Query multiple LLMs, have them peer-review and rank each other's responses, then synthesize a final answer via a chairman model. Based on karpathy/llm-council.
Trigger
User invokes /council with a question or topic.
How It Works
Three stages, all via OpenRouter API:
Stage 1 — First Opinions: GPT-5.1, Gemini 3 Pro, Claude Sonnet 4.5, and Grok 4 each answer the query independently (in parallel).
Stage 2 — Peer Review: Each model receives all anonymized responses and ranks them by quality. Rankings are aggregated.
Stage 3 — Chairman Synthesis: Gemini 3 Pro (chairman) reads all responses + all rankings and produces a single final answer representing the council's collective wisdom.
Usage
Run the council script:
OPENROUTER_API_KEY="$OPENROUTER_API_KEY" uv run --with httpx python3 .claude/skills/llm-council/council.py "YOUR QUESTION HERE"
The script:
Prints truncated Stage 1 responses (first 500 chars each)
Prints parsed rankings from each model in Stage 2
Prints the full chairman synthesis in Stage 3
Saves the complete session to .claude/skills/llm-council/last_session.json
Workflow
User asks /council "question"
Run the command above with the user's question
Wait for it to complete (~60-120s depending on model latency)
Present the Stage 3 final answer to the user
If the user wants to see individual model responses or rankings, read last_session.json
Config
Edit council.py to change:
COUNCIL_MODELS — which models sit on the council
CHAIRMAN_MODEL — which model synthesizes the final answer
Models use OpenRouter identifiers (e.g., openai/gpt-5.1)
Requirements
OPENROUTER_API_KEY in environment (set in ~/.zshrc)
uv installed (manages Python + dependencies)
OpenRouter account with credits loaded
1---2name: llm-council3description: LLM Council4---5# LLM Council67Query multiple LLMs, have them peer-review and rank each other's responses, then synthesize a final answer via a chairman model. Based on [karpathy/llm-council](https://github.com/karpathy/llm-council).89## Trigger1011User invokes `/council` with a question or topic.1213## How It Works1415Three stages, all via OpenRouter API:16171. **Stage 1 — First Opinions**: GPT-5.1, Gemini 3 Pro, Claude Sonnet 4.5, and Grok 4 each answer the query independently (in parallel).182. **Stage 2 — Peer Review**: Each model receives all anonymized responses and ranks them by quality. Rankings are aggregated.193. **Stage 3 — Chairman Synthesis**: Gemini 3 Pro (chairman) reads all responses + all rankings and produces a single final answer representing the council's collective wisdom.2021## Usage2223Run the council script:2425```bash26OPENROUTER_API_KEY="$OPENROUTER_API_KEY" uv run --with httpx python3 .claude/skills/llm-council/council.py "YOUR QUESTION HERE"27```2829The script:30- Prints truncated Stage 1 responses (first 500 chars each)31- Prints parsed rankings from each model in Stage 232- Prints the full chairman synthesis in Stage 333- Saves the complete session to `.claude/skills/llm-council/last_session.json`3435## Workflow36371. User asks `/council "question"`382. Run the command above with the user's question393. Wait for it to complete (~60-120s depending on model latency)404. Present the **Stage 3 final answer** to the user415. If the user wants to see individual model responses or rankings, read `last_session.json`4243## Config4445Edit `council.py` to change:46- `COUNCIL_MODELS` — which models sit on the council47- `CHAIRMAN_MODEL` — which model synthesizes the final answer48- Models use OpenRouter identifiers (e.g., `openai/gpt-5.1`)4950## Requirements5152- `OPENROUTER_API_KEY` in environment (set in `~/.zshrc`)53- `uv` installed (manages Python + dependencies)54- OpenRouter account with credits loaded
Run npx skillmds@latest add cdeistopened/llm-council in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.
LLM Council It is listed under AI & ML on SkillMD.
This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.
This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.
Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.
cdeistopened (@cdeistopened) published this skill. Their other Agent Skills are listed on their SkillMD profile.
cdeistopened