The Council — Conselho Consultivo Multi-Modelo
DantesPeak85/the-council
Segundas opiniões de Codex e Gemini em paralelo, com contexto completo
Ver no GitHubSkill para Claude Code que convida OpenAI Codex e Google Gemini como conselheiros técnicos paralelos. Ambos analisam seu projeto em sandboxes isoladas (read-only) e síntesizam recomendações unificadas. Inclui loop de aprendizado que registra insights para futuras sessões.
Instalação
npx convez-ai DantesPeak85/the-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
- Code review e análise de arquitetura com múltiplas perspectivas
- Debug colaborativo entre três modelos de IA independentes
- Validação de decisões técnicas high-stakes antes de implementação
Veredito

Ferramenta ambiciosa que requer múltiplas credenciais CLI e atenção a segurança (versions <1.3.0 têm falha crítica em permissões). Valor real para equipes que querem síntese técnica genuína, mas curva de setup é íngreme e dependência de terceiros é alta.
README
A tradução para o português deste README ainda está na fila.
The Council
Parallel minds. Singular clarity.
Security advisory (2026-05-25): Versions 1.2.0 and 1.2.1 do NOT enforce read-only on the Gemini advisor as their docs claim —
agy's--dangerously-skip-permissionsauto-approves tool calls rather than withholding them, and a 2026-05-24 incident saw agy ghost-write 8 files into a project directory. Upgrade to 1.3.0 or later for real OS-level sandbox enforcement (sandbox-execon macOS) and a pre/post-invocation diff safety net. See SKILL.md → Permissions and Safety for the full incident note.
A Claude Code skill that convenes OpenAI Codex and Google Gemini as an advisory board. Both run in parallel via their CLIs, with full project context, and return independent analyses that Claude synthesizes into a unified recommendation.
What It Does
When you say "ask the council" or request a second opinion, Claude:
- Syncs your project context (CLAUDE.md) to both advisors
- Composes a prompt using the appropriate template (code review, architecture, debugging, or general)
- Invokes Codex and Gemini in parallel, both in read-only sandboxes
- Synthesizes the responses — highlighting consensus, divergence, and Claude's own recommendation
Both advisors run at maximum capability (Codex with xhigh reasoning effort, Gemini with HIGH thinking level) and have full codebase access in read-only mode.
Learning Loop
After each council session, Claude reflects on what the advisors revealed — gaps, blind spots, or better approaches — and generalizes those learnings into CLAUDE.md and AGENTS.md. This means future sessions in the same repository start with lessons from previous council deliberations already loaded.
Installation
Via Claude Code Plugin Marketplace
/plugin marketplace add DantesPeak85/the-council
/plugin install the-council@the-council-marketplace
Manual Installation
Clone this repo and register it as a local plugin in your Claude Code configuration.
Prerequisites
You need at least one of these CLIs installed and authenticated:
OpenAI Codex CLI
npm i -g @openai/codex
codex auth
Google Gemini CLI
npm i -g @google/gemini-cli
gemini # Authenticates via OAuth on first run
Enable experimental plan mode in ~/.gemini/settings.json:
{ "experimental": { "plan": true } }
Graceful Degradation
The Council runs a preflight check on first invocation. If only one CLI is available, it operates in single-advisor mode and notes this in the synthesis. If neither is available, it shows installation instructions.
Usage
Just tell Claude:
- "Ask the council to review this PR"
- "Get a second opinion on this architecture"
- "Council review" on any code changes
- "Debug this with the council"
The council is also invoked proactively during high-stakes architectural decisions or security-sensitive code reviews.
Recommended Configuration
For maximum advisor capability:
Codex (~/.codex/config.toml):
model_reasoning_effort = "xhigh"
Gemini (~/.gemini/settings.json):
{
"modelConfigs": {
"customAliases": {
"council": {
"modelConfig": {
"model": "gemini-3-pro-preview",
"generateContentConfig": {
"thinkingConfig": { "thinkingLevel": "HIGH" }
}
}
}
}
}
}
Environment Variables
| Variable | Default | Description |
|---|---|---|
CODEX_MODEL | auto (CLI default) | Model for Codex advisor |
GEMINI_MODEL | gemini-3-pro-preview | Model for Gemini advisor |
COUNCIL_TIMEOUT | 300 | Max seconds per advisor |
GEMINI_MAX_TURNS | 50 | Max Gemini session turns |
Safety
Both advisors run in read-only mode and cannot modify your project:
- Codex:
--sandbox read-only - Gemini:
--approval-mode plan+--policydeny list (write/execute tools removed from model memory)
Author
Created by Tom Wolinitz with Claude Code
More projects at github.com/DantesPeak85
License
MIT