Stop Building Alone: Using GStack to Run a Solo Startup Like a Real Team

I've been testing GStack for a few weeks now and I think it's worth writing about.

It's a framework that drops a "team" into your Claude Code setup — CEO, design lead, engineering manager, QA. You stay the founder. The agents argue with you so you don't have to argue with yourself at 2am.

That's the actual pitch. Everything else is detail.

GStack is a folder of skills you clone into your project. You point Claude Code at two folders — your codebase, and the GStack folder — and now your AI assistant has opinions backed by a structured playbook instead of vibes.

Setup is three steps:

That's it. Your code is the body. GStack is the brain.

There's one skill called Office Hours that I keep coming back to.

Most AI tools agree with you. That's the problem. You ship a feature, the AI tells you it's great, you ship another, same response — and three weeks later you've built something nobody asked for.

Office Hours pushes back. It asks where you are (pre-revenue? scaling?), what your customers actually complain about, and whether the feature you're excited about solves any of that. If the answer is no, it tells you. Then it gives you a design doc, GitHub issues, and a prioritised list of what to build instead.

One tip: tell Claude to use the built-in question UI for any clarifying questions. Otherwise the terminal turns into an essay you have to read. With the UI you tap an option and move on.

Once you have a direction, two more skills do the heavy lifting.

Auto Plan runs four agents — CEO, design, engineering, dev experience — and they fight. Then a set of decision principles breaks the tie and merges everything into one plan. The "borderline scope" principle is the one I care about most. It's the anti-feature-creep guardrail. Keeps the MVP from quietly becoming v3.

Design Shotgun is for the UI phase. It generates multiple design variants of a page in parallel and dumps them into a designboard.html file you can open in the Claude Code preview. You pick a direction before you write any real CSS. Cheap iteration before expensive iteration. That's the whole point.

GStack isn't magic. It won't write your business model. It won't find you customers. What it will do is stop you from making the cheap, obvious mistakes a solo founder makes when there's nobody in the room to push back.

That's worth a lot when you're building alone.

If you're working in Claude Code and you're shipping anything that matters, try it for a week. The setup takes ten minutes. You'll know by day three whether it's helping.

GStack Github link: https://github.com/garrytan/gstack