Council

Five AI advisors for Claude Code and Codex. Catch the problems you didn't think to look for — before you build.

npm install -g @kidoncio/council

"The argument you had before building is the one you don't have in production."

Why Council · The Council · Getting Started · Skills · Output Files

Why Council

You've shipped a feature that looked solid in the plan — and then watched it fall apart for a reason nobody thought to check.

The auth flow that seemed fine until a security engineer pointed out it leaks session tokens. The architecture that worked great until someone asked what happens at 10x load. The product decision that made sense until a user couldn't figure out what the feature was for.

Council gives you five permanent advisors — each with a distinct worldview and a mandate to disagree — embedded directly in Claude Code and Codex as skills. They review your plan before you build it: independently, then against each other, then together.

The problems they surface are the ones that cost the most to fix after the fact. That's the point.

The Council

Five advisors. Five lenses. All permanently disagreeing with at least one of the others.

| Advisor | Role | Their question | | ------------- | ------------------- | --------------------------------------------------------------------- | | TURING | Pragmatist Engineer | "What happens when this breaks at 3am?" | | LOVELACE | Product Strategist | "Are we solving the right problem?" | | TORVALDS | Security Engineer | "What's the worst thing a malicious user can do with this?" | | FOWLER | Refactoring Expert | "How many places does the next change have to touch?" | | CASSANDRA | Pre-Mortem Strategist | "What will be the first line of the post-mortem when this fails in 6 months?" |

Each advisor has a philosophy, a blind spot, and a debate style. They produce independent reports, argue with each other, and converge on a verdict only when the argument is actually settled.

Getting Started

npm install -g @kidoncio/council

postinstall installs skills for Claude and Codex.

• Claude skills: ~/.claude/skills/council-*
• Codex skills: ~/.agents/skills/council-*

Confirm it worked:

council list
# → Lists all installed council skills

Then open Claude Code or Codex and run your first plan with the council-plan skill.

Skills

Setup

| Command | What it does | | ---------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | council-init | Reads your project's existing files (CLAUDE.md, AGENTS.md, README.md, package.json, etc.) and asks a few targeted questions to write .council/PROJECT.md — a stable snapshot of your stack, structure, and conventions that every subsequent council agent uses as shared context. Run once per project. |

Planning

The skills are designed to run in sequence — but each one is independently useful. You end up with a plan your whole team can poke holes in, backed by research and a full adversarial review.

| Command | What it does | | ------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------- | | council-brainstorming | Brainstorm a feature before committing to a plan. Weighs 2-3 approaches with tradeoffs, runs sectioned design with per-section approval, and writes BRAINSTORMING.md. Hands off to /plan or /research (user picks). | | council-research | Spawn parallel research agents on a feature or problem. Produces RESEARCH.md without committing to a plan. Reused automatically by /plan. | | council-plan | 5-phase structured planning: research → technical sketch → UX mapping → task plan → adversarial council review. Produces a complete file set in .council/[feature]/. | | council-execute | Reads PLAN.md, implements tasks in dependency order, and updates ROADMAP.md after each one. Pauses for input when blocked or ambiguous. |

council-plan detects and reuses an existing RESEARCH.md automatically — no duplicate work. If .council/PROJECT.md exists, it is injected into every agent's context automatically.

Council Reviews

Convene one advisor or all five — your choice.

| Command | Advisor | What they focus on | | ---------------------------- | --------- | ----------------------------------------------------------- | | council-review | All five | Independent reports → debate → unified verdict | | council-senior-engineer | TURING | Operational complexity, blast radius, maintainability, naming, test quality, convention drift | | council-security-engineer | TORVALDS | Attack surface, data exposure, specific CVE classes | | council-product-strategy | LOVELACE | User outcomes, delivery speed, the right problem | | council-refactoring | FOWLER | Code smells, technical debt, refactoring discipline, pattern fit | | council-pre-mortem | CASSANDRA | Failure scenarios, leading indicators, second-order effects |

How the Review Works

council-review runs a 3-phase adversarial process:

Phase 1 — Independent reports
Each advisor reviews the plan separately, with no knowledge of the others' findings.

Phase 2 — Debate
Advisors challenge each other's conclusions directly.
TORVALDS flags an auth bypass; LOVELACE argues it's out of scope for v1;
the verdict names the trade-off and forces you to decide.
Consensus achieved too easily is a failure of the process.

Phase 3 — Unified verdict
A synthesized report that names the real trade-offs and what you must decide before committing.

The single-advisor skills skip the debate and give you one sharp opinion fast.

Output Files

Plans are written to .council/[feature-slug]/ inside your project. Everything is plaintext and lives in your repo — version it, diff it, hand it to a teammate. No external state, no accounts, no dashboard.

.council/
├── PROJECT.md           ← project snapshot: stack, structure, conventions (from /init)
└── my-feature/
    ├── BRAINSTORMING.md     ← goals, chosen approach, approved design (from /brainstorming)
    ├── RESEARCH.md          ← synthesized research (from /research or /plan)
    ├── TECHNICAL_SKETCH.md  ← feasibility constraints and chosen direction (from /plan)
    ├── UX.md                ← personas and user journeys (informed by technical constraints)
    ├── PLAN.md              ← tasks with BDD acceptance criteria
    ├── ROADMAP.md           ← live progress tracker (updated by /execute)
    ├── SUMMARY_OF_COUNCIL.md
    ├── research/            ← individual agent reports (from /research)
    │   ├── market-solutions.md
    │   ├── ux-patterns.md
    │   ├── technical-approaches.md
    │   ├── failure-modes.md
    │   └── security-and-compliance.md
    └── council/
        ├── TURING.md
        ├── LOVELACE.md
        ├── TORVALDS.md
        ├── FOWLER.md
        └── CASSANDRA.md

Install

Global — available in every project

npm install -g @kidoncio/council

postinstall installs skills in Claude/Codex paths.

Optional:

• Install only for Claude: council install --target=claude
• Install only for Codex: council install --target=codex

To uninstall:

council uninstall
# or
npm uninstall -g @kidoncio/council

Project — scoped to one repo

Install into project-local skills paths: .claude/skills/council-* and .agents/skills/council-*. Commit them to version-control if you want project-scoped behavior.

npx @kidoncio/council install --local

# only Claude (local)
npx @kidoncio/council install --local --target=claude

# only Codex (local)
npx @kidoncio/council install --local --target=codex

To uninstall:

npx @kidoncio/council uninstall --local

curl

curl -fsSL https://raw.githubusercontent.com/kidoncio/council/main/install.sh | bash

Clone

git clone https://github.com/kidoncio/council.git
cd council
./install.sh

To uninstall:

./uninstall.sh

Requirements

• Claude Code CLI or Codex
• Node.js 18+
• curl (for the curl install method)

License

MIT

Five advisors. One plan. Ship with confidence.