linkdrop-x-tom-doerr-multi-llm-council-deliberation

name	linkdrop-x-tom-doerr-multi-llm-council-deliberation
description	18 AI personas (Aristotle, Socrates, Sun Tzu, Feynman, Kahneman, Karpathy, Sutskever, Taleb, etc.) deliberate hard decisions via structured multi-round debate across Claude/OpenAI/Gemini/Ollama. Use when making a high-stakes call where genuine multi-model disagreement beats single-model consensus.
source	https://x.com/tom_doerr/status/2045839498634727752 → https://github.com/0xNyk/council-of-high-intelligence
allowed-tools	Read, Write, Edit, Bash, Glob, Grep

linkdrop-x-tom-doerr — Council of High Intelligence

Overview

0xNyk/council-of-high-intelligence is a Claude Code + Codex skill (CC0 license, 311+ stars) that installs 18 named council members — each a philosophical / disciplinary archetype — and runs structured multi-round deliberation across multiple LLM providers. The pitch: a single LLM gives you "one reasoning path dressed up as confidence." The council gives you structured disagreement instead.

The real differentiator vs our existing debate patterns (/mundi:debate, /mundi:consensus, stochastic-multi-agent-consensus):

1. Named members with polarity pairs. Not generic debate roles — specific archetypes paired for deliberate tension (Socrates↔Feynman, Torvalds↔Watts, Karpathy↔Sutskever, 10 more).
2. Multi-provider auto-routing. Members distributed across Claude, OpenAI, Gemini, Ollama. No single model's bias dominates.
3. Pre-built triads and profiles. 20 domain-specific 3-member combos + 3 pre-built panels for different decision types.

Install

git clone https://github.com/0xNyk/council-of-high-intelligence.git
cd council-of-high-intelligence
./install.sh           # Claude Code
# OR
./install.sh --codex   # Codex

Then in Claude Code:

/council Should we open-source our agent framework?
/council --quick Should we add caching here?
/council --duo Should we use microservices or monolith?
/council --triad ai-safety What's our AI policy?

The 18 members (actual, from repo)

Agent	Figure	Domain	Default model	Polarity
council-aristotle	Aristotle	Categorization & structure	opus	Classifies everything
council-socrates	Socrates	Assumption destruction	opus	Questions everything
council-sun-tzu	Sun Tzu	Adversarial strategy	sonnet	Reads terrain & competition
council-ada	Ada Lovelace	Formal systems & abstraction	sonnet	What can/can't be mechanized
council-aurelius	Marcus Aurelius	Resilience & moral clarity	opus	Control vs acceptance
council-machiavelli	Machiavelli	Power dynamics & realpolitik	sonnet	How actors actually behave
council-lao-tzu	Lao Tzu	Non-action & emergence	opus	When less is more
council-feynman	Feynman	First-principles debugging	sonnet	Refuses unexplained complexity
council-torvalds	Linus Torvalds	Pragmatic engineering	sonnet	Ship it or shut up
council-musashi	Miyamoto Musashi	Strategic timing	sonnet	The decisive strike
council-watts	Alan Watts	Perspective & reframing	opus	Dissolves false problems
council-karpathy	Andrej Karpathy	Neural network intuition	sonnet	How models actually learn and fail
council-sutskever	Ilya Sutskever	Scaling frontier & AI safety	opus	When capability becomes risk
council-kahneman	Daniel Kahneman	Cognitive bias & decision science	opus	Your own thinking is the first error
council-meadows	Donella Meadows	Systems thinking & feedback loops	sonnet	Redesign the system, not the symptom
council-munger	Charlie Munger	Multi-model reasoning & economics	sonnet	Invert — what guarantees failure?
council-taleb	Nassim Taleb	Antifragility & tail risk	opus	Design for the tail, not the average
council-rams	Dieter Rams	User-centered design	sonnet	Less, but better — the user decides

Polarity pairs (13 intentional counter-weights)

• Socrates ↔ Feynman — destroys top-down vs rebuilds bottom-up
• Aristotle ↔ Lao Tzu — classifies everything vs structure IS the problem
• Sun Tzu ↔ Aurelius — wins external games vs governs the internal one
• Ada ↔ Machiavelli — formal purity vs messy human incentives
• Torvalds ↔ Watts — ships concrete solutions vs questions whether the problem exists
• Musashi ↔ Torvalds — waits for the perfect moment vs ships it now
• Karpathy ↔ Sutskever — build + iterate vs pause + research + ensure safety
• Karpathy ↔ Ada — empirical ML intuition vs formal systems theory
• Kahneman ↔ Feynman — your cognition is the first error vs trust first-principles reasoning
• Meadows ↔ Torvalds — redesign the feedback loop vs fix the symptom and ship
• Munger ↔ Aristotle — multi-model lattice vs single taxonomic system
• Taleb ↔ Karpathy — hidden catastrophic tails vs smooth empirical scaling curves
• Rams ↔ Ada — what the user needs vs what computation can do

Three deliberation modes

1. Full (default) — 3 rounds: independent analysis → cross-examination → final positions. Use for high-stakes one-way-door decisions.
2. --quick — 2 rounds, no cross-examination. Use for simpler decisions where you still want divergence but not the full 3-round cost.
3. --duo — 2-member dialectic using a polarity pair. Use when you want to explore a specific tension (e.g., --duo --members torvalds,ada).

