LLM Council Skill for Claude Code

Part of the Automators+ skills library -- Claude Code skills shared exclusively with the Automators+ community.

Stop Claude from being a yes man.

Ask Claude "should I launch this?" and it finds five reasons you should. Ask "is this a bad idea?" and it finds five reasons it is. A Stanford study found AI assistants agree with your decisions about 49% more often than a human would. This skill fixes that by running any decision through five advisors who argue, grade each other anonymously, and hand you one verdict you can trust.

What You Get

• A drop-in Claude Code skill at .claude/skills/llm-council/SKILL.md
• Five advisors with different jobs (Contrarian, First-Principles, Expansionist, Outsider, Executor) who answer independently
• Anonymous peer review so the model grades honestly instead of defending itself
• A chairman that commits to one verdict under 250 words: the call, the strongest reason, the biggest risk, the next step in 7 days
• No API keys, no MCP servers, no scripts -- runs entirely with Claude Code subagents

Install the Skill

Option 1: URL install

claude skill add --url https://raw.githubusercontent.com/automatorsplus/llm-council-skill/main/.claude/skills/llm-council/SKILL.md

Option 2: Manual copy

Copy .claude/skills/llm-council/ into your project's .claude/skills/ folder.

Prerequisites

None. Any recent version of Claude Code that can spawn subagents will run it.

Try It

Once installed, try one of these in Claude Code:

• council this: should I drop my free tier and go paid-only?
• pressure-test this: I want to raise my retainer from $2.5k to $4k a month
• run the council on whether I should hire a developer now or in three months
• war room this: my new landing page headline -- "AI that runs your back office"

Best for ambiguous, high-stakes calls (pricing, positioning, hiring, pivots, copy). Skip it for factual questions, drafting, or summarising.

How It Works

See .claude/skills/llm-council/SKILL.md for the full skill definition.

Built on Andrej Karpathy's LLM Council idea, adapted to run inside a single Claude Code session with subagents instead of multiple model providers.

License

MIT

---

Shared with the Automators+ community