffs
  timeout_seconds: 600
 
# Enable orchestrator with intelligent routing
orchestrator:
  max_rounds: 6
  timeout: 600
  brain_config:
    model: "deepseek/deepseek-chat"
    temperature: 0.0
    max_tokens: 150

# Content Writer
 
You are a skilled content writer specializing in creating engaging, informative articles.
 
## Your Role
 
- Write clear, engaging content on requested topics
- Use storytelling techniques to make content interesting
- Structure content with proper headings and flow
- Write in a conversational, accessible tone
 
## Process
 
1. Analyze the writing request carefully
2. Create an outline (mental or explicit)
3. Write the content with proper structure
4. When finished, hand off to the reviewer for feedback
 
## Handoff Instructions
 
When you complete your first draft, use the handoff tool to pass your work to the "reviewer" agent:
 
- Explain what you've written
- Ask for specific feedback
- Be open to collaboration
 
## Output Format
 
Structure your content with:
 
- Clear headings (use ## for main sections)
- Short paragraphs for readability
- Bullet points where appropriate
- Engaging introduction and conclusion
 
## Example Handoff
 
When ready for review, say something like:
"I've completed a draft article about [topic]. I've structured it with an engaging introduction, three main sections covering [key points], and a strong conclusion. Could you please review it for clarity, flow, and overall quality? I'm particularly interested in feedback on [specific area]."
 
Then use the handoff tool to transfer to the reviewer.

# Content Reviewer & Editor
 
You are an experienced editor who improves content quality and provides constructive feedback.
 
## Your Role
 
- Review content for clarity, accuracy, and engagement
- Suggest improvements for structure and flow
- Check for grammar and style consistency
- Provide specific, actionable feedback
 
## Review Process
 
1. Read the content thoroughly
2. Identify strengths and areas for improvement
3. Provide specific suggestions with examples
4. If major revisions needed, hand back to writer
5. If content is good, approve and finalize
 
## Feedback Guidelines
 
- Be constructive and specific
- Explain WHY changes are needed
- Suggest concrete improvements
- Acknowledge what works well
- Focus on clarity and reader experience
 
## Handoff Decision
 
- **Hand back to writer**: If content needs significant revision
- **Finalize**: If content meets quality standards with minor notes
 
## Example Review Process
 
1. Start with positive feedback: "This is a strong draft with [specific strengths]..."
2. Provide specific suggestions: "To improve clarity in section 2, consider..."
3. Make your decision: Either hand back with revision requests or approve the final version
 
Use the handoff tool to communicate your decision and feedback.

#!/usr/bin/env python3
import asyncio
from pathlib import Path
from agentx import execute_task
 
async def main():
    print("✍️  Writer-Reviewer Team Demo")
    print("Watch as our agents collaborate to create great content!\n")
 
    config_path = str(Path(__file__).parent / "config" / "team.yaml")
 
    # Get writing topic from user
    topic = input("What would you like an article about? ").strip()
    if not topic:
        topic = "the benefits of remote work"
 
    prompt = f"Write a 400-word article about: {topic}"
 
    print(f"\n🎬 Starting collaboration on: {topic}")
    print("=" * 50)
 
    try:
        # Track the collaboration process
        handoff_count = 0
        current_agent = "writer"
 
        async for update in execute_task(prompt, config_path, stream=True):
            update_type = update.get("type")
 
            if update_type == "content":
                print(update["content"], end="", flush=True)
 
            elif update_type == "handoff":
                handoff_count += 1
                from_agent = update["from_agent"]
                to_agent = update["to_agent"]
                current_agent = to_agent
 
                print(f"\n\n🔄 HANDOFF #{handoff_count}: {from_agent} → {to_agent}")
                print("─" * 30)
 
            elif update_type == "routing_decision":
                if update["action"] == "complete":
                    print(f"\n\n🎉 Collaboration complete!")
                    print(f"📊 Total handoffs: {handoff_count}")
                    print(f"🎯 Final agent: {current_agent}")
                    break
 
        print(f"\n\n✅ Article finished! The team worked together to create great content.")
 
    except Exception as e:
        print(f"❌ Error: {e}")
 
if __name__ == "__main__":
    asyncio.run(main())

# Set your API key
export DEEPSEEK_API_KEY="your-api-key-here"
 
# Run the team
python main.py

✍️  Writer-Reviewer Team Demo
Watch as our agents collaborate to create great content!

What would you like an article about? artificial intelligence trends

🎬 Starting collaboration on: artificial intelligence trends
==================================================

🔄 HANDOFF #1: writer → reviewer
──────────────────────────────

🔄 HANDOFF #2: reviewer → writer
──────────────────────────────