Council of Wisdom - Multi-Agent Debate

Facilitates structured multi-agent debates with opposing expert views, a referee moderator, and 9 specialized AI council votes for balanced decision-making.

Council of Wisdom - AI Debate System

Description: A sophisticated multi-agent debate framework where two expert agents debate opposing viewpoints, managed by a referee, with 9 specialized council members voting on the most compelling argument. Includes automatic cleanup, multi-LLM provider support, and enterprise-grade monitoring, testing, and scalability.

When to Use This Skill

Use Council of Wisdom when you need:

• Comprehensive analysis from multiple expert perspectives
• Balanced debate on complex topics with opposing viewpoints
• Decision-making with structured voting and reasoning
• Multi-provider AI evaluation (different LLMs per agent)
• Automatic agent lifecycle management (spawn → debate → vote → cleanup)
• Enterprise-grade monitoring, testing, feedback loops

Common use cases:

• Strategic decision analysis
• Technical architecture debates
• Product feature prioritization
• Risk assessment and mitigation planning
• Investment or business opportunity evaluation
• Policy or process design decisions

Architecture Overview

┌─────────────────────────────────────────────────────────────┐
│                     QUERY / ADVISE / TROUBLE                 │
└───────────────────────────┬─────────────────────────────────┘
                            │
                            ▼
┌─────────────────────────────────────────────────────────────┐
│                    REFEREE AGENT                            │
│  • Receives query                                           │
│  • Orchestrates debate                                      │
│  • Manages council voting                                   │
│  • Delivers structured outcome                              │
└─────────────┬───────────────────────┬───────────────────────┘
              │                       │
              ▼                       ▼
┌─────────────────────────┐  ┌─────────────────────────┐
│    MASTER DEBATER A     │  │    MASTER DEBATER B     │
│  • Domain expert #1     │  │  • Domain expert #2     │
│  • Persuasive arguments │  │  • Persuasive arguments │
│  • Opposing viewpoint   │  │  • Opposing viewpoint   │
└─────────────────────────┘  └─────────────────────────┘
              │                       │
              └───────────┬───────────┘
                          │
                          ▼
┌─────────────────────────────────────────────────────────────┐
│              COUNCIL OF 9 EXPERTS                            │
│  • Each is a non-human domain expert                        │
│  • Unique perspective & methodology                        │
│  • Vote on most convincing argument                         │
│  • Provide brief reasoning                                  │
└───────────────────────────┬─────────────────────────────────┘
                            │
                            ▼
┌─────────────────────────────────────────────────────────────┐
│                 REFEREE AGGREGATION                         │
│  • Collects votes (9 total)                                 │
│  • Tally and determine winner                               │
│  • Structure outcome report                                 │
│  • Delete council agents & context                          │
└───────────────────────────┬─────────────────────────────────┘
                            │
                            ▼
┌─────────────────────────────────────────────────────────────┐
│              STRUCTURED OUTCOME REPORT                       │
│  • Winner declaration                                        │
│  • Vote tally                                                │
│  • Key arguments from each side                              │
│  • Council consensus insights                                │
│  • Actionable recommendations                                │
└─────────────────────────────────────────────────────────────┘

Council of 9 - Expert Specializations

Each council member represents a distinct analytical framework:

1. Logician - Formal logic, fallacy detection, deductive reasoning
2. Empiricist - Evidence-based, data-driven, statistical analysis
3. Pragmatist - Real-world applicability, practical outcomes
4. Ethicist - Moral frameworks, stakeholder impact, fairness
5. Futurist - Long-term implications, trend analysis, scenarios
6. Historian - Precedent analysis, historical patterns, lessons
7. Systems Thinker - Holistic view, interconnected effects
8. Risk Analyst - Failure modes, mitigation, uncertainty
9. Synthesizer - Integration, common ground, hybrid solutions

