Debate Agents Example

This example demonstrates how to create a sophisticated debate system with multiple agents playing different roles. It showcases dynamic agent interactions, role-based responses, and structured debate flow.

Overview

The Debate Agents example shows how to: - Create multiple specialized agents for different debate roles - Implement dynamic agent interactions - Manage debate flow and turns - Handle structured debate content - Integrate judging and evaluation

Implementation

Here's the complete implementation:

from datetime import datetime
from yosrai import Agent, Action, ContextFunctions
import agents_debate_prompts

# Selection Agent - Chooses debate participants
selection_agent = Agent(
    agent_code='selection_agent',
    agent_name='Selection Agent',
    prompt_template=agents_debate_prompts.SELECTION_PROMPT,
    outputs={'players_json': ContextFunctions.DEFAULT, 'players': ContextFunctions.JSON}
)

# Greeting Agent - Introduces the debate and participants
greeting_agent = Agent(
    agent_code='greeting_agent',
    agent_name='Greeting Agent',
    agent_type='tool',
    prompt_template=agents_debate_prompts.GREETING_PROMPT,
    outputs={'greeting': ContextFunctions.DEFAULT}
)

# Player One Agent - First debate participant
player_one_agent = Agent(
    agent_code='player_one_agent',
    agent_name='Player One Agent',
    prompt_template=agents_debate_prompts.PLAYER_ONE_PROMPT,
    outputs={
        'history': ContextFunctions.REPLIES, 
        'current_response': ContextFunctions.DEFAULT, 
        'messages': ContextFunctions.REPLIES
    }
)

# Player Two Agent - Second debate participant
player_two_agent = Agent(
    agent_code='player_two_agent',
    agent_name='Player Two Agent',
    prompt_template=agents_debate_prompts.PLAYER_TWO_PROMPT,
    outputs={
        'history': ContextFunctions.REPLIES, 
        'current_response': ContextFunctions.DEFAULT, 
        'messages': ContextFunctions.REPLIES
    }
)

# Next Player Agent - Determines who speaks next
next_player_agent = Agent(
    agent_code='next_player_agent',
    agent_name='Next Player Agent',
    agent_type='tool',
    prompt_template=agents_debate_prompts.NEXT_PLAYER_PROMPT,
    outputs={'next_player': ContextFunctions.DEFAULT}
)

# Judge Agent - Evaluates the debate
judge_agent = Agent(
    agent_code='judge_agent',
    agent_name='Judge Agent',
    prompt_template=agents_debate_prompts.JUDGE_PROMPT,
    outputs={'jury_response': ContextFunctions.DEFAULT, 'messages': ContextFunctions.REPLIES}
)

# Set up the workflow
action = Action('Debate Between Two Players')
action.Context(
    query='',
    greeting='',
    players={'player_one': '', 'player_two': ''},
    history=[],
    no_of_rounds=1,
    jury_response='',
    date=datetime.now().strftime('%d-%m-%Y %H:%M:%S')
)

# Add agents to the action
action.add_agent(selection_agent)
action.add_agent(greeting_agent)
action.add_agent(player_one_agent)
action.add_agent(player_two_agent)
action.add_agent(next_player_agent)
action.add_agent(judge_agent)

# Define the workflow
action.add_link(selection_agent, greeting_agent)
action.add_link(greeting_agent, player_one_agent)
action.add_link_dynamic(player_one_agent, [player_two_agent, judge_agent], next_player_agent)
action.add_link_dynamic(player_two_agent, [player_one_agent, judge_agent], next_player_agent)
action.add_link(judge_agent, "END")

Workflow Visualization

The workflow shows the complex interaction between debate participants:

---
title: Two Players Debate
---
graph LR
START((Start)) --> selection_agent[Selection Agent]
selection_agent(Selection Agent) --> greeting_agent(Greeting Agent)
greeting_agent(Greeting Agent) --> player_one_agent(Player One Agent)
player_one_agent(Player One Agent) --> player_two_agent(Player Two Agent)
player_one_agent(Player One Agent) --> judge_agent(Judge Agent)
player_two_agent(Player Two Agent) --> player_one_agent(Player One Agent)
player_two_agent(Player Two Agent) --> judge_agent(Judge Agent)
judge_agent(Judge Agent) --> END((End))

Key Components

Selection Agent

Responsible for choosing debate participants:

selection_agent = Agent(
    agent_code='selection_agent',
    agent_name='Selection Agent',
    prompt_template=agents_debate_prompts.SELECTION_PROMPT,
    outputs={'players_json': ContextFunctions.DEFAULT, 'players': ContextFunctions.JSON}
)

Greeting Agent

Introduces the debate and sets the stage:

greeting_agent = Agent(
    agent_code='greeting_agent',
    agent_name='Greeting Agent',
    agent_type='tool',
    prompt_template=agents_debate_prompts.GREETING_PROMPT,
    outputs={'greeting': ContextFunctions.DEFAULT}
)

Player Agents

Represent debate participants with structured responses:

player_one_agent = Agent(
    agent_code='player_one_agent',
    agent_name='Player One Agent',
    prompt_template=agents_debate_prompts.PLAYER_ONE_PROMPT,
    outputs={
        'history': ContextFunctions.REPLIES, 
        'current_response': ContextFunctions.DEFAULT, 
        'messages': ContextFunctions.REPLIES
    }
)

Next Player Agent

Controls debate flow and turn management:

next_player_agent = Agent(
    agent_code='next_player_agent',
    agent_name='Next Player Agent',
    agent_type='tool',
    prompt_template=agents_debate_prompts.NEXT_PLAYER_PROMPT,
    outputs={'next_player': ContextFunctions.DEFAULT}
)

Judge Agent

Evaluates debate performance and determines outcome:

judge_agent = Agent(
    agent_code='judge_agent',
    agent_name='Judge Agent',
    prompt_template=agents_debate_prompts.JUDGE_PROMPT,
    outputs={'jury_response': ContextFunctions.DEFAULT, 'messages': ContextFunctions.REPLIES}
)

Example Interaction

Here's a sample debate interaction:

Selection Agent: "Selected participants: Data Analyst John vs Sports Strategist Sarah"

Greeting Agent:

Hello! Today 06-01-2025 21:00:16, we will witness a debate between
- Data Analyst John
- Sports Strategist Sarah

The topic of the debate is:
ManU vs Liverpool

Let's start the debate 🏁

Player One: "As a data analyst, I believe Manchester United has better historical performance metrics..."

Player Two: "From a strategic perspective, Liverpool's current squad depth and tactical flexibility..."

Judge: "After analyzing both arguments, Sarah's strategic analysis was more comprehensive..."

Best Practices

When implementing the Debate Agents pattern:

Clear Roles: Define distinct responsibilities for each agent
Turn Management: Implement robust turn-taking mechanisms
Context Preservation: Maintain debate history and context
Evaluation Criteria: Define clear judging criteria
Dynamic Flow: Handle debate flow variations effectively

Use Cases

This pattern is ideal for: - Educational debate platforms - Decision-making systems - Argument analysis tools - Training simulations - Multi-perspective analysis

Next Steps

After implementing the Debate Agents pattern, you can: 1. Add more debate formats 2. Implement audience participation 3. Add real-time fact-checking 4. Enhance judging criteria 5. Add debate analytics