LLM Council

Install

npx skillfoxx add skills/llm-council

npx skills add aiwithremy/claude-skills-llm-council --skill llm-council -a claude-code -y

npx skills add aiwithremy/claude-skills-llm-council --skill llm-council -a cursor -y

npx skills add aiwithremy/claude-skills-llm-council --skill llm-council -a github-copilot -y

npx skills add aiwithremy/claude-skills-llm-council --skill llm-council -a codex -y

npx skills add aiwithremy/claude-skills-llm-council --skill llm-council -a gemini-cli -y

npx skills add aiwithremy/claude-skills-llm-council --skill llm-council -a cline -y

npx skills add aiwithremy/claude-skills-llm-council --skill llm-council -a roo -y

npx skills add aiwithremy/claude-skills-llm-council --skill llm-council -a opencode -y

This is third-party code. Review the repository files before installing.

What it does

The skill turns one question into a five-angle review. On a trigger phrase the agent gathers context from workspace files, writes a neutral framing and spawns five advisors as parallel sub-agents: a contrarian hunts for flaws, a first-principles thinker reframes the task, an expansionist looks at the upside, an outsider reacts without your context, an executor checks feasibility. Their answers are anonymized and the advisors peer-review each other. A chairman then produces a structured verdict: where the council agrees, where it clashes, which blind spots surfaced, a recommendation and a first step. The method follows Andrej Karpathy's LLM Council idea, but sub-agents with different thinking lenses stand in for different models.

Who it is for. For founders, product managers and marketers who need to pressure-test a high-stakes decision from several angles at once.

Example request

Council this: launch a 97 dollar workshop or a 497 dollar course

Limitations

The skill gives judgment, not fact-checking: the verdict is only as good as the context you provide. It spawns several sub-agents per query, so it uses more tokens than a plain answer. It works where the agent can create sub-agents and read project files, such as Claude Code and claude.ai. The repository states no license.

How to disable. Remove the llm-council skill folder from the agent's skills directory, for example ~/.claude/skills/.

Security check

• Scans the workspace and reads project context files
• Spawns several sub-agents per query and uses more tokens

README in short

The README explains that the skill runs a question through five independent advisors who peer-review each other anonymously, after which a chairman gives the final answer. It suits high-stakes decisions and is not meant for simple one-answer questions or creative tasks. You install it by asking the agent to fetch SKILL.md from the repository, or by downloading the file into the skills directory. It triggers on phrases like 'council this' or 'pressure-test this'. The method is adapted from Andrej Karpathy's LLM Council; the skill is by Ole Lehmann.

SKILL.md

---
name: llm-council
description: "Run any question, idea, or decision through a council of 5 AI advisors who independently analyze it, peer-review each other anonymously, and synthesize a final verdict. Based on Karpathy's LLM Council methodology. MANDATORY TRIGGERS: 'council this', 'run the council', 'war room this', 'pressure-test this', 'stress-test this', 'debate this'."
---

# LLM Council

You ask one AI a question, you get one answer. That answer might be great. It might be mid. You have no way to tell because you only saw one perspective.

The council fixes this. It runs your question through 5 independent advisors, each thinking from a fundamentally different angle. Then they review each other's work. Then a chairman synthesizes everything into a final recommendation that tells you where the advisors agree, where they clash, and what you should actually do.

This is adapted from Andrej Karpathy's LLM Council. He dispatches queries to multiple models, has them peer-review each other anonymously, then a chairman produces the final answer. We do the same thing inside Claude using sub-agents with different thinking lenses instead of different models.

Related