llm-council

Deliberates complex questions through a three-stage multi-model council with anonymized peer ranking.

Updated Mar 4, 2026
One-click install
npx skills add https://github.com/gmolike/Claude-Template --skill llm-council-gmolike
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: llm-council
Source: https://github.com/gmolike/Claude-Template/tree/main/.claude/skills/llm-council
Command: npx skills add https://github.com/gmolike/Claude-Template --skill llm-council-gmolike

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Single-model answers to high-stakes or wide-solution-space questions carry a single point of failure: one model's confident judgment may miss tradeoffs or fail in ways it cannot self-detect. This Skill convenes a council of tier-diverse Claude models that answer independently, blindly rank each other's anonymized answers, and let a chairman synthesize a final answer with disagreements made explicit. ## Core Features & Use Cases - Three-Stage Deliberation: Independent first opinions, anonymized peer ranking (Response A/B/C), and chairman synthesis in one orchestrated workflow. - Tier-Diverse Line-up: Opus 5, Opus 4.8, and Sonnet 5 produce diverse failure modes so the ranking stage catches what any single model would miss. - Disagreement Mapping: The chairman explicitly notes where the council disagreed, showing which parts of the answer are solid and which are judgment calls. - Use Case: Facing an architecture decision with several defensible approaches — such as choosing a state management strategy for a large React app — convene the council to get independent proposals, blind cross-evaluation, and a synthesized recommendation. ## Quick Start Ask the AI to convene the LLM council on your specific architecture or design decision, for example whether to adopt event sourcing for the billing service.

Frequently Asked Questions about llm-council

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

FAQPage Schema
How do I get multiple AI models to evaluate a decision?

Convene an LLM council: each member answers the question independently, then blindly ranks the other members' anonymized answers, and a chairman model synthesizes the final response. This structure catches errors a single model would miss.

When should I use multi-model deliberation instead of one model?

Use it for wide-solution-space decisions, high-stakes or irreversible changes, and contested tradeoffs where one model's judgment is a single point of failure. Skip it for trivial, fully specified, or easily reversible tasks.

Why are council answers anonymized before peer ranking?

Anonymization strips model identity so members rank answers as Response A/B/C purely on accuracy and insight. This prevents favoritism by brand or tier and stops a member from recognizing its own answer.

What models make up the LLM council line-up?

The council seats Opus 5, Opus 4.8, and Sonnet 5, with Opus 5 as chairman. The tier diversity produces different failure modes, which makes the peer-ranking stage informative rather than redundant.

What are the limitations of multi-model council deliberation?

It adds orchestration overhead and cost, so it is wasted on trivial or reversible work. It also cannot replace research: questions answerable from primary sources should be verified against ground truth first.