claude-consensus

Multi-model code review, plan review, and general review for Claude Code and Codex

Multiple AI models independently review code, implementation plans, documents, or technical proposals, then converge through structured synthesis and approval rounds. The lead host participates directly: Claude leads in Claude Code, and Codex leads in Codex.

Quick Start

Claude Code

Tell Claude:

"Set up claude-consensus: https://github.com/AltimateAI/claude-consensus"

Manual install:

/plugin marketplace add AltimateAI/claude-consensus
/plugin install consensus
/consensus-setup

Then use:

/code-review "staged changes"
/plan-review "Add caching to the API layer"
/consensus:review docs/architecture.md

Codex

From a local clone, add this repo marketplace to Codex and install the consensus plugin:

git clone https://github.com/AltimateAI/claude-consensus.git

Use the repo marketplace at:

.agents/plugins/marketplace.json

Then configure and run:

consensus-setup
consensus-code-review staged changes
consensus-plan-review Add caching to the API layer
consensus-review docs/architecture.md

Alias skills are also provided for code-review, plan-review, plan-reviwe, and claude-consensus.

Prerequisites

Requirement	Required?	Notes
Claude Code or Codex	Yes	The host environment determines the lead reviewer
Kilo CLI	Recommended	Routes OpenRouter models with one API key
OpenRouter API key	Recommended	Required for Kilo/OpenRouter models
Native CLIs	Optional	Claude supports codex, gemini, and qwen; Codex supports non-Codex native CLIs
CodeRabbit CLI	Optional	Supplementary static analysis for code reviews

Minimal setup: the host lead plus 1 external model is enough for consensus reviews.

Supported Panels

Model	Provider	OpenRouter ID	Claude native	Codex external
Claude	Anthropic	built-in	lead	not used
Codex / GPT	OpenAI	openai/gpt-5.4-codex	codex	lead, not external
Gemini 3.1 Pro	Google	google/gemini-3.1-pro-preview	gemini	gemini
Kimi K2.6	Moonshot	moonshotai/kimi-k2.6	Kilo	Kilo
Grok 4.20	xAI	x-ai/grok-4.20-beta	Kilo	Kilo, disabled by default
MiniMax M2.7	MiniMax	minimax/minimax-m2.7	Kilo	Kilo
GLM-5.1	Zhipu AI	zai-coding-plan/glm-5.1	Kilo	Kilo
Qwen 3.6 Plus	Alibaba	qwen/qwen3.6-plus	qwen or Kilo	Kilo by default
MiMo V2 Pro	Xiaomi	xiaomi/mimo-v2-pro	Kilo	Kilo
DeepSeek V4 Pro	DeepSeek	deepseek/deepseek-v4-pro	Kilo	Kilo

Codex intentionally excludes Codex/GPT from the external panel because Codex is already the lead reviewer.

Configuration

Host	User config	Default config	Env file
Claude Code	~/.claude/consensus.json	plugins/consensus/consensus.config.json	~/.claude/.env
Codex	~/.codex/consensus.json	plugins/consensus/consensus.codex.config.json	~/.codex/.env first, then ~/.claude/.env read-only fallback

Claude setup can smoke-test live model calls. Codex setup does not run live smoke tests unless you ask, because they consume credits.

Example Codex config:

{
  "version": "1.6.0",
  "lead": {
    "id": "codex",
    "name": "Codex",
    "excluded_model_ids": ["gpt", "codex"]
  },
  "models": [
    {
      "id": "gemini",
      "name": "Gemini 3.1 Pro",
      "command": "gemini",
      "resume_flag": "",
      "enabled": true
    },
    {
      "id": "kimi",
      "name": "Kimi K2.6",
      "command": "kilo run -m openrouter/moonshotai/kimi-k2.6 --auto",
      "resume_flag": "-c",
      "enabled": true
    }
  ],
  "min_quorum": 3
}

Commands And Skills

Host	Command / skill	Purpose
Claude	/consensus-setup	Configure models, API keys, and quorum
Claude	/code-review [target]	Multi-model code review with convergence
Claude	/plan-review [task]	Multi-model implementation planning
Claude	/consensus:review [target]	General document/topic review
Codex	consensus-setup	Configure the Codex consensus panel
Codex	consensus-code-review [target]	Codex-led code review with convergence
Codex	consensus-plan-review [task]	Codex-led implementation plan review
Codex	consensus-review [target]	Codex-led document/topic review

How It Works

Host lead  Model 1  Model 2  Model N
    |         |        |        |
    v         v        v        v
Independent review or planning
    |_________|________|________|
              |
              v
Synthesis: consensus, conflicts, comparison table
              |
              v
Convergence: APPROVE / CHANGES NEEDED, max 2 rounds
              |
              v
Final result with attribution

• Quorum: configurable per host.
• Graceful degradation: unavailable models are skipped if quorum still holds.
• Session artifacts: saved in /tmp/ for debugging and auditability.
• Independent reviews: each model starts from the same prompt with no cross-contamination.
• Codex isolation: Codex uses separate config and never writes Claude user config.

Contributing

Contributions are welcome. Please read the Contributing Guide before submitting a PR.

This project follows the Contributor Covenant Code of Conduct.

License

MIT License. Copyright (c) 2026 Altimate AI.