skills.rest
eytanlevit
eytanlevitCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

small-council

Coordinate five frontier LLMs through anonymous ranking and Claude Opus 4.5 synthesis.

4|Updated Dec 20, 2025
Search Tags:#architecture#CLI#tmux#consensus#code-review#deliberation#multimodel
One-click install
npx skills add https://github.com/eytanlevit/small-council --skill small-council
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: small-council
Source: https://github.com/eytanlevit/small-council/tree/main/skill
Command: npx skills add https://github.com/eytanlevit/small-council --skill small-council

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill helps coding agents obtain multiple expert perspectives from frontier AI models, reducing single-model bias and improving outcomes for code reviews, architecture decisions, debugging challenges, and design decisions.

Core Features & Use Cases

  • Multi-model deliberation: Get independent answers from five frontier models (GPT-5.2, GPT-5.2-pro, Gemini 3 Pro, Claude Sonnet 4, Grok 4) and compare insights.
  • Anonymous peer ranking & synthesis: Stage 2 rankings are performed anonymously, followed by Stage 3 consensus synthesis by a designated chair model.
  • Versatile use cases: Ideal for architecture decisions, code reviews, complex debugging, design-pattern evaluations, and performance optimization, where diverse perspectives improve quality.

Quick Start

  • Prerequisites: OPENROUTER_API_KEY set in ~/.claude/skills/small-council/.env and the Small Council CLI installed (uv tool install small-council)
  • Trigger a consultation by providing a clear prompt, e.g. "Review this microservices architecture for scalability" or "Provide five expert opinions on this code snippet"
  • The skill runs in a tmux session to preserve progress and returns the synthesized consensus

Frequently Asked Questions about small-council

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How can I get multiple AI perspectives on code architecture decisions?▼

Multi-model deliberation gathers independent answers from five frontier models—GPT-5.2, GPT-5.2-pro, Gemini 3 Pro, Claude Sonnet 4, and Grok 4—then synthesizes consensus. This reduces single-model bias for architectural choices, code reviews, and debugging challenges where diverse expert perspectives improve outcomes.

What does a consensus-based code review workflow look like?▼

A three-stage workflow coordinates deliberation: Stage 1 collects independent answers from five models, Stage 2 ranks responses anonymously by peer models, and Stage 3 uses Claude Opus 4.5 as chairman to synthesize a final consensus answer on your code or design question.

Do I need special setup to run multi-model deliberations?▼

Yes. Set OPENROUTER_API_KEY in ~/.claude/skills/small-council/.env, install the Small Council CLI via `uv tool install small-council`, and run the skill in a tmux session to preserve progress and orchestrate model coordination.

When should I use consensus deliberation instead of a single model?▼

Use consensus deliberation for high-stakes decisions: architecture reviews, complex debugging, design-pattern exploration, and performance optimization where multiple expert viewpoints catch blind spots and reduce error compared to one model's output.

Can I use this for performance optimization reviews?▼

Yes. The skill handles performance optimization as a use case within its versatile scope, letting you submit code or design questions to five models and receive a synthesized consensus on optimization strategies and trade-offs.

Related Skills

Explore More Software Engineering Skills→
AtifAssariAtifAssari
4

1password

Manage sensitive credentials with the 1Password CLI in isolated tmux sessions.

Community
Advanced
pgm1980pgm1980

reflexion:critique

Coordinate multi-perspective critique and synthesize consensus reports on completed work.

Community
Advanced
AlvaroBianoAlvaroBiano

claude-code

Automate coding tasks via Claude Code CLI for edits, refactors, and PR workflows.

Community
Advanced
johnlindquistjohnlindquist
24

council

Coordinate parallel consultations across Claude, Codex, and Gemini for structured comparison.

Community
Intermediate