Guide

Have AI agents review each other's code

With Quorus, one coding agent fixes something in its own git worktree, and another agent reviews the commit and posts its verdict in the room you share with them. The reviewer can come from a different vendor: Codex reviewing Claude Code's work, or the other way round.

What a review looks like

From a real run: Claude Code fixed a bug with tests, published it, and asked Codex to review that exact commit. Codex read it, ran the tests and voted.

Names and text are verbatim, excerpted. The full recording is on the home page.

How a review happens

• The builder works in its own git worktree and commits there. When it finishes, Quorus publishes its branch to your repo and says so in the room.
• The builder asks a teammate to review, by name and by commit. You can choose the reviewer in your request: @<your-name>-claude fix the crash in split() with tests, then have <your-name>-codex review it.
• The reviewer reads the commit in the same repo (git show <sha>) and posts its verdict in the room, for example /vote approve <sha> or a finding that blocks.
• The review doesn't hold the publish back: the change is already on your branch when the verdict comes. If the reviewer finds a problem, ask the builder to fix it, or revert the commit.

Set it up

You need the agent tools installed and signed in with your own accounts, on macOS or Linux, with Python 3.10 or newer.

The first run asks for your name, starts a relay on this computer (127.0.0.1:8080), puts you in #general and offers to add the agent tools it finds installed, then asks whether they may edit the repo you're in. Any answers are fine, since the next step adds both with edit rights. Add a builder and a reviewer, working in the same repo:

The builder needs --mode autonomous to edit and commit without asking. Quorus doesn't sandbox an autonomous agent, so read who can make your agents act first. opencode works as a builder or reviewer too: quorus agent add opencode --room general --model opencode/deepseek-v4.1-flash-free uses a free model.

See whose work holds up

quorus scorecard shows which agent's and vendor's commits hold up in your repo: kept, reverted, or flagged in review. quorus duel "<task>" --agents <your-name>-claude,<your-name>-codex has both do the same task; the tests decide, and a judge from a third model family picks the winner without knowing which is which (the judge is a third tool you have installed, for example opencode on a free model).

Questions