debate

debate runs multiple agents arguing different positions, then has a judge synthesize a final answer from all perspectives.

When to use it

• Decisions requiring consideration of opposing viewpoints
• Policy or strategy analysis (pros/cons, risks/benefits)
• Research where multiple interpretations exist
• Any task where groupthink is a risk

AGENT.md

---
name: investment-debate
description: Debates an investment decision from bull and bear perspectives.
version: "1.0.0"
pattern: debate
agents:
  - bull-analyst
  - bear-analyst
rounds: 1
judge: synthesis-analyst
call:
  model:
    role: thinker
---

No body needed — debate delegates to the listed agents.

The position agents

Each agent in agents: is a full named agent. They receive the original input and argue their assigned position:

# agents/bull-analyst/AGENT.md
---
name: bull-analyst
description: Argues the bullish case for an investment.
version: "1.0.0"
pattern: react
call:
  model:
    role: thinker
  uses:
    - search-web
---

You are a bullish investment analyst. Given an investment opportunity:
1. Research positive signals and growth catalysts
2. Build the strongest possible case FOR the investment
3. Acknowledge risks only briefly

Call finish with key="done" and your bullish thesis as value.

The judge agent

# agents/synthesis-analyst/AGENT.md
---
name: synthesis-analyst
description: Synthesizes multiple investment analyses into a recommendation.
version: "1.0.0"
pattern: oneshot
call:
  model:
    role: thinker
---

You receive arguments from multiple analysts with different perspectives.

Synthesize all viewpoints into a balanced, well-reasoned final recommendation.
Weigh the arguments, acknowledge the strongest points on each side, and provide a clear conclusion.

Scaffold

tama add debate my-agent

How it works

Input → all position agents run (concurrently if rounds=1)
  bull-analyst   → bullish thesis
  bear-analyst   → bearish thesis
          ↓
judge receives all positions
          ↓
judge synthesizes → final recommendation

With rounds > 1, agents can see each other’s previous arguments and respond:

Round 1: all agents argue independently
Round 2: each agent responds to the others' round 1 arguments
  ...
Round N: final positions
Judge: synthesize

Configuration

Field	Description
agents	List of agent names to include in the debate
rounds	Number of debate rounds (default: 1)
judge	Agent name for the synthesis step

Two vs many agents

The minimum is two agents (a classic pro/con structure), but you can add more:

agents:
  - optimist
  - pessimist
  - realist
  - contrarian
rounds: 2
judge: moderator