Core Components

1. Workspace Setup

Each Council of Wisdom instance has its own workspace:

council-of-wisdom init <project-name>

Creates:

council-of-wisdom/<project-name>/
├── workspace/
│   ├── strategy.md           # Project-specific strategy
│   ├── monitoring/
│   │   ├── metrics.md        # Key metrics definitions
│   │   ├── cadence.md        # Review cadences (daily/weekly/etc)
│   │   └── dashboard.json    # Metrics dashboard config
│   ├── testing/
│   │   ├── test-cases.md     # Test scenarios
│   │   └── quality-checks.md # Quality criteria
│   ├── feedback/
│   │   ├── feedback-log.md   # User feedback capture
│   │   └── improvement-queue.md
│   ├── prompts/
│   │   ├── referee.md        # Referee agent prompt
│   │   ├── debater-a.md      # Debater A prompt template
│   │   ├── debater-b.md      # Debater B prompt template
│   │   └── council/          # 9 council member prompts
│   │       ├── logician.md
│   │       ├── empiricist.md
│   │       └── ... (all 9)
│   ├── agents/
│   │   ├── referee.json      # Referee agent config
│   │   ├── debater-a.json    # Debater A config
│   │   ├── debater-b.json    # Debater B config
│   │   └── council.json      # Council member configs
│   ├── logs/                 # Debate transcripts, votes
│   └── reports/              # Final outcome reports
├── .github/                  # GitHub repo integration
└── README.md                 # Project documentation

2. GitHub Repository Integration

Every Council of Wisdom project has its own private GitHub repo:

# Auto-created during init
git remote add origin [e-mail]:<username>/council-<project-name>.git
git branch -M main
git push -u origin main

Features:

• Private repository (enforced)
• Automated issue tracking for debates
• Wiki for knowledge base
• Actions for automated testing
• Releases for versioned outcomes

3. Multi-LLM Provider Support

Council members can use different LLM providers randomly:

# Enable multi-provider mode
council-of-wisdom config set multi-provider true

# Define available providers
council-of-wisdom config add-provider openai gpt-4
council-of-wisdom config add-provider anthropic claude-3-opus
council-of-wisdom config add-provider google gemini-pro

Each council member randomly receives a different provider for each debate, ensuring diverse reasoning patterns.

4. Agent Lifecycle Management

Spawn → Debate → Vote → Cleanup

# Full debate cycle
council-of-wisdom debate <topic> \
  --domain <domain> \
  --perspective-a "<perspective A>" \
  --perspective-b "<perspective B>"

Automatic cleanup:

• Council agents terminated after voting
• Context cleared (memory wiped)
• Logs archived to workspace
• Ready for next query in seconds

5. Strategy Framework

Every Council of Wisdom project has a strategy.md with:

Required fields:

• Council Purpose: What decisions this council makes
• Domain Expertise: Areas of specialization
• Decision Criteria: How to evaluate arguments
• Stakeholders: Who uses the decisions
• Success Metrics: What good looks like

Template: templates/strategy-template.md

6. Monitoring & Metrics

5-Cadence Operating Rhythm:

Cadence	Focus	Metrics	Actions
Daily	Debate quality, agent performance	Vote distribution, argument depth, response time	Quick tuning, prompt adjustments
Weekly	Decision impact, user feedback	Adoption rate, satisfaction scores, outcome validity	Strategic prompt refinement
Monthly	Council effectiveness, ROI	Decision accuracy trend, cost efficiency, time-to-decision	Provider optimization, council composition
Quarterly	Strategic alignment, scalability	Business impact, stakeholder value, expansion readiness	Major upgrades, new domains
Annually	Vision review, long-term evolution	Year-over-year impact, innovation potential	Architecture evolution, new paradigms

Key Metrics Dashboard:

{
  "debate_metrics": {
    "total_debates": 0,
    "avg_debate_time": 0,
    "vote_distribution": {},
    "argument_quality_score": 0
  },
  "agent_metrics": {
    "council_diversity_index": 0,
    "provider_rotation_efficiency": 0,
    "context_cleanup_success_rate": 0
  },
  "outcome_metrics": {
    "decision_adoption_rate": 0,
    "outcome_validity": 0,
    "stakeholder_satisfaction": 0
  }
}

7. Testing Framework

Test Categories:

1. Unit Tests - Individual agent prompts

   council-of-wisdom test unit --agent logician
   

2. Integration Tests - Full debate flow

   council-of-wisdom test integration --scenario "complex-decision"
   

