This is a multi-step software factory intensive, built on Gas City to:
- Provide a fast path to the simplest possible factory.
- Demonstrate key components of a software factory.
- Teach you how to customize those components so that you can create the agents and workflows you need to power your own software factory.
You'll start with out-of-the-box examples that ships with Gas City and two reference implementation agents from Gas Town, mol-polecat-work and mol-refinery-patrol. And you'll evolve a multi-agent workflow. You'll start with a simple implement -> review workflow. Then as the tutorials progress you'll layer on:
- GitHub pull requests
- Opinionated quality gates
- Specialize reviewer agents
- An architecture-best-practices loop
- A multi-model
code-review-loop - Etc.
Stuck on a step, want to share what you've built, or looking to collaborate with other participants? Join the Actual AI User Community Slack! Here you can share what you've built, ask questions, and get help from other members of the community.
By the end of the Progression, you'll have:
- A Gas City factory named
factory1 - An example project or "rig" named
ascii-artpushed to your own GitHub repo - Gas City packs with formulas and agents that wire the
ascii-artproject to your workflows
Factory1 is totally decoupled from the ASCII Art example project. If you're happy with its functionality at the end of these tutorials, you can start using it with your own projects immediately. If not, you'll have what you need to customize it further, and then you can start using it. You don't need to create a different factory to get started with Gas City.
The example project is intentionally simple. The point is to learn the workflows.
Even though you're learning how to build a software factory, our examples simply generate text files, specifically ASCII Art (like this) for letters A-Z and numbers 1-100, along with little rhymes in each.
All the same principles and processes apply. In your projects, your technical requirements will be related to your technologies stack rather than how many characters are allowed in each file, and your design system will focus on your brand and user experience rather than styling of markdown files. These details are project- or rig- specific. They do not live in the factory.
Tools and versions:
gc1.0+bd1.0+gh2.x (authenticated:gh auth status)git- macOS or Linux with
tmuxavailable
GitHub:
- One fresh GitHub repo for the rig — your own org or a sandbox account. The rig will push to this repo and branch protection will be applied to it.
There are two paths through this repo.
Each step unlocks the next. Every page is self-contained: prereqs, walkthrough with copy-pasteable commands, verification block, troubleshooting, and a "what's next" link.
- Setup: Create a Gas City named factory1 —
gc init, import a setup pack (based on thegastownreference pack), configurecity.toml - Setup: Add ASCII Art rig —
gc rig add, push to GitHub, create 138 beads - Setup: env vars — handy variables for easier copy-and-paste-ing
- Basic flow (OOTB) — one agent claims a bead and implements (
mol-polecat-work) + another reviews (mol-refinery-patrol) with no extras - First review loop — extend the refinery patrol with a required round of feedback before any bead becomes a PR
- Branch protection — only approved humans merge to
main - Architect agent — add an architecture-aware reviewer between the polecat and the refinery
- Bead gate checks — add a project-manager agent at the front of the factory
- Bead gate checks — Grill Me — refine beads at the source with a local Grill-Me skill
These are independent follow-ons. Pick the ones that interest you; they don't have to be done in order.
- Bead-creation formula extensions — auto-link/create design, testing, and documentation references
- Specialize reviewers per domain — split the single reviewer into ADR, design, testing, and docs reviewers
- Architecture best practices loop — per-principle scoring with an append-only audit trail
- Strengthen review system — multi-vendor reviewers, a synthesizer, and an iterate loop
- Clone this repo and
cdinto it. - Read this README end to end (you're nearly done).
- Open
progression/00.1-setup-foundation.mdand work through it. - Each page ends with a "what's next" link. Follow it.
- After the Basic Progression, browse the Hardening Exercises and pick what's useful.
Every page is designed to be copy-paste-runnable. Commands are exact. When a page introduces a new artifact (a formula, an agent definition, a script), the artifact lives under artifacts/ and the page tells you exactly where to put it.
Ready? Head to progression/00.1-setup-foundation.md.