quorum

Quorum

Convene a quorum of AI models from your terminal. A set of portable Agent Skills that let any coding agent summon other AI CLIs — OpenAI Codex, opencode (DeepSeek / Kimi / Qwen), and a fresh Claude — as an independent panel for review, research, and hard calls. Plus turbo-review: one command that runs them all at once and synthesizes the verdicts.

Where the models agree, you have high confidence. Where they disagree, you get a precise list of things to look at.

Why

A single model — even a great one — has blind spots and talks itself into things. A second opinion from a different model, in a clean context, catches what the in-session model glossed over. Quorum makes that one keystroke, and turbo-review turns it into a panel vote.

The skills

External passes are read-only by default: they review, research and propose; your host agent synthesizes and makes the edits. The diff stays coherent, and nothing is touched without you.

Requirements

Each skill needs its CLI on your PATH — install only the ones you want; turbo-review uses whichever are present and skips the rest.

• codex — for codex and codex-imagegen
• opencode — for opencode (free models built in, or bring your own provider)
• claude — for claude

Install

These are standard Agent Skills (SKILL.md folders), so they work in Claude Code, Codex, and opencode alike.

Quick install (all agents)

git clone https://github.com/alesha-pro/quorum
cd quorum
./install.sh

install.sh symlinks each skill into the shared ~/.agents/skills directory (read by all three agents) plus the native skills dir of every agent it detects. Restart your agent — or start a new session — afterwards. Undo any time with ./install.sh --uninstall.

Manual install

Symlink (or copy) the skill folders into your agent's skills directory:

ln -s "$PWD/skills/turbo-review" ~/.claude/skills/turbo-review

Claude Code plugin

Prefer the plugin manager? This repo is also a Claude Code marketplace:

/plugin marketplace add alesha-pro/quorum
/plugin install quorum@quorum

Installed via the plugin, the skills are namespaced (/quorum:turbo-review, /quorum:codex, …). Installed as loose skills, they're bare (/turbo-review, /codex, …).

Use

/turbo-review   "review my staged diff before I merge — correctness, regressions, prod risk"
/codex          "challenge this migration plan: <plan>"
/opencode       "research the trade-offs of X vs Y for our use case"
/claude         "fresh-eyes pass on this design — what am I missing?"
/codex-imagegen "a 32x32 transparent pixel-art coin sprite → ./assets/coin.png"

Your agent will also reach for these on its own — running turbo-review before a risky change, or asking codex to pressure-test a plan — because that's what the skill descriptions tell it to do.

Configuration

• opencode model — defaults to the free opencode/deepseek-v4-flash-free (opencode Zen, no API key). Run opencode models to see what's available to you and pass --model for a stronger one you've configured; details in skills/opencode/SKILL.md.
• codex model — defaults to your Codex config (e.g. gpt-5.5, xhigh). Override per call with -m.
• claude model — defaults to your Claude Code config; pick a specific one with --model.

Compatibility

License

MIT © alesha-pro