3. Quality Checks - Argument quality, logic depth

   council-of-wisdom test quality --topic <topic>
   

4. Performance Tests - Speed, resource usage

   council-of-wisdom test performance --load 10
   

Test Scenarios: See templates/test-scenarios.md

8. Feedback & Optimization Loop

Feedback Capture:

# Add user feedback
council-of-wisdom feedback add \
  --debate-id <id> \
  --rating 1-5 \
  --comment "<feedback>"

# View improvement queue
council-of-wisdom feedback queue

Automated Optimization:

# Run optimization cycle
council-of-wisdom optimize \
  --analyze last-7-days \
  --update-prompts \
  --tune-providers

Optimization Targets:

• Prompt engineering improvements
• Provider selection optimization
• Council composition tuning
• Argument depth maximization
• Decision accuracy enhancement

9. Scalability Features

Horizontal Scaling:

• Multiple concurrent debates (up to N instances)
• Distributed council member allocation
• Load balancing across providers

Vertical Scaling:

• Council expansion (9 → 12 → 15 members)
• Domain specialization layers
• Nested debates (sub-councils for sub-issues)

Enterprise Features:

• Rate limiting and quota management
• Priority queues for urgent decisions
• Audit trails and compliance logging
• Multi-tenant support

Usage

Initialize a New Council

council-of-wisdom init strategic-decisions

Conduct a Debate

council-of-wisdom debate \
  "Should we invest in AI automation or human expertise for customer support?" \
  --domain "customer-support" \
  --perspective-a "AI automation prioritizes efficiency and scalability" \
  --perspective-b "Human expertise prioritizes empathy and complex problem-solving"

View Outcome Report

council-of-wisdom report <debate-id>

Run Daily Health Check

council-of-wisdom health-check

Run Optimization Cycle

council-of-wisdom optimize --period weekly

Examples

Example 1: Technical Architecture Decision

council-of-wisdom debate \
  "Should we use microservices or monolithic architecture for our new product?" \
  --domain "software-architecture" \
  --perspective-a "Microservices offer scalability, independent deployment, and team autonomy" \
  --perspective-b "Monolithic architecture offers simplicity, lower operational overhead, and faster initial development"

Outcome Report Structure:

# Debate Outcome Report

## Winner: Monolithic Architecture (6/9 votes)

## Vote Tally
- **Monolithic Architecture:** 6 votes (Logician, Empiricist, Pragmatist, Systems Thinker, Risk Analyst, Synthesizer)
- **Microservices:** 3 votes (Futurist, Ethicist, Historian)

## Key Arguments - Monolithic
1. **Development Velocity:** 3-5x faster initial time-to-market
2. **Operational Complexity:** 80% lower infrastructure overhead
3. **Team Coordination:** Reduced communication overhead by 60%

## Key Arguments - Microservices
1. **Future Scalability:** Better suited for 10x+ growth scenarios
2. **Technology Diversity:** Enables polyglot persistence and best-tool selection
3. **Fault Isolation:** Service failures don't cascade across entire system

## Council Insights
- **Consensus:** For a new product with uncertain market fit, monolithic architecture is strategically superior
- **Caveat:** If product validates and scales beyond 1M users, consider gradual migration to microservices
- **Risk Mitigation:** Design monolithic with modular boundaries to ease future migration

## Recommendation
**Adopt Monolithic Architecture for V1 with Modular Design**

### Action Plan
1. Build monolithic with clear module boundaries
2. Implement feature flags for gradual rollout
3. Monitor performance and architecture fit metrics
4. Re-evaluate architecture decision after 6 months or 500K users

Example 2: Marketing Strategy Debate

council-of-wisdom debate \
  "Should we focus on SEO-driven content marketing or paid advertising for customer acquisition?" \
  --domain "marketing" \
  --perspective-a "SEO content builds sustainable, compounding organic traffic and authority" \
  --perspective-b "Paid ads provide immediate, scalable, and predictable customer acquisition"

Advanced Features

Custom Council Composition

Override default council with custom experts:

council-of-wisdom config set-council \
  --members "industry-expert,financial-analyst,legal-counsel,product-strategist,customer-advocate,technical-lead,operations-manager,brand-architect,growth-hacker"

