OpenRig: A Harness Wraps a Model. A Rig Wraps Your Harnesses
A harness wraps a model. A rig wraps your harnesses. That is the one-line pitch of OpenRig, and it names a layer most people running coding agents have already built by hand: a pile of terminal tabs, some with Claude Code, some with Codex, and a human copy-pasting between them.
OpenRig turns that into a YAML file. You define the team, boot it with one command, and every agent gets a stable address like dev-owner@first-project. The starter rig is two Codex seats, an owner and a checker. You send the owner a bounded outcome, it records the task in a queue, asks the checker to review the exact candidate, and reports back. Claude Code and Codex can sit in the same rig. There is a TUI that shows the team as a graph, then as a table of seats with runtime, model, context and state. It runs on Node and tmux, installs from npm as @openrig/cli, and is Apache 2.0.
It is trending on GitHub today at around 900 stars, and it ships fast: v0.5.15, v0.5.16 and v0.5.17 all landed between Friday and Sunday morning. The part worth copying even if you never install it is the README section titled what OpenRig changes on your machine. It lists every hook, trust setting and skill file the tool writes, including into ~/.claude/skills, and tells you to back them up first. That is rarer than it should be for tools that install executable hooks into your agents.
Why it matters: the owner-plus-checker default is the independent-verifier idea turning into a product default rather than a research recommendation. Mixing vendors is the other quiet point. A Codex seat reviewing a Claude seat's work, or the reverse, is the cheapest way to get a checker that does not share the author's blind spots. Small project, sharp idea.
Link: github.com/mvschwarz/openrig
← Back to all articles
OpenRig turns that into a YAML file. You define the team, boot it with one command, and every agent gets a stable address like dev-owner@first-project. The starter rig is two Codex seats, an owner and a checker. You send the owner a bounded outcome, it records the task in a queue, asks the checker to review the exact candidate, and reports back. Claude Code and Codex can sit in the same rig. There is a TUI that shows the team as a graph, then as a table of seats with runtime, model, context and state. It runs on Node and tmux, installs from npm as @openrig/cli, and is Apache 2.0.
It is trending on GitHub today at around 900 stars, and it ships fast: v0.5.15, v0.5.16 and v0.5.17 all landed between Friday and Sunday morning. The part worth copying even if you never install it is the README section titled what OpenRig changes on your machine. It lists every hook, trust setting and skill file the tool writes, including into ~/.claude/skills, and tells you to back them up first. That is rarer than it should be for tools that install executable hooks into your agents.
Why it matters: the owner-plus-checker default is the independent-verifier idea turning into a product default rather than a research recommendation. Mixing vendors is the other quiet point. A Codex seat reviewing a Claude seat's work, or the reverse, is the cheapest way to get a checker that does not share the author's blind spots. Small project, sharp idea.
Link: github.com/mvschwarz/openrig
Comments