llm-council

Orchestrate four-model Claude deliberation with anonymized peer-review and president synthesis.

10|3|Updated May 26, 2026

Search Tags:#workflow #llm #council #claude-code #deliberation #anonymized-review

Ready to run

npx skills add https://github.com/agentik-os/OmegaOS --skill llm-council-agentik-os

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: llm-council
Source: https://github.com/agentik-os/OmegaOS/tree/main/skills/llm-council
Command: npx skills add https://github.com/agentik-os/OmegaOS --skill llm-council-agentik-os

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Convene a multi-model Claude deliberation to answer a single question by having four Claude models respond independently, anonymously peer-review, and have an Opus 4.8 president synthesize a final verdict that preserves dissent.

Core Features & Use Cases

Independent responses from four Claude models on the same prompt.
Anonymized peer-review with blind labeling to surface true differences.
Deterministic in-process synthesis by an Opus president that preserves minority views.
A runnable Workflow script example to reproduce the council end-to-end.
Suitable for high-stakes decisions, architecture choices, and risk-sensitive verification tasks.

Quick Start

Invoke the llm-council workflow in a Claude Code session to run a four-model deliberation and receive a synthesized verdict with dissent preserved.

Frequently Asked Questions about llm-council

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

FAQPage Schema

How do I run a multi-model LLM deliberation for an architecture decision?▼

To run a multi-model LLM deliberation, you invoke a four-model Claude council workflow in-process. Four models independently answer your architecture prompt, anonymously peer-review each other, and an Opus president synthesizes a final verdict preserving dissent.

What is anonymized peer-review in multi-model LLM workflows?▼

Anonymized peer-review in multi-model LLM workflows is a blind labeling process where multiple Claude models evaluate each other's independent responses without knowing the source. This surfaces true differences and biases before final synthesis.

Do I need external API keys to run a Claude deliberation council?▼

No, you do not need external API keys or network calls to run a Claude deliberation council. The workflow requires only in-process Workflow agents for four Claude models to perform the independent responses and synthesis.

When should I use a four-model council instead of a single LLM prompt?▼

Use a four-model council instead of a single LLM prompt for high-stakes decisions, complex verification tasks, and risk-sensitive architecture choices. The deliberation process captures diverse perspectives and preserves minority views that a single prompt would miss.

Does the deliberation council workflow work with Claude Code sessions?▼

Yes, the deliberation council workflow works directly within a Claude Code session. You invoke the workflow in-process to execute the four-model deliberation and receive the synthesized verdict without external dependencies.

What's the best way to preserve dissent in LLM-generated architecture reviews?▼

The best way to preserve dissent in LLM-generated architecture reviews is using a deterministic in-process synthesis by an Opus president model. It synthesizes the final verdict from anonymized peer-reviews while intentionally preserving minority views.