🏛️ LLM Council Skill Tutorial

What is the LLM Council?
A structured multi-perspective decision-making system that pressure-tests complex choices by running them through multiple AI advisors with different thinking styles and models, then synthesizing their insights into a clear recommendation.

🎯 When to Use the Council

✅ Perfect for:
❌ Don't use for:

🗣️ How to Trigger a Council

Simply use one of these trigger phrases:

Or be explicit about the mode:

🎭 The Three Council Modes

Lehmann Mode (Default)

What it does: Five advisors with different thinking styles analyze your question, then peer-review each other's responses. A chairman synthesizes everything into a clear verdict.

Best for: Most decisions. Strategic analysis. When you want diverse thinking styles applied to your problem.
The Contrarian
Actively looks for what's wrong, what's missing, what will fail. Your friend who saves you from a bad deal.
The First Principles Thinker
Strips away assumptions and rebuilds the problem from the ground up. Sometimes says you're asking the wrong question entirely.
The Expansionist
Looks for upside everyone else is missing. What could be bigger? What adjacent opportunity is hiding?
The Outsider
Has zero context about your field or history. Catches the curse of knowledge and blind spots.
The Executor
Only cares about one thing: can this actually be done? What do you do Monday morning?

Karpathy Mode

What it does: Four different AI models analyze your question independently, then rank each other's responses. Includes quantitative ranking aggregation.

Best for: When you want to leverage different AI model capabilities. Technical decisions. When you want measurable comparison between different AI perspectives.

Models used:

Hybrid Mode (Experimental)

What it does: Combines both approaches — five thinking styles each running on different AI models, with both qualitative review and quantitative ranking.

Best for: Maximum scrutiny decisions. When cost and time aren't constraints. Research and analysis where you want the most comprehensive perspective possible.
⚠️ Important Limitation:
In Hybrid mode, it's hard to tell whether insights came from the thinking style or the AI model's natural tendencies. Results are comprehensive but less interpretable.

⚙️ How the Council Works

Detailed Workflow (7 Steps)
Step 1: Context Enrichment
The system scans your workspace for relevant files and context, then frames your question with all necessary background information.
Step 2: Advisor Round
Multiple advisors (5 in Lehmann/Hybrid, 4 in Karpathy) analyze your question in parallel, each bringing their unique perspective or model capability.
Step 3: Anonymization
Advisor responses are anonymized and randomized to prevent bias in the review phase.
Step 4: Peer Review
Fresh reviewers evaluate all advisor responses, identifying strengths, blind spots, and what everyone missed.
Step 5: Ranking (Karpathy/Hybrid)
In modes that include ranking, reviewers also rank responses from best to worst, creating quantitative comparison data.
Step 6: Chairman Synthesis
A chairman receives all advisor responses, peer reviews, and rankings, then synthesizes everything into a clear verdict with specific structure.
Step 7: Report Generation
The system generates both an HTML report (for easy reading) and a markdown transcript (for detailed analysis).

📊 Mode Comparison

Aspect Lehmann Karpathy Hybrid
Advisors 5 thinking styles, same model 4 different models, no specific styles 5 thinking styles across different models
Review Type Qualitative peer review only Ranking and evaluation Combined qualitative + ranking
Output Structured 5-section verdict Synthesis + ranking leaderboard Comprehensive analysis + rankings
Best For Strategic decisions, diverse perspectives Technical analysis, model comparison Maximum scrutiny, research
Time/Cost Medium Medium High
Interpretability High (clear thinking styles) High (clear model differences) Lower (style/model confounding)

💡 Example: Course Creation Decision

User Question:
"Council this: I'm thinking of building a $297 course on Claude Code for beginners. My audience is mostly non-technical solopreneurs. Is this the right move?"
See How Each Advisor Responded

The Contrarian: "The market is flooded with Claude courses right now. At $297, you're competing with free YouTube content. Your audience is non-technical, which means high support burden and refund risk..."

The First Principles Thinker: "What are you actually trying to achieve? If it's revenue, a course is one of the slowest paths. If it's authority, a free resource might do more..."

The Expansionist: "Beginner Claude for solopreneurs is a massive underserved market. Everyone's teaching advanced stuff. If you nail the beginner angle, you own the entry point to this entire space..."

The Outsider: "I don't know what Claude Code is. If I saw '$297 course on Claude Code for beginners,' I wouldn't know if this is for me. The name means nothing to someone outside your world..."

The Executor: "A full course takes 4-8 weeks to produce properly. Before building anything, run a live workshop at $97 to 50 people. You validate demand, generate testimonials..."

Chairman's Verdict: Don't build the course yet. Validate with a lower-commitment offer first. But reframe entirely: sell the outcome (automate your business), not the tool. One thing to do first: Run a $97 live workshop called "How to automate your first business task with AI" to 50 people.

📋 What You Get

Every council run produces two artifacts:

HTML Report (Interactive)
Markdown Transcript (Complete Record)

🔧 Technical Details

Architecture & Implementation

The LLM Council uses a hybrid architecture:

Why this design? Previous pure-agent approaches drifted and made mistakes. This design moves all deterministic work to code while keeping judgment calls with the agent.

Reliability & Testing

The system includes three layers of validation:

Failure Handling: Built-in retry logic, minimum completion thresholds, and graceful degradation when some advisors fail.

🚀 Getting Started

  1. Identify a real decision you're facing (not a trivial question)
  2. Ask Atlas using a trigger phrase: "council this: [your question]"
  3. Provide clarification if Atlas asks for more context
  4. Wait for the council to complete (typically 3-5 minutes)
  5. Review the HTML report that opens automatically
  6. Use the recommendation as input for your decision
💡 Pro Tip:
The council works best when you provide context about what's at stake and any constraints you're working within. Don't just ask "Should I do X?" — explain why you're considering it and what success looks like.

📚 Advanced Usage

Choosing the Right Mode
Interpreting Results

Pay attention to:

Remember: The council provides input for your decision, not the decision itself. You still need to weigh the advice against your specific situation and values.

LLM Council Skill • Generated by Atlas •