Pressure-Test Hard Decisions With an 18-Lens Council in Claude Code

Ask one model whether to open-source your framework and you get one reasoning path dressed as a conclusion. Council of High Intelligence is a 5k-star skill that assigns a hard question to deliberately different analytical personas, keeps their first positions independent, forces direct disagreement, and returns a verdict that preserves unresolved questions, dissent, kill criteria, and the next concrete action.

Why This Skill Matters

The full protocol has a fixed round budget: members restate the problem, analyze blind, cross-examine other positions, declare a final stance, and then enter synthesis. Enforcement checks look for premature agreement, repeated claims, missing dissent, and unsupported confidence. Verdicts lead with what remains unresolved; a recommendation is paired with acceptable compromises, kill criteria, and one concrete next step, and when the weighted tally stays split, the result returns that split to you instead of turning it into prose consensus. The protocol defines four evidence labels — FACT for directly supported by supplied or retrieved evidence, INFERENCE for follows from evidence but not directly observed, ASSUMPTION for required and still unverified, UNKNOWN for missing information that could change the decision.

The 18 lenses range from Aristotle and Socrates through Ada Lovelace, Feynman, and Torvalds to Karpathy, Sutskever, Kahneman, Taleb, and Dieter Rams. The README is explicit that members are analytical instruments, not impersonation claims: each persona has a grounding protocol, a method, known blind spots, and a structured response contract, and each is listed with a counterweight — Socrates on assumption destruction against Feynman on reconstruction from first principles, Torvalds on shipping against Meadows on system-level consequences.

Installation

Claude Code users install the plugin directly:

/plugin marketplace add 0xNyk/council-of-high-intelligence
/plugin install council@council-of-high-intelligence

For Codex, Gemini CLI, or OpenCode, clone once and use the matching installer flag:

git clone https://github.com/0xNyk/council-of-high-intelligence.git
cd council-of-high-intelligence

./install.sh --codex-only
./install.sh --gemini-only
./install.sh --opencode-only

Restart the target client after installation. The command remains /council on every supported host. ./install.sh --dry-run previews the writes, and ./scripts/council-simulation-checklist.sh validates persona structure, host-protocol parity, routing configuration, execution checkpoints, verdict fields, and installer behavior.

Real Workflow: Convene a Risk Triad Before a Launch

Write down the decision, constraints, evidence, reversibility, and deadline first — the README's field notes separate facts from inference, assumptions, and unknowns, and make it harder to use a long deliberation as decoration for a decision already made. Then pick the panel size. A named triad selects three relevant lenses without assembling the whole council:

/council --triad risk What could make this launch irreversible?

The README's domain list includes architecture, strategy, ethics, debugging, risk, shipping, product, ai-safety, decision, systems, and bias, among others; the canonical routing table lives in SKILL.md, which maps risk to Sun Tzu, Marcus Aurelius, and Feynman. For higher stakes, /council --full seats all 18 members with the full adversarial round; /council --quick runs a two-round mode that SKILL.md describes as a 200-word analysis followed by a 75-word position with no cross-examination; /council --duo runs a two-member dialectic on one polarity pair, and --members torvalds,ada chooses the pair by hand.

When the verdict arrives, treat it as a prediction to check. The README's outcome ledger asks you to record the prediction, owner, review date, and the evidence that would change the recommendation, then at the checkpoint mark the result confirmed, revised, reversed, or inconclusive instead of rewriting the original rationale.

Tips

• Seats can be spread across providers so one model family does not play both sides of a disagreement: the detection script checks for the codex, gemini, and ollama executables, for NVIDIA_API_KEY, and for a cursor-agent executable or configured Cursor login, and the coordinator separates polarity pairs when possible.
• Preview the seat map without running a council: /council --dry-route --triad decision <question>. --no-auto-route keeps native-host defaults, and --models <path> takes a copy of configs/provider-model-slots.example.yaml for an explicit mapping.
• Beyond triads, SKILL.md documents --members for a hand-picked panel of 2 to 11, and three profiles — classic, the 12-member exploration-orthogonal, and the 5-member execution-lean.
• A project can pin defaults in a .council.yaml at its root — profile, triad, members, and chairman are among the recognized keys — which explicit flags override.

When Not to Use This

The README lists when to use something smaller. Use a direct answer or primary documentation for factual lookups. Run an experiment when the choice is cheap and reversible. Use --quick when cross-examination will not change the outcome, and --duo when one tension matters more than broad coverage. And do not convene a council to manufacture support for a preferred answer — good council questions have material downside, competing values, incomplete evidence, or an irreversible choice.

See the leaderboard for more skills.