openrig: one harness for Claude Code and Codex

TL;DR: openrig is a multi-agent harness that runs Claude Code and Codex together as one system, per its own one-line description. The topic list spells out the shape: agent-harness, agent-orchestration, agent-skills, ai-coding, cli, codex-cli, and tmux, all under an Apache-2.0 license.

What it is

Most agent tooling gives you one agent in one terminal. openrig’s pitch is a different one: put Claude Code and Codex in the same system and orchestrate them together. The description says “as one system”, and that phrase is doing the heavy lifting.

The topic list does more work here than the description does, and it is unusually complete for a project this young:

• agent-harness and agent-orchestration frame it as a control plane, not a thin wrapper
• agent-skills puts skills in as a first-class concept
• claude-code and codex-cli name both agents explicitly
• multi-agent and multi-agent-systems make the plural intent unambiguous
• tmux implies panes rather than a web UI
• cli and ai-coding place it firmly in terminal work

That tmux entry is the most informative one. Agent tooling built on tmux panes tends to be inspectable and scriptable, which is a meaningfully different choice from a closed graphical interface that hides each agent’s state. You can look at every pane, scroll back through any agent’s transcript, and reattach to a session after your laptop sleeps.

The two agents side by side

	Claude Code	Codex CLI
Named in repository topics	yes	yes
Runs under one harness	yes	yes
Orchestration layer	tmux panes	tmux panes
Shared skill concept	agent-skills	agent-skills

The interesting question, and one the description does not answer, is what happens when the two agents disagree. Two coding agents pointed at the same worktree will produce conflicting edits unless something arbitrates between them. If openrig partitions work by pane, that is simple and safe. If it lets both agents touch the same files, you want to understand the merge story before you trust it with anything you care about.

The numbers

Field	Value
Stars	1,017
Forks	105
Open issues	51
Language	TypeScript
License	Apache-2.0
Created	2026-04-01
Last push	2026-09-28
Homepage	openrig.dev

A fork count of 105 against 1,017 stars is roughly a 10 percent fork rate, high for a TypeScript CLI. Forking an agent harness usually means re-pointing it at a different agent or a different provider, so that number reads as adoption rather than idle curiosity.

Created 2026-04-01 and pushed 2026-09-28, the project is about six months old and still moving daily.

How to try it

gh repo view mvschwarz/openrig

The project site is openrig.dev. To follow the source:

git clone https://github.com/mvschwarz/openrig.git
cd openrig
ls

When to use it

openrig fits if you already use more than one coding agent and have felt the cost of switching between them. The multi-agent framing pays off when tasks genuinely parallelise, for example one agent on tests while another works on the implementation.

It does not fit if a single agent with a solid workflow already covers your work. A harness that coordinates two agents adds process, and process you do not need is just overhead. The 51 open issues on a six-month-old project are worth a skim before you build a daily habit around it.

Source and Accuracy Notes

• Project page: https://github.com/mvschwarz/openrig
• Homepage: https://openrig.dev
• Source repository: mvschwarz/openrig (verified via GitHub API html_url)
• License: Apache-2.0 (verified via GitHub API license.spdx_id)
• Stars at time of writing: 1,017