Skip to content
Get packs

Guide · Agent skill

How to use the OpenRig Software Factory agent skill

Updated · written with AI from the project's own documentation and checked against our catalogue data

The OpenRig Software Factory skill is an instruction package that helps an AI coding agent set up a small team of agents for a real code repository, such as a builder and an independent reviewer, inside the OpenRig orchestration tool. It is written for people who want reviewed, repeatable work on a codebase, and for beginners who want to start with one bounded change. After this guide you will know what the skill does, how to install it, how to ask for it in plain language, and what to check before letting a team launch.

OpenRig Software Factorymvschwarz/openrig · Apache-2.0

Quick start

  1. 1Download and unzip the skill. Download the ZIP from the item page and unzip it into ~/.claude/skills/ for personal use, or into .claude/skills/ in your project. In the Claude apps, upload it under Settings → Capabilities → Skills instead.
  2. 2Install prerequisites and log in. Make sure OpenRig with its rig CLI is installed, and that Claude Code and/or Codex CLI is installed. Log in with claude auth login or codex login for the providers your team uses.
  3. 3Ask for a team in plain language. Describe the repository, the outcome you want and how much you want to spend. Mention whether you are new to agent teams so the skill starts small.
  4. 4Choose the team and providers. Let the agent recommend starter, workshop or factory, and confirm which providers each seat uses. It will not add a provider login you do not need.
  5. 5Preview, then launch. Run rig up with the --plan flag to review the plan. Then run rig up without it to launch, once you have agreed the work boundary and stop conditions.
  6. 6Hand off reviewed work. Use rig queue create and rig queue handoff to pass candidates and evidence between the builder and reviewer. Stop when the checked result is ready or a real blocker appears.

What the skill makes the agent do

Once installed, the agent uses the skill to guide you through setting up a team of AI coding agents on your repository. It helps you choose providers (Claude Code, Codex), launch the team with OpenRig rig commands, agree on a work boundary and stop conditions, and decide how work is handed between agents.

The skill also sets guardrails. It asks before launching or assigning work, and it avoids copying credentials. It recommends starting simple: a beginner can complete reviewed work without the optional Workflow feature.

  • Recommends one of three teams: starter (a Claude builder and a Codex reviewer, for one bounded change), workshop, or factory (seven agents for sustained product work).
  • Explains when to use queue handoffs, and when an optional Workflow is worth the extra setup.
  • Helps you grow a running team later by adding seats.

Install the skill from the ZIP or the Claude apps

Download the skill ZIP from the item page and unzip it into one of two locations. Use ~/.claude/skills/ to make the skill available across your projects, or use .claude/skills/ inside a single project so it applies only there.

If you use the Claude apps instead, upload the skill under Settings → Capabilities → Skills.

  • Personal install: unzip into ~/.claude/skills/
  • Project install: unzip into .claude/skills/ in the project folder
  • Claude apps: Settings → Capabilities → Skills, then upload the skill

Check what you need before you start

The skill expects OpenRig installed with its rig command line tool, plus Claude Code and/or the Codex CLI, depending on which providers your chosen team uses. You also need a login for each provider the team needs, and a repository to work on.

You only need the logins for providers your selected team uses. If one is missing, the skill will request the matching login command once.

  • OpenRig with its rig CLI installed
  • Claude Code and/or Codex CLI, as the team requires
  • Logins: claude auth login or codex login, for the providers in use
  • A repository you are authorised to change

Trigger the skill with a natural request

You do not need special syntax. Describe the outcome and the kind of help you want, and mention that you want a team on a real repository. The skill's description matches requests for a continuing software team, or for a repeatable path for reviewed work on an existing OpenRig team.

Be specific about the repository and the result. The agent will then read the repository instructions and current work before proposing a team.

  • Example: Set up a small reviewed team to add input validation to this repository.
  • Example: I already have an OpenRig team. Give me a repeatable way to review and hand off new tasks.
  • Example: I am new to agent teams. Start me with one bounded change on this repo.

Launch the team and agree the work boundary

Before anything launches, the skill shows the chosen team and the resolved runtimes, then has you preview it with the plan flag. Only after you confirm does it run the launch command. Lifecycle commands such as rig up and rig down still ask for permission in Claude team seats.

Agree the work boundary, your time and spend limit, who answers open questions, and when the team should stop: a checked result, no authorised next work, an exhausted budget, or a real blocker.

  • Show the plan first, then launch.
  • Keep purpose, acceptance criteria and evidence in existing project files.
  • Permission changes are offered as a choice. You can decline and keep the team default.
rig up <team> --cwd . --plan rig up <team> --cwd .

Hand off work, and grow the team later

For continuing work, the skill uses queues: a worker creates or claims work with rig queue create, then passes the candidate and its evidence to the next owner with rig queue handoff. Real blockers and continuations are recorded. The optional Workflow path (rig workflow compile and rig workflow instantiate-lifecycle) is for repeated steps that need an explicit dependency graph.

To add capacity, the skill points you to the grow steps in the worked example. Adding seats does not assign work or change permissions by itself.

  • Start with the two-agent starter, then add one or two seats when the workload needs it.
  • Use the worked example at references/worked-example.md for the concrete queue loop and grow commands.
  • Use rig context get to read installed reference files.
rig queue create rig queue handoff

Safety notes: scripts, network and token spend

This skill runs scripts and uses the network. It launches agent sessions through rig commands, and those sessions spend model tokens. Background daemon checks are not model turns, but delivered wakes and resumed work can spend tokens, so prefer event-driven waiting over frequent reminders.

The skill may propose additions to your native agent permission settings. Read each proposed change before you accept it. Wakes cannot answer questions or clear permission prompts, and they do not guarantee progress.

  • Review any permission change; you can say No and keep the team default.
  • Set a time and spend budget before launch.
  • Do not use this for quick one-off code questions, since the setup adds overhead.

What people use it for

  • Adding a validation or parsing feature to an existing repository, with a builder and an independent reviewer checking the change.
  • Keeping an ongoing team on a codebase, where tasks are queued and handed off with recorded evidence.
  • Onboarding a new developer to agent teams by starting with one bounded change before adding more seats.

Frequently asked questions

Is the OpenRig Software Factory skill free?

The skill is licensed under Apache-2.0. Its source does not state prices for OpenRig or the model providers, and agent sessions spend model tokens, so check your own provider usage and billing.

What do I need installed to use this skill?

You need OpenRig with its rig CLI, Claude Code and/or the Codex CLI depending on your team, the matching logins (claude auth login or codex login), and a repository you are allowed to change.

Where do I put the skill folder?

Unzip the downloaded ZIP into ~/.claude/skills/ for personal use, or into .claude/skills/ inside a project. In the Claude apps, upload it under Settings → Capabilities → Skills.

Does the skill run scripts and use the internet?

Yes. The skill is marked as running scripts and using the network, and it launches agent sessions that spend tokens. Review the commands before launch and check any permission changes it proposes.

How is this different from just asking one coding agent for help?

A single agent answers a question or makes one change. This skill sets up a small team with a builder and an independent reviewer, plus a queue for handing off work. It is more setup, so it suits repeatable reviewed work better than quick one-off questions.

Why does the agent ask me before launching?

The skill asks once before launching or assigning work, because launching sessions spends tokens and may change settings. You can decline permission changes and keep the team default.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Use this skill if you want a repeatable, reviewed multi-agent workflow on a real repository, or if you already run OpenRig and want to add seats and ongoing tasks. Skip it for quick one-off code questions, or if you are not comfortable with agents launching sessions and spending model tokens; its own guidance is to start simple with one bounded change.

Download the skill