The Council — Painel de Validação de Ideias com Multi-Agentes
lizandroambriz-ops/council
Painel de críticos especializados para validar ideias com pesquisa web
Ver no GitHubSistema de validação de ideias que usa múltiplos agentes especializados (Adversário, Pré-Mortem, Realista de Recursos) para stress-testar propostas com pesquisa web em tempo real. Sintetiza um veredito estruturado com scores de confiança, condições bloqueantes e ações com prazo definido, ideal para validar projetos antes de investir recursos.
Instalação
npx convez-ai lizandroambriz-ops/council1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Validar viabilidade de uma ideia ou projeto antes de iniciar desenvolvimento
- Comparar duas propostas alternativas (A vs. B) para tomar decisão informada
- Identificar riscos, modos de falha e objeções antes de executar um plano
Veredito

The Council é uma ferramenta robusta e bem pensada para validação estruturada de ideias, combinando múltiplas perspectivas críticas com pesquisa ativa. Recomendado para times que precisam tomar decisões de alto risco ou alocar recursos de forma inteligente.
README
A tradução para o português deste README ainda está na fila.
The Council
A multi-agent idea-validation panel for Claude Code. Three permanent critics — The Adversary, The Pre-Mortem, The Resource Realist — plus domain specialists cast on the fly independently research and stress-test an idea with live web search, and a Chair synthesises a structured verdict with confidence scores, blocking conditions, and time-bound next actions.
Two ways to run: validate one idea or compare two ideas (A vs. B). Once it understands your request, the Council recommends an effort level (Quick / Standard / Deep) and the best model, shows a token + wall-time estimate before the expensive phase, then runs. A logbook keeps past sessions with outcome tracking, and you can follow up over any completed verdict.
Prerequisites
- Claude Code
- Node.js (v18+). All deterministic work — scoring, checkpoint I/O, rendering, config — runs through a tested Node helper (
lib/cli.mjs); the skill never re-implements it in prose.
Install
The Council ships as a Claude Code plugin. There are two ways to get it; the plugin route is recommended.
Option A — Install as a plugin (recommended)
This repo is also a plugin marketplace, so you can install without cloning. In Claude Code:
/plugin marketplace add lizandroambriz-ops/council
/plugin install council@council
Then run it with the namespaced command:
/council:council
Plugin skills are always namespaced (/<plugin>:<skill>), which is why the command is /council:council. Updates are a single /plugin marketplace update away.
Option B — Install as a standalone skill (clone & copy)
Prefer to vendor the skill directly (e.g. commit it into a project)? Clone the repo and run the bundled installer:
git clone https://github.com/lizandroambriz-ops/council.git
cd council
./install.sh # all projects → ~/.claude/skills/council/
./install.sh --project # current project → ./.claude/skills/council/
Or copy it by hand — the installer just places the skill folder:
# personal (all projects)
cp -r skills/council ~/.claude/skills/council
# or per-project (commit it alongside your code)
mkdir -p .claude/skills && cp -r skills/council .claude/skills/council
Restart Claude Code so it discovers the skill, then run /council (standalone installs are not namespaced).
The same
SKILL.mdworks in both modes: it locates its helper CLI via${CLAUDE_PLUGIN_ROOT:-.claude}/skills/council/lib/cli.mjs, which resolves to the plugin directory when installed as a plugin and to.claude/skills/council/when installed standalone.
Verify the install (optional):
node --test skills/council/lib/*.test.mjs # expect: 82 pass
Usage
Start it in Claude Code:
- Plugin install:
/council:council - Standalone install:
/council
You'll get a short menu. Pick what you want to do:
| Mode | What it does |
|---|---|
| Validate an idea | Stress-test a single idea or plan and get a scored verdict with the strongest objections, failure modes, and next actions. |
| Compare two ideas | Put two options head-to-head (A vs. B) and get a reasoned recommendation instead of a coin flip. |
| Past sessions | Reopen the logbook of previous verdicts, record how each one turned out, or follow up with new questions over a completed verdict. |
How a run works
- Describe your idea in a sentence — the Council asks a few sharp clarifying questions first.
- Pick the depth — it recommends an effort level and you can change it:
- Quick — a fast gut-check.
- Standard — the balanced default.
- Deep — adds cross-examination between the experts for high-stakes calls.
- See the cost before you spend — a token + wall-time estimate appears at the launch gate, so you can abort before the expensive research phase.
- It runs — expert personas research and stress-test your idea in parallel; the Chair synthesises the verdict. If a run is interrupted, it auto-resumes where it left off.
- Read the result — every run produces a Markdown transcript and a self-contained HTML report you can open or share.
Good things to know
- Where your data lives: checkpoints, transcripts, and HTML reports are written to
.claude/council/in your current project — never inside the plugin, so they survive updates. Delete that folder any time to clear your history. - Model choice: the heavy thinking runs on your chosen primary model (Opus 4.7, Sonnet 4.6, or Haiku 4.5 — recommended per query, overridable). Cheap utility steps always run on Haiku 4.5, so picking a bigger model only spends where it actually changes the verdict.
- Tip: press Shift+Tab to enable auto-accept mode before a run — the Council makes many small local calls, and approving each one slows things down.
Layout
.claude-plugin/
plugin.json plugin manifest
marketplace.json marketplace catalog (this repo is its own marketplace)
skills/council/
SKILL.md router — startup, resume detection, menu, phase routing
phases/ one file per phase (prebrief, casting, research, cross_exam,
chair, resolve, export, followup, logbook, compare, settings)
lib/ tested Node helpers (*.mjs) + their tests (*.test.mjs)
install.sh standalone installer (Option B)
Tests
node --test skills/council/lib/*.test.mjs
License
MIT © 2026 Lizandro Ambriz — you're free to use, modify, and redistribute it. Attribution appreciated but not required.