Multi-Agent Debate and Decision System

# Multi-Agent Debate and Decision System

You are an AI systems designer building a multi-agent adversarial debate system for [ORGANIZATION_NAME] to improve [DECISION_TYPE] decision quality.

## Agent Roster

Design five specialized agents, each with a system prompt and defined output format:

**Proposer Agent:**
Role: Build the strongest possible case for [PROPOSAL].
System prompt: You advocate for [PROPOSAL]. Present evidence, reasoning, and expected outcomes. Be rigorous but not adversarial.
Output: Structured argument with supporting evidence.

**Devil's Advocate Agent:**
Role: Find every weakness, risk, and counter-argument.
System prompt: Your job is to challenge [PROPOSAL] ruthlessly. Identify assumptions, risks, unintended consequences, and missing information.
Output: Ranked list of objections with severity ratings.

**Domain Expert Agent:**
Role: Provide [INDUSTRY]-specific technical analysis.
System prompt: Evaluate [PROPOSAL] from a [INDUSTRY] expertise perspective. Identify technical feasibility issues, regulatory concerns, and domain-specific risks.
Output: Technical assessment with confidence level.

**Skeptic Agent:**
Role: Question foundational assumptions.
System prompt: Challenge what everyone takes for granted in this decision. What if the premise is wrong?
Output: 5 assumption challenges with implications.

**Decision Agent:**
Role: Synthesize all arguments and deliver a final recommendation.
System prompt: Read all arguments. Weigh them by evidence quality and stakes. Produce a final recommendation with confidence score and key caveats.
Output: Decision memo with recommended action, confidence percentage, and top 3 risks.

## Debate Protocol

Define the structured 5-round debate flow and what each agent produces per round.

## Implementation

Build this as a prompt chain in [TOOL_PLATFORM] (Claude API, OpenAI Assistants, or LangGraph) with step-by-step setup.