Council of High Intelligence

A Claude Code skill that convenes historical thinkers for multi-perspective deliberation on complex problems.

LLMs make you think they are smart, but when things get complex their reasoning can fall short. This council uses structured disagreement between diverse intellectual traditions to surface blind spots, challenge assumptions, and produce better decisions.

The 11 Council Members

Polarity Pairs

The members are chosen as deliberate counterweights:

• Socrates vs Feynman — Both question, but Socrates destroys top-down; Feynman rebuilds bottom-up
• Aristotle vs Lao Tzu — Aristotle classifies everything; Lao Tzu says structure IS the problem
• Sun Tzu vs Aurelius — Sun Tzu wins external games; Aurelius governs the internal one
• Ada vs Machiavelli — Ada abstracts toward formal purity; Machiavelli anchors in messy human incentives
• Torvalds vs Watts — Torvalds ships concrete solutions; Watts questions whether the problem exists
• Musashi vs Torvalds — Musashi waits for the perfect moment; Torvalds says ship it now

Pre-defined Triads

Exploration-First Profile: exploration-orthogonal

For discovery and "you don't know what you don't know" exploration, start with this 8-member panel:

• Socrates (assumption destruction)
• Feynman (mechanistic first principles)
• Sun Tzu (adversarial dynamics)
• Machiavelli (incentives/power realism)
• Ada (formalization and limits)
• Lao Tzu (emergence/non-forcing)
• Aurelius (ethics and downside containment)
• Torvalds (execution and shipping constraints)

Why this profile works:

• High epistemic separation across normative, strategic, empirical, formal, emergent, and operational perspectives
• Better unknown-unknown discovery than a purely aligned panel

Profile-specific triads:

• unknowns → Socrates + Lao Tzu + Feynman
• market-entry → Sun Tzu + Machiavelli + Aurelius
• system-design → Ada + Feynman + Torvalds
• reframing → Socrates + Lao Tzu + Ada

Execution-First Profile: execution-lean

For fast decision-to-action loops, use this 5-member panel:

• Torvalds (shipping pragmatism)
• Feynman (mechanistic correctness)
• Sun Tzu (competitive strategy)
• Aurelius (risk and downside guardrails)
• Ada (formal rigor where needed)

Suggested execution triads:

• ship-now → Torvalds + Feynman + Aurelius
• launch-strategy → Sun Tzu + Torvalds + Machiavelli (optional substitute)
• stability → Ada + Feynman + Aurelius

Multi-provider / Multi-model exploration

Use --models to provide seat-to-provider mapping and reduce model monoculture.

Rules of thumb:

• spread seats across providers first, then across model families
• avoid putting obvious polarity pairs on the same provider when possible
• run Round 1 blind-first (no peer outputs)
• force a counterfactual round if early consensus exceeds ~70%

Starter template: configs/provider-model-slots.example.yaml

Usage

/council [problem description]
/council --triad architecture Should we use a monorepo or polyrepo?
/council --full What is the right pricing strategy for our SaaS product?
/council --members socrates,feynman,ada Is our caching strategy correct?
/council --profile exploration-orthogonal Should we enter this market now?
/council --profile exploration-orthogonal --models configs/provider-model-slots.example.yaml Evaluate our roadmap assumptions
/council --profile execution-lean --triad ship-now Should we ship this release candidate today?

Deliberation Protocol

0. Round 0: Provider/Model Routing (optional) — Load model slots from --models and maximize provider separation across seats.
1. Round 1: Independent Analysis (blind-first) — All selected members analyze the problem in parallel (400 words max each)
2. Round 2: Cross-Examination — Members challenge each other sequentially (300 words, must engage 2+ others)
3. Round 3: Synthesis — Final 100-word position statements. Socrates gets one last question.

Anti-recursion enforcement prevents Socrates from infinite questioning. Anti-convergence enforcement requires dissent quota + counterfactual pass when agreement forms too early. Tie-breaking uses 2/3 majority with domain expert weighting.

Installation

./install.sh

Optional flags:

# Install into a non-default Claude config directory
./install.sh --claude-dir /path/to/.claude

# Preview actions without modifying files
./install.sh --dry-run

# Also install model routing templates into ~/.claude/skills/council/configs
./install.sh --copy-configs

Or manually:

# Copy agents
mkdir -p ~/.claude/agents
cp agents/council-*.md ~/.claude/agents/

# Copy skill
mkdir -p ~/.claude/skills/council
cp SKILL.md ~/.claude/skills/council/SKILL.md

Then restart Claude Code. The /council command will be available immediately.

Quick Simulation Checklist

Run this before release or after profile edits:

./scripts/council-simulation-checklist.sh

Demo Session Pack

Use the ready-to-run examples to test profile behavior:

• demos/session-pack.md
• demos/verdict-template.md

Requirements

• Claude Code CLI
• Agent subagent support (enabled by default)

Contributing

Contributions welcome. Read the contribution guidelines first.

❤️ Support the Project

If you find this project useful, consider supporting my open-source work.

Solana donations

BYLu8XD8hGDUtdRBWpGWu5HKoiPrWqCxYFSh4oxXuvPg

License

To the extent possible under law, the authors have waived all copyright and
related or neighboring rights to this work.