![\"Council](\"assets/header.jpeg\")
\n\n \n
\n Structured multi-perspective deliberation for decisions that deserve more than one reasoning path.\n
\n\n\n ![\"Lint\"](\"https://github.com/0xNyk/council-of-high-intelligence/actions/workflows/lint.yml/badge.svg\"){kind=icon}
\n ![\"Release\"](\"https://img.shields.io/github/v/release/0xNyk/council-of-high-intelligence\")
![\"License\"](\"https://img.shields.io/badge/license-MIT-blue\"){kind=icon}
\n Claude Code · Codex · Gemini CLI · OpenCode\n
\n\nThe council assigns a hard question to deliberately different analytical personas, keeps\ntheir first positions independent, forces direct disagreement, and returns a verdict that\npreserves unresolved questions, dissent, kill criteria, and the next concrete action.\n\n## Try a council\n\nClaude Code users can install the plugin directly:\n\n```text\n/plugin marketplace add 0xNyk/council-of-high-intelligence\n/plugin install council@council-of-high-intelligence\n```\n\nThen convene the full council, a faster panel, or a two-member dialectic:\n\n```text\n/council Should we open-source our agent framework?\n/council --quick Should we add caching here?\n/council --duo Should we use microservices or a monolith?\n```\n\nFor Codex, Gemini CLI, or OpenCode, clone once and use the matching installer flag:\n\n```bash\ngit clone https://github.com/0xNyk/council-of-high-intelligence.git\ncd council-of-high-intelligence\n\n./install.sh --codex-only\n./install.sh --gemini-only\n./install.sh --opencode-only\n```\n\nRestart the target client after installation. The command remains `/council` on every\nsupported host.\n\n## Use it for the decision boundary\n\nGood council questions have material downside, competing values, incomplete evidence, or\nan irreversible choice. Write down the decision, constraints, evidence, reversibility, and\ndeadline before adding personas.\n\n\n\nThe field notes distinguish facts from inference, assumptions, and unknowns. They also make\nit harder to use a long deliberation as decoration for a decision already made.\n\n### When to use something smaller\n\n- Use a direct answer or primary documentation for factual lookups.\n- Run an experiment when the choice is cheap and reversible.\n- Use `--quick` when cross-examination will not change the outcome.\n- Use `--duo` when one tension matters more than broad coverage.\n- Do not convene a council to manufacture support for a preferred answer.\n\n## Choose a mode\n\n| Mode | Shape | Use it when |\n|---|---|---|\n| Full | Independent analysis, cross-examination, final stance, synthesis | Stakes are high and competing frames need contact |\n| Quick | Restate, rapid analysis, final positions | The decision needs breadth but not a full adversarial round |\n| Duo | Opening positions, direct response, final statements | One polarity defines the decision |\n\n\n\n```text\n/council --full What is the right pricing model?\n/council --quick --triad shipping Should we release today?\n/council --duo --members torvalds,ada Is this abstraction worth it?\n```\n\nNamed triads select three relevant lenses without assembling the whole council:\n\n```text\n/council --triad strategy Where is our defensible advantage?\n/council --triad risk What could make this launch irreversible?\n/council --triad ai-product Which capability belongs in the product?\n```\n\nAvailable domains include `architecture`, `strategy`, `ethics`, `debugging`, `risk`,\n`shipping`, `product`, `founder`, `ai`, `ai-product`, `ai-safety`, `decision`, `systems`,\n`uncertainty`, `design`, `economics`, and `bias`. The canonical routing table lives in\n[`SKILL.md`](SKILL.md).\n\n## What the protocol protects\n\nThe full protocol has a fixed round budget. Members first restate the problem, analyze\nblind, cross-examine other positions, declare a final stance, and then enter synthesis.\nEnforcement checks look for premature agreement, repeated claims, missing dissent, and\nunsupported confidence.\n\n\n\nVerdicts lead with what remains unresolved. A recommendation is paired with acceptable\ncompromises, kill criteria, and one concrete next step. When the weighted tally remains\nsplit, the result returns that split to the user instead of turning it into prose consensus.\n\nEvidence labels separate:\n\n- `FACT`: directly supported by supplied or retrieved evidence;\n- `INFERENCE`: follows from evidence but is not directly observed;\n- `ASSUMPTION`: required for the argument and still unverified;\n- `UNKNOWN`: missing information that could change the decision.\n\n## Track the outcome\n\nA verdict is useful only if it can be checked later. Before acting, record the prediction,\nowner, review date, and evidence that would change the recommendation. At the checkpoint,\nmark the result confirmed, revised, reversed, or inconclusive instead of rewriting the\noriginal rationale.\n\n\n\n## The 18 lenses\n\nMembers are analytical instruments, not impersonation claims. Each persona has a grounding\nprotocol, a method, known blind spots, and a structured response contract.\n\n\n\n| Member | Primary lens | Useful counterweight |\n|---|---|---|\n| Aristotle | Categories and structure | Lao Tzu on emergence and excess structure |\n| Socrates | Assumption destruction | Feynman on reconstruction from first principles |\n| Sun Tzu | Terrain and adversarial strategy | Aurelius on internal control and moral cost |\n| Ada Lovelace | Formal systems and abstraction | Machiavelli on incentives and informal power |\n| Marcus Aurelius | Resilience and moral clarity | Sun Tzu on external competition |\n| Machiavelli | Power and incentives | Ada on formal consistency |\n| Lao Tzu | Non-action and emergence | Aristotle on explicit categories |\n| Richard Feynman | Explanation and empirical debugging | Socrates on the premise itself |\n| Linus Torvalds | Shipping and maintainability | Meadows on system-level consequences |\n| Miyamoto Musashi | Timing and decisive action | Torvalds on acting before the ideal moment |\n| Alan Watts | Reframing and false problems | Torvalds on concrete implementation |\n| Andrej Karpathy | Empirical ML behavior | Sutskever on frontier risk |\n| Ilya Sutskever | Scaling and AI safety | Karpathy on observation and iteration |\n| Daniel Kahneman | Cognitive bias | Feynman on explicit causal reasoning |\n| Donella Meadows | Feedback loops and high-impact interventions | Torvalds on local fixes |\n| Charlie Munger | Inversion and model lattices | Aristotle on single-system classification |\n| Nassim Taleb | Tail risk and fragility | Karpathy on smooth empirical trends |\n| Dieter Rams | User clarity and restraint | Ada on what can be formalized |\n\nThe repository stores each member contract under [`agents/`](agents/). Custom panels can\nuse `--members`, named triads, or the `classic`, `exploration-orthogonal`, and\n`execution-lean` profiles documented in [`SKILL.md`](SKILL.md).\n\n## Multi-provider routing\n\nThe detection script checks which supported providers are available, then the coordinator\ndistributes seats across them. Polarity pairs are separated when possible so a single model\nfamily does not play both sides of a disagreement.\n\n\n\n| Provider path | Detection |\n|---|---|\n| Native host subagents | Available through the active supported client |\n| OpenAI | `codex` executable |\n| Google | `gemini` executable |\n| Ollama | `ollama` executable |\n| NVIDIA NIM | `NVIDIA_API_KEY` |\n| Cursor | `cursor-agent` executable or configured login |\n\nPreview routing without running a council:\n\n```text\n/council --dry-route --triad decision Should we accept this acquisition offer?\n```\n\nUse `--no-auto-route` to keep native-host defaults. Use `--models\n ![\"Council](\"assets/star-history-dark.svg\")
\n