Claude Council

Claude Council

An LLM Council for Claude Code — with zero API keys.

Ask a question and a council of independent AI advisors debates it: each member answers in parallel from a completely isolated context, the members anonymously peer-review and rank each other's answers, and a chairman synthesizes the final recommendation — with attribution, consensus, and dissent.

The twist: every council member is a Claude subagent running on the Claude Code subscription you already have. No OpenAI key, no Gemini key, no separate billing. If Claude Code runs on your machine, the council runs.

Why

karpathy/llm-council and gcpdev/llm-council-skill are great — if you have API keys for multiple vendors. Lots of people just have a Claude Pro/Max subscription and no API keys at all. Claude Council recreates the council experience with what actually makes opinions independent:

- Fresh contexts — members can't see your conversation or each other's first drafts, so they can't anchor on each other
- Distinct personas — The Architect, The Skeptic, The Pragmatist, The Researcher (or any custom lineup you ask for)
- Anonymized peer review — answers are judged as "Response A/B/C/D", so nobody defers to a name
- Repo awareness — members run in your working directory and can actually read your code before opining, something an external-API council can never do

Honest caveat: all members are Claude, so you get independence from isolation and personas, not cross-vendor diversity. It won't tell you "what GPT thinks" — it gives you four genuinely independent, adversarial Claude takes plus a ranked synthesis.

How it works

1. Opinions — N members (default 4) answer your question in parallel, each through its persona lens.
2. Peer review — each member critiques all answers anonymized as Response…