20 pre-defined domain triads

architecture   = Aristotle + Ada + Feynman       (classify + formalize + simplicity-test)
strategy       = Sun Tzu + Machiavelli + Aurelius (terrain + incentives + moral grounding)
ethics         = Aurelius + Socrates + Lao Tzu    (duty + questioning + natural order)
debugging      = Feynman + Socrates + Ada         (bottom-up + assumption testing + formal verification)
innovation     = Ada + Lao Tzu + Aristotle
conflict       = Socrates + Machiavelli + Aurelius
complexity     = Lao Tzu + Aristotle + Ada
risk           = Sun Tzu + Aurelius + Feynman
shipping       = Torvalds + Musashi + Feynman
product        = Torvalds + Machiavelli + Watts
founder        = Musashi + Sun Tzu + Torvalds
ai             = Karpathy + Sutskever + Ada
ai-product     = Karpathy + Torvalds + Machiavelli
ai-safety      = Sutskever + Aurelius + Socrates
decision       = Kahneman + Munger + Aurelius
systems        = Meadows + Lao Tzu + Aristotle
uncertainty    = Taleb + Sun Tzu + Sutskever
design         = Rams + Torvalds + Watts
economics      = Munger + Machiavelli + Sun Tzu
bias           = Kahneman + Socrates + Watts

Invoke with /council --triad <domain> <question>.

Three council profiles

• classic (default) — all 18 members with domain triads. Broad deliberation.
• exploration-orthogonal — 12-member panel for "unknown unknowns" (Socrates, Feynman, Sun Tzu, Machiavelli, Ada, Lao Tzu, Aurelius, Torvalds, Karpathy, Sutskever, Kahneman, Meadows).
• execution-lean — 5-member panel for fast decision-to-action (Torvalds, Feynman, Sun Tzu, Aurelius, Ada).

Multi-provider auto-routing

The install script detects which LLM provider CLIs are present (claude, openai, gemini, ollama) and distributes members across them automatically. No config required. The whole point is genuine model diversity — 18 correlated voices on one provider gives you groupthink.

When to use

• High-stakes one-way-door decisions: architectural pivots, hiring, major spend, IC memos.
• Auditing a recommendation from a single model before acting.
• Strategy sessions where diverse priors genuinely matter (ethics, regulatory, competitive).
• Pre-mortem style risk review — let Taleb + Sun Tzu + Sutskever hunt tail risks.

Do NOT use when:

• The decision is reversible and cheap — just act.
• You only have one LLM provider configured (no real diversity possible).
• The stakes don't justify 18× token cost vs single-model output.

Cost structure

Per the repo's own guidance: full 18-member 3-round deliberation is substantial spend. Use --quick or --duo or a triad for routine calls. Reserve full classic profile for the decisions that genuinely warrant it.

Install strategy for this monorepo

Option A — vendored install (recommended):

1. Clone into ~/sandbox/council-of-high-intelligence/ (not global).
2. Review install.sh before running — it writes to ~/.claude/.
3. Run ./install.sh. Verify no overwrites of existing skills (our local linkdrop-x-tom-doerr-* should be untouched since names don't collide).
4. Test with /council --quick Is this migration safe? on a low-stakes question first.

Option B — pattern extraction: If you want the concept without the install:

• Read agents/council-*.md files from the repo for the 18 system prompts.
• Adapt into our existing /mundi:debate or /mundi:consensus with a --personas flag.

Integration with our stack

Our existing analogues — keep or replace:

• /mundi:debate — 2-side debate. Keep; /council --duo is the install-free equivalent.
• /mundi:consensus — N-agent consensus. Complements; council adds named archetypes + polarity.
• stochastic-multi-agent-consensus — stochastic aggregation. Complementary; use for non-named-persona votes.
• saraev-iterative-multi-agent-debate-visualization — visualization layer. Pair with council for artifact presentation.
• superpowers:dispatching-parallel-agents — the underlying parallel-exec primitive. Prerequisite.

Safety + gotchas

• License: CC0 / public domain (per repo badge). Safe to vendor with attribution.
• Install review: install.sh writes to your ~/.claude/ directory. Always read before running in a shared environment.
• Provider monoculture: if only Anthropic is configured, the "multi-provider" promise degrades to single-provider routing. Configure at least 2 for real diversity.
• Budget blowouts: 18 members × 3 rounds = 54 LLM calls per decision. Cap with --quick or --duo unless stakes justify full mode.
• Persona drift on weaker models: if some members route to smaller models, their system-prompt distinctions can collapse. The repo's sonnet / opus defaults are tuned to avoid this.
• Not a substitute for human judgment: the council is an input to your decision, not a replacement. Save the artifact; make the call yourself.

Source

• Tweet: https://x.com/tom_doerr/status/2045839498634727752 (2026-04-19 — Tom Doerr)
• Repo: https://github.com/0xNyk/council-of-high-intelligence (311 stars, CC0, MIT-like)
• Captured: 2026-04-18 via link-drop-pipeline; rewritten 2026-04-20 against live repo README.

Cross-references

• /mundi:debate, /mundi:consensus, /mundi:debate-then-verify
• stochastic-multi-agent-consensus
• saraev-iterative-multi-agent-debate-visualization
• superpowers:dispatching-parallel-agents, superpowers:subagent-driven-development
• ~/.claude/rules/agents.md
• docs/knowledge-base/wiki/01-ai-development-and-agents/INDEX.md

Related KB entry

• docs/knowledge-base/wiki/01-ai-development-and-agents/linkdrop-x-tom-doerr-multi-llm-council-deliberation.md