debate-chamber

Files (3)

• README.md
• SKILL.md
• hero.png · assets/

SKILL.md (first lines)

# Debate Chamber

## Overview

**Model-Chat Pattern:** Spawn 5 agents into a shared conversation room where they debate across 3 rounds, each round building on the previous. Agents see each other's prior responses and are instructed to find flaws, challenge assumptions, and synthesize stronger positions. A Team Lead agent then converges the full debate transcript into a single unified implementation plan.

Unlike consensus-brainstormer (parallel, independent agents), the debate chamber is **sequential and iterative** — ideas sharpen through friction.

## When to Use

- Problems where the first answer is rarely the best answer
- Implementation planning where trade-offs need to be argued through
- When you want ideas to collide and improve, not just be collected
- Architecture decisions, technical design, strategy refinement
- When consensus-brainstormer gives you a MODE but you want to stress-test it further

**When NOT to use:** Simple questions, time-sensitive decisions (3 rounds takes longer), factual lookups, tasks where independent perspectives are more valuable than iterative debate.

**Comparison with consensus-brainstormer:**

| Dimension         | Consensus Brainstormer              | Debate Chamber                          |
| ----------------- | ----------------------------------- | --------------------------------------- |
| Agent interaction | Independent, no cross-talk          | Iterative, agents respond to each other |
| Rounds            | 1 (parallel)                        | 3 sequential + convergence              |
| Strength          | Breadth of ideas, outlier discovery | Depth of analysis, idea refinement      |
| Speed             | Faster (1 parallel dispatch)        | Slower (3 sequential rounds)            |
| Best for          | "What are our options?"             | "Which option is best and why?"         |

## The 5 Debaters

Each agent has a distinct argumentative role designed to create productive tension:

| #   | Role              | Mandate                                                             | Style                       |
| --- | ----------------- | ------------------------------------------------------------------- | --------------------------- |
| 1   | **Architect**     | Big-picture design, system coherence, long-term implications        | Structured, principled      |
| 2   | **Skeptic**       | Find flaws, challenge assumptions, probe weak points                | Adversarial, rigorous       |
| 3   | **Pragmatist**    | Implementation feasibility, timelines, resource constraints         | Grounded, practical         |
| 4   | **Innovator**     | Creative alternatives, unconventional approaches, what-if scenarios | Bold, exploratory           |
| 5   | **User Champion** | End-user impact, usability, adoption, unintended consequences       | Empathetic, outcome-focused |

## Process

```dot
digraph debate {
  rankdir=TB;
  node [shape=box];

  start [label="Receive problem" shape=doublecircle];
  clarify [label="Clarify problem with user\n(1 round max)"];
  r1 [label="ROUND 1: Opening Positions\n5 agents respond independently\n(parallel)"];
  compile1 [label="Compile Round 1 transcript"];
  r2 [label="ROUND 2: Critique & Challenge\n5 agents see Round 1, attack flaws\n(parallel)"];
  compile2 [label="Compile Rounds 1-2 transcript"];
  r3 [label="ROUND 3: Synthesis & Refinement\n5 agents see Rounds 1-2, synthesize\n(parallel)"];
  compile3 [label="Compile full transcript"];
  lead [label="TEAM LEAD: Convergence\n1 agent reads all 3 rounds\nproduces unified plan"];
  present [label="Present unified plan\nto user" shape=doublecircle];

  start -> clarify -> r1 -> compile1 -> r2 -> compile2 -> r3 -> compile3 -> lead -> present;
}
```

### Step 1: Clarify the Problem

Before launching the debate, ensure you have a well-defined problem. If ambiguous, ask ONE round of clarifying questions:

- What specific problem or decision needs to be resolved?
- What constraints exist (time, budget, team, tech stack)?
- What has already been tried or decided?
- What does success look like?

### Step 2: Round 1 — Opening Positions

Dispatch all 5 agents in a **single message** (parallel). Each agent gets:

**Round 1 prompt template:**

```
You are a debater in a structured argumentation chamber analyzing a problem.

## Your Role: [ROLE_NAME]
[ROLE_MANDATE]. Your style is [STYLE].

## Problem
[USER'S PROBLEM STATEMENT]

## Context
[ANY CONSTRAINTS OR BACKGROUND]

## Round 1 Instructions
This is the OPENING ROUND. Present your initial position on this problem.

1. **Your Position** (2-3 paragraphs): How do you see this problem through your [ROLE_NAME] lens? What matters most?
2. **Your Proposal**: A concrete, actionable recommendation. Be specific enough that someone could start implementing it.
3. **Your Strongest Argument**: The single most compelling reason your approach is right.
4. **Your Acknowledged Weakness**: One legitimate concern about your own proposal (intellectual honesty earns credibility in later rounds).

Keep your response under 400 words. Be direct and opinionated — bland consensus positions will be torn apart in Round 2.
```

**Use `model: "sonnet"` for all 5 debaters.** Reserve Opus for the Team Lead convergence.

After all 5 agents return, compile the **Round 1 Transcript** — a numbered document with each agent's full response, labeled by role.

### Step 3: Round 2 — Critique & Challenge

Dispatch all 5 agents again in a **single message** (parallel). Each agent now sees the full Round 1 transcript:

**Round 2 prompt template:**

```
You are a debater in Round 2 of a structured argumentation chamber.

## Your Role: [ROLE_NAME]
[ROLE_MANDATE]. Your style is [STYLE].

## Problem
[USER'S PROBLEM STATEMENT]

From kabirdos/claude-toolkit, MIT.

More from kabirdos/claude-toolkit