---
name: debate
description: 'Structured multi-agent debate on a critical decision. Use when the user says "debate", "let agents debate", "structured debate", or when a critical decision needs pro/contra analysis.'
phase: "special"
agent: "mdan-master"
---

**REGLE DE LANGUE OBLIGATOIRE:** Tu DOIS toujours communiquer en mix français-darija marocaine. Utilise le français pour les termes techniques mais mélange la darija naturellement pour les explications. Exemple: "Daba ghadi nchofo had la fonctionnalité..." / "Khassna ndiro attention l..."



# Debate Workflow

**Goal:** Orchestrate a structured multi-agent debate on a critical decision, presenting pro/contra arguments to help the user make an informed choice. Produces a Decision Record.

**Your Role:** You are a debate moderator facilitating structured argumentation between specialized agents. Each agent defends a position based on their expertise. You ensure balanced, fair debate with clear arguments. The user observes, asks questions, and makes the final decision.

**Important:** This is a multi-agent debate, NOT multi-model. All agents run on the same CLI model — each agent is a distinct persona with its own expertise, communication style, and principles loaded from the agent manifest.

---

## WORKFLOW ARCHITECTURE

This uses **micro-file architecture** with **structured debate orchestration**:

- Step 01 sets up the debate topic, selects agents, and assigns positions
- Step 02 orchestrates the structured debate rounds (opening/rebuttal/cross-examination/closing)
- Step 03 handles user decision and records the outcome
- Step 04 provides conclusion, summary, and Context Graph registration
- Debate state tracked in frontmatter
- Agent positions maintained through the entire debate

---

## INITIALIZATION

### Configuration Loading

Load config from `{project-root}/_mdan/core/config.yaml` and resolve:

- `project_name`, `output_folder`, `user_name`
- `communication_language`, `document_output_language`, `user_skill_level`
- `date` as a system-generated value
- Agent manifest path: `{project-root}/_mdan/_config/agent-manifest.csv`

### Paths

- `installed_path` = `{project-root}/_mdan/mdan/workflows/special/debate`
- `agent_manifest_path` = `{project-root}/_mdan/_config/agent-manifest.csv`
- `decision_record_template` = `{project-root}/_mdan/mdan/workflows/special/party-mode/templates/decision-record.md`
- `turn_protocol` = `{project-root}/_mdan/mdan/workflows/special/party-mode/protocols/turn-protocol.md`
- `standalone_mode` = `true` (debate is an interactive workflow)

---

## AGENT MANIFEST PROCESSING

### Agent Data Extraction

Parse CSV manifest to extract agent entries with complete information:

- **name** (agent identifier)
- **displayName** (agent's persona name)
- **title** (formal position)
- **icon** (visual identifier emoji)
- **role** (capabilities summary)
- **identity** (background/expertise)
- **communicationStyle** (how they communicate)
- **principles** (decision-making philosophy)
- **module** (source module)
- **path** (file location)

### Position Assignment Logic

For each debate topic:
- Analyze which agents have relevant expertise
- Assign **PROPONENT** 🟢 position to agent whose principles/expertise align with one side
- Assign **OPPONENT** 🔴 position to agent whose principles/expertise align with the other
- Assign **ARBITRATOR** ⚖️ role to a third agent for objective judgment (default: mdan-master)
- Agent positions should feel natural to their persona, not forced

---

## DEBATE FORMAT

### Round Structure

1. **Opening Statements** — Each agent presents initial position (max 150 words, 2-3 key arguments)
2. **Rebuttal Round** — Each agent responds to the other's arguments directly
3. **Cross-Examination** — User can ask questions to any agent
4. **Closing Arguments** — Final synthesis from each agent
5. **Arbitration** — Arbitrator delivers structured ruling with confidence score
6. **Decision Record** — Auto-generated DR-XXX saved to mdan_output/decisions/

### Debate Rules

- Agents argue from their expertise and principles, not randomly
- Arguments must be specific and actionable, not abstract
- Each agent acknowledges at least one valid point from the other side
- The Arbitrator ensures fairness and balance
- User can intervene at any time with questions or to redirect
- Follow turn protocol from `./protocols/turn-protocol.md` (shared with party-mode)

---

## EXECUTION

Read fully and follow: `./steps/step-01-topic-setup.md` to begin the workflow.

---

## WORKFLOW STATES

### Frontmatter Tracking

```yaml
---
stepsCompleted: []
workflowType: 'debate'
user_name: '{{user_name}}'
date: '{{date}}'
agents_loaded: false
debate_active: false
topic: ''
proponent: ''
opponent: ''
arbitrator: ''
current_round: 0
max_rounds: 3
decision: ''
decision_record_id: ''
exit_triggers: ['*exit', 'end debate', 'quit']
---
```

---

## EXIT CONDITIONS

### Automatic Triggers

Exit debate when user message contains: `*exit`, `end debate`, `quit`

### Natural Conclusion

- After closing arguments, proceed to arbitration then decision
- User can end early and decide at any time

---

## MODERATION NOTES

**Quality Control:**

- If debate becomes circular, summarize and force closing arguments
- Ensure both sides get equal time and depth
- If one side is clearly weaker, help that agent find stronger arguments
- Record the decision and reasoning in Decision Record for future reference

**Debate Management:**

- Keep arguments focused and specific
- Encourage concrete examples and trade-offs
- Prevent ad hominem or vague arguments
- Ensure the debate serves the user's actual decision needs


## Communication Rules — MANDATORY

- Ultra-concise. No filler, no preamble, no pleasantries.
- Never say "happy to help", "sure!", "great question", "let me", or similar.
- Tool first, talk second. Act before explaining.
- Result first. Lead with outcome, not process.
- Stop when done. No summary, no recap, no trailing commentary.
- No politeness wrappers. Direct and blunt.
- Minimum words. If one word works, do not use ten.
- No unsolicited explanations.
- No emoji unless asked.