OpenRig

Rigs

Install a team someone already debugged.

Teams of agents you can install by pasting one block. Each one shows what it will do and how much OpenRig has tested it, for the harnesses you have.

Or set up with your agent
npx skills add https://github.com/mvschwarz/openrig/tree/v0.7.20/skills/rigs --skill rigs --agent claude-code

Run it in your project folder, then type /rigs in Claude Code. The skill helps install OpenRig and get a team working on your project.

Needs Node.js 22 or 24 (on Apple silicon, 22), which provides npx. If your shell says npx isn't found, Node isn't installed or isn't on your PATH. How to install Node.js →

Contributing to someone else's project? Workshop follows the repository's own conventions and CONTRIBUTING guide, so the pull request you send reads like the maintainers' own.

  • Every seat is a real Claude Code, Codex or Pi session, no wrapper
  • Pinned to a commit
  • Shows what it will do before it starts
  • Reviewed for listing, not a security audit

Picked by OpenRig maintainers as the best place to start. Each seat runs in a harness, the agent app you already use: Claude Code, Codex or Pi. What the labels mean

Workshop

Listed by OpenRig maintainers

A four-agent software team for your repository: a lead and a reviewer on Codex, a builder and QA on Claude Code. If you only use one of them, you can run every seat on the one you have.

4 seats · runs on Claude Code, Codex or Pi

The workshop team has 4 agents in 2 pods. In orch, lead runs on Codex. In dev, build runs on Claude Code, review on Codex and qa on Claude Code. Lead hands work to build; build hands work to review and qa. You talk to lead, at orch-lead@workshop.
1Rig
The whole team. One name, installed and started with one command.
2Pod
Seats that work on the same part of the job and share what they know about it.
3Seat
One agent with one role, in its own terminal. Its address is how you and the other seats reach it.
4Harness
The agent app the seat runs in: Claude Code, Codex or Pi. A preset swaps it per seat.

How much OpenRig has tested it

  • Harnesses tested by OpenRig on macOS and LinuxClaude Code 2.1.290 and Codex 0.160.0 on macOS, OpenRig 0.6.6, 6 Oct 2026; Claude Code 2.1.299 (updated to 2.1.302) and Codex 0.163.0 on Linux, OpenRig 0.6.11, 18 Oct 2026OpenRig checked each harness on its own, but hasn't run this team as a whole at commit f4579db3ab7e, or on OpenRig 0.7.20.

Permission prompts

Permission prompts: off

Options

  • Workshop with a review pair · 5 seatsOptional five-seat Workshop variant with two independent reviewers, one Claude Code and one Codex. The four-seat Workshop remains the recommended default. Both install as workshop in the same directory; choose one variant.
  • Recommended
  • All Claude
  • All Codex
  • All Pi
See the team and install →

Reviewed for listing · not a security audit

Reference teams

Larger teams OpenRig keeps as worked examples. Install them the same way; they're not tested by OpenRig unless the label says so.

Community listings

Teams other people built and submitted. Each one is reviewed and pinned to a commit before it appears here. What the labels mean

No community listings yet.

Share your team and it can be the first.

What the labels mean

Every team shows two things before you install it: how much OpenRig has tested it, and whether its agents ask before they act. Both are per harness setup, so they can change when you pick different harnesses.

How much OpenRig has tested it

A ladder of assurance, from most to least. It sets your expectation; none of these is a warning.

Tested by OpenRig
OpenRig ran this team and debugged it, and it worked for us. The date and version are under the label.
Tested with help (N)
It worked, but someone had to step in N times to get it there.
Partly tested
Only part of it was run. A note on the team's page says what was skipped.
Harnesses tested by OpenRig on macOS
OpenRig has run this team, and the harnesses it uses are tested on that platform. The line under the label gives the versions and what hasn't been run yet.
Harnesses partly tested by OpenRig on macOS
The same, for some of its harnesses. The label names the ones OpenRig hasn't run yet.
Untested
Shared as a structure to read, try and build on: an idea, an experiment, or a team OpenRig hasn't tested recently. The note under the label says what's known.
Known problem
It fails in a way OpenRig has seen. The note says how.
Status unavailable
OpenRig's records for this setup couldn't be read, or are missing.

What counts as help. An assist is one action a person or the test runner took to get past a problem during the run, such as restarting a seat or re-running a command; following the bundle's own documented steps doesn't count.

Community reports from people who ran a team appear beside the label and never change it. No label is green: each one says what happened, not that a team is safe.

Permission prompts

Declared by the team's author for each seat, and shown before you install.

On
Each agent asks you before it runs a command or changes a file.
Off
The agents act without asking you first: they run commands and change files on their own. The first time on a machine, Claude Code asks once to accept this.
Your machine's default
On Claude Code seats, reads, searches, rig commands and tests run without asking, and edits and other commands ask first. Codex seats follow their own sandbox and approval settings. Your own settings can change either.
Policy <name>
A named rule set that allows some actions and asks about others. The team's page links to what it allows.
Not stated by the bundle
The team doesn't say. Pi seats always show this, since the bundle can't tell which Pi you run.

Before you install a team with prompts off, read what it will do on its page: which agents start, what they're told, what they write and which addresses they mention. Agents run as you, with your access.

To report a problem with a team, use "Tell us how it went" on its page. A withdrawn team's page says so, with the date, and no longer offers an install command. When seats differ, the page lists them, for example "off for dev.build; on for the other seats". Every listing is reviewed before it appears. Review is not a security audit.

Share your team

Publish your team on GitHub and submit the link. After review it appears here automatically. Your files stay in your repository; the listing records the exact commit a maintainer reviewed. You get your name and repository on the listing, a page at openrig.dev/rigs/your-team, and a badge for your README.

  1. Step 1

    Put the team on GitHub

    A folder with rig.yaml, configurations.yaml (which harness each seat can use) and bundle.yaml (what people need before they install). Check what each configuration will do before you submit.

    # the workshop's own output
    $ rig bundle configurations rig.yaml
    recommended: dev.build=claude-code,dev.qa=claude-code,dev.review=codex,orch.lead=codex  (recommended, as rig.yaml is written)
    all-claude: dev.build=claude-code,dev.qa=claude-code,dev.review=claude-code,orch.lead=claude-code
    all-codex: dev.build=codex,dev.qa=codex,dev.review=codex,orch.lead=codex
    all-pi: dev.build=pi,dev.qa=pi,dev.review=pi,orch.lead=pi
    $ rig bundle check .
  2. Step 2

    Submit the link

    Open a pull request on mvschwarz/openrig-registry that adds one small file. The registry check replies "Submission received", or tells you the name is taken. How to submit →

    # registry/submissions/your-team.yaml
    repository: https://github.com/you/your-repo
    folder: rigs/your-team
    ref: main
  3. Step 3

    Reviewed, pinned, listed

    A maintainer pins your ref to a full commit, generates what each configuration will do, and lists it. A later push changes nothing here until an update is reviewed.

    ✓ Listed: openrig.dev/rigs/your-team
    Badge: [![Listed on OpenRig](https://openrig.dev/rigs/your-team/badge.svg)](https://openrig.dev/rigs/your-team)
    Reviewed for listing on <date> at commit <short>. Review is not a security audit.

    Workshop: listed on OpenRig The Workshop team's badge