Architecture Studio 1.3: A Course in the Terminal

The numbers

We open-sourced Architecture Studio on March 2, 2026. At three months it was 179 stars and 39 skills. Four and a half months in:

• 255 GitHub stars and 54 forks (179 and 40 at three months)
• 1,148 unique visitors and 343 unique cloners in the last 14 days — visitors roughly flat since June, but unique cloners up a third. Fewer browsers, more installers; for a tool repo that’s the trade you want.
• Eight releases, now at v1.3.0, and 40 skills

The 40th skill is a tutor

Every skill we’d shipped until now assumed you were already comfortable in a terminal. That assumption is exactly backwards for this profession: the architects who’d benefit most from the catalog have spent twenty years in Revit and Rhino and zero minutes at a command line. No install guide fixes that — a guide tells you what to type, not what this thing is or why the blank screen isn’t broken.

So v1.3’s headline isn’t an analysis skill. It’s /learn — a six-module, hands-on course that teaches Claude Code itself, from “what is this blank screen” through building your own skills, to running a real task on a real project. Most modules are 15–20 minutes built around one exercise; the capstone runs longer — you bring a real project. Progress lives in a plain PROGRESS.md, so you can stop after any module and the course remembers — which is itself the first lesson in how memory works here: files.

The practice project is a fictional Brooklyn art museum planning a rooftop expansion — the Greenpoint Museum of Art, six deliberately messy files that ship with the skill: raw site-visit notes, a space-program spreadsheet with a duplicate row and mixed units, a voice-memo transcript hiding in a file named like a photo, a scanned zoning memo that cuts off mid-page. Every module runs on the same project, so by the final drill you know it well enough to catch the machine being wrong about it.

Module 6 is the point of the whole thing. The tutor summarizes the project’s zoning excerpt and you interrogate it: pick any claim and ask “show me where the document says that” — then ask about something the excerpt never covers, and hear the only honest answer: the document doesn’t say. Silence and “no requirement” are different things, and learning to tell them apart on a fictional zoning memo is cheaper than on a real filing. The question works on any AI output, forever.

The course ends where the sandbox stops: the capstone sets up a copy of one of your actual projects, writes its standards binder, and runs one real task end to end. If you type /studio before you’re ready for any of it, the router now notices and points you at the course.

Also in 1.3

The ten workflow-stage plugins collapsed into one — claude plugin install architecture-studio now brings all 40 skills and 7 agents in a single install, with the old groups surviving as documentation. Under the hood: a stricter lint (every slash command and referenced file must resolve, counts can’t drift from the files on disk), the disclaimer pipeline actually wired end to end, and a batch of content corrections across zoning references, EPD baselines, and occupancy examples. Details in the CHANGELOG.

What’s next

The standalone harness — durable studio memory, firm-authored skills, governed connectors — is in development; the product page has the shape. If you’re new to all of it: install the plugin, type /learn, and give it fifteen minutes.