Overview
The debate workflow creates structured arguments between multiple AI agents to explore different perspectives on a topic. A judge moderates and evaluates arguments.
Quick Start
# Spawn fresh debaters with pre-assigned positions
hcom run debate "AI will replace programmers" --spawn
# Use existing agents (they pick their own positions)
hcom run debate "tabs vs spaces" -w agent1,agent2,agent3
# Custom rounds and context
hcom run debate "microservices vs monolith" --spawn --rounds 4 \
--context "E-commerce platform with 50 engineers"
Modes
Launches fresh PRO and CON debaters with debate-specific system prompts.hcom run debate "topic" --spawn
- PRO debater (argues in favor)
- CON debater (argues against)
- Judge (moderates and evaluates)
Best for:
- Clean slate debates
- Controlled positions (PRO/CON)
- Academic/formal arguments
Uses existing agents who dynamically choose their positions.hcom run debate "topic" -w luna,nova,kira
- Using agents with domain knowledge
- More than 2 debaters
- Organic position selection
Debate Structure
Setup
Judge announces topic and rules. Debaters confirm ready.
Opening Statements
PRO presents argument first, then CON responds.
Rebuttals
Multiple rounds of back-and-forth arguments (default: 2 rounds).
Evaluation
Judge scores both sides and declares winner with reasoning.
Options
hcom run debate [OPTIONS] TOPIC
| Option | Description | Default |
|---|
-s, --spawn | Spawn fresh PRO/CON instances | - |
-w, --workers NAMES | Use existing instances (comma-separated) | - |
--tool TOOL | AI tool for spawned agents | claude |
-r, --rounds N | Number of rebuttal rounds | 2 |
-t, --timeout N | Response timeout in seconds | 120 |
-c, --context TEXT | Additional context for debaters | - |
-i, --interactive | Launch in terminal windows | false |
--name NAME | Your identity | - |
Examples
Simple Debate
hcom run debate "Remote work is better than office work" --spawn
- PRO debater launches with system prompt favoring remote work
- CON debater launches opposing remote work
- Judge coordinates debate rounds
- All messages go to shared thread for visibility
Extended Debate
hcom run debate "Should we migrate to microservices?" --spawn --rounds 5 \
--context "Current: Django monolith, 20 engineers, 5 year old codebase"
- Ground arguments in reality
- Reference specific constraints
- Make practical recommendations
Multi-Agent Debate
hcom run debate "Best programming language for startups" \
-w alice,bob,charlie,diana
- Agents pick different languages/positions
- More diverse perspectives
- Judge has harder evaluation task
Interactive Debate
hcom run debate "Vim vs Emacs" --spawn -i
-i flag:
- Debaters launch in terminal windows (not headless)
- You can watch live conversation
- Useful for debugging or learning
Watching Debates
Follow in real-time
# Get thread name from script output
hcom events --wait 600 --sql "msg_thread='debate-1234567890'"
Use TUI
hcom
# Navigate to Messages tab
# Filter by thread: debate-1234567890
Query after completion
# Get thread from script output, then:
hcom events --thread debate-1234567890 --type message
Custom Debate Script
For more control, create your own debate script:
# ~/.hcom/scripts/my-debate.sh
#!/usr/bin/env bash
set -euo pipefail
topic="$1"
thread="debate-$(date +%s)"
# Launch 3 debaters with different expertise
hcom 1 claude --tag tech-lead --go --headless \
--hcom-system-prompt "You are a tech lead. Argue from engineering perspective."
hcom 1 claude --tag product --go --headless \
--hcom-system-prompt "You are a product manager. Argue from user/business perspective."
hcom 1 claude --tag designer --go --headless \
--hcom-system-prompt "You are a designer. Argue from UX/design perspective."
# Custom judge
hcom 1 claude --tag judge --go --hcom-prompt "
Topic: ${topic}
Thread: ${thread}
1. Announce topic to debaters:
hcom send '@tech-lead @product @designer Topic: ${topic}. \
State your position and opening argument.' --thread ${thread}
2. Collect 3 opening arguments (wait 180s):
hcom events --wait 180 --thread ${thread} --type message
3. Two rounds of rebuttals:
- Prompt each to respond to others
- Collect responses
4. Final judgment:
- Score each perspective: technical merit, user impact, feasibility
- Synthesize best path forward
- Send verdict to @team
"
echo "Debate started: ${thread}"
Advanced: Contextual Debates
Provide shared files/context for grounded debates:
# Create context bundle
hcom bundle create "Database migration context" \
--description "Current schema, performance metrics, migration plan" \
--files schema.sql,perf-report.csv,migration-plan.md
# Send to debaters before debate
hcom send '@pro- @con- Context for debate attached. Review before arguing.' \
--title "Database migration context" \
--description "Schema, metrics, and plan" \
--files schema.sql,perf-report.csv,migration-plan.md
# Then start debate
hcom run debate "Should we migrate to PostgreSQL?" --spawn
Judging Criteria
The judge evaluates on:
- Logical coherence: Do arguments follow logically?
- Evidence quality: Are claims backed by data/examples?
- Persuasiveness: How compelling is the rhetoric?
- Responsiveness: Do they address opponent’s points?
- Structure: Are arguments organized clearly?
Tips
Use --context to ground debates in real scenarios instead of abstract theory.
Increase --rounds for complex topics that need deeper exploration (4-6 rounds).
For technical debates, use workers mode with agents that have loaded relevant code.
Don’t set timeout too short. Complex arguments need time (120-180s per response).
For/Against Decision
hcom run debate "Should we rewrite in Rust?" --spawn \
--context "Current: Python, 100k LOC, performance issues"
Best Option (3+ agents)
hcom run debate "Choose frontend framework" \
-w react-expert,vue-expert,svelte-expert
Pros/Cons Exploration
hcom run debate "Evaluate serverless architecture" --spawn --rounds 4 \
--context "Current: Docker on EC2, considering Lambda migration"
Troubleshooting
Debaters not responding
Check if agents are idle:
hcom list | grep -E 'pro-|con-|judge-'
hcom events --agent pro- --last 5
Judge skipped rounds
Query thread to see what happened:
hcom events --thread debate-1234567890 --type message --last 20
hcom send '@judge- Continue with round 2' --thread debate-1234567890
Debate timeout
Increase timeout for complex topics:
hcom run debate "topic" --spawn --timeout 240 # 4 minutes per response
No clear winner
Judge should declare TIE with reasoning:
VERDICT: TIE
Both sides presented strong arguments:
- PRO excelled at [aspects]
- CON excelled at [aspects]
The decision depends on [factors]
Recommendation: [synthesis]
Example Output
Typical debate flow:
Thread: debate-1709564222
Topic: Remote work is better than office work
Tool: claude
Mode: Spawning fresh PRO/CON debaters
Rounds: 2
PRO debater launched (claude)
CON debater launched (claude)
Judge launched (claude)
Watch debate:
hcom events --wait 600 --sql "msg_thread='debate-1709564222'"
Or in TUI:
hcom
hcom events --thread debate-1709564222 --type message
#42 14:30:22 message judge-luna "PRO: Present your opening argument..."
#43 14:31:05 message pro-nova "Remote work offers flexibility..."
#44 14:31:47 message judge-luna "CON: Present your argument..."
#45 14:32:30 message con-kira "Office work enables collaboration..."
#46 14:33:15 message judge-luna "Round 1 rebuttal..."
...
#52 14:38:42 message judge-luna "VERDICT: PRO wins 7/10 vs 6/10..."