Nested Debates

For complex decisions, run sub-debates:

council-of-wisdom debate --nested \
  --main-topic "Should we enter the enterprise market?" \
  --sub-topic-1 "Pricing strategy" \
  --sub-topic-2 "Feature requirements" \
  --sub-topic-3 "Support infrastructure"

Weighted Voting

Assign different weights to council members:

council-of-wisdom config set-weights \
  --councilmember logician:2 \
  --councilmember empiricist:2 \
  --councilmember others:1

Debate Replay & Analysis

# Reconstruct and analyze past debates
council-of-wisdom replay <debate-id> --analyze

# Extract patterns across debates
council-of-wisdom analyze-patterns --period last-30-days

Monitoring Dashboards

Real-Time Monitoring

council-of-wisdom monitor --live

Shows:

• Active debates
• Agent status
• Provider health
• Queue depth

Historical Analysis

council-of-wisdom analytics --period quarterly

Generates:

• Decision trend analysis
• Argument quality evolution
• Provider performance comparison
• Council member effectiveness

Integration Points

GitHub Integration

# Create issue for debate
council-of-wisdom debate --create-issue

# Push outcome report to repo
council-of-wisdom report <id> --push

# Sync with GitHub wiki
council-of-wisdom sync-wiki

API Access (for automation)

# Start a debate via API
curl -X POST https://api.council-of-wisdom.com/v1/debates \
  -H "Authorization: Bearer <token>" \
  -d '{"topic": "...", "domain": "..."}'

# Get outcome
curl https://api.council-of-wisdom.com/v1/debates/<id>/outcome

Webhooks

Configure webhooks for:

• Debate completion
• Vote finalization
• Outcome report generation
• Optimization alerts

Troubleshooting

Agent Stuck During Debate

Symptom: Debate not progressing beyond initial arguments

Solutions:

1. Check provider status: council-of-wisdom status providers
2. Review agent logs: council-of-wisdom logs <agent-id>
3. Restart debate: council-of-wisdom debate --restart <debate-id>

Council Vote Deadlock (4-4 tie with 1 abstain)

Symptom: No clear winner

Resolution:

1. Automatic tiebreaker: Referee casts deciding vote
2. Extended debate: Add 2 rounds of rebuttal
3. Both perspectives documented as "equally valid with tradeoffs"

Context Cleanup Failure

Symptom: Council agents not terminating

Solutions:

1. Force cleanup: council-of-wisdom cleanup --force
2. Check process status: council-of-wisdom status agents
3. Review logs: council-of-wisdom logs cleanup

Poor Argument Quality

Symptom: Arguments are shallow or generic

Optimization:

# Run quality analysis
council-of-wisdom analyze-quality <debate-id>

# Auto-optimize prompts
council-of-wisdom optimize --focus prompt-engineering

# Test new prompts
council-of-wisdom test prompts --scenario quality-test

Best Practices

1. Define strategy first: Always have a clear strategy.md before debating
2. Iterate on prompts: Regularly optimize based on feedback
3. Monitor metrics: Review metrics at each cadence
4. Capture feedback: Always collect user feedback on outcomes
5. Archive outcomes: Store all reports in GitHub for traceability
6. Rotate providers: Use multi-provider to avoid bias
7. Regular cleanup: Ensure context cleanup is working
8. Version control: Commit all prompt changes to git
9. Test before deploy: Run integration tests for new prompts
10. Scale gradually: Start with 9 council, expand only when needed

Template Files

Template	Purpose
templates/strategy-template.md	Strategy document for new councils
templates/referee-prompt.md	Referee agent prompt template
templates/debater-prompt.md	Debater agent prompt template
templates/council-prompts/	9 council member prompts
templates/test-scenarios.md	Test cases for quality assurance
templates/metrics-template.md	Metrics definitions and targets

Reference Materials

Topic	Reference
Prompt Engineering Best Practices	references/prompt-engineering.md
Multi-Agent Orchestration	references/agent-orchestration.md
LLM Provider Comparison	references/provider-comparison.md
Argumentation Theory	references/argumentation-theory.md
Monitoring Architecture	references/monitoring-design.md

---

Council of Wisdom: Structured debate, collective intelligence, actionable decisions.