Repo Week covered three pieces of the AI design stack: a real brand’s design system as a file, a taste skill, and a design vocabulary. But the stack this all came from — a five-piece setup for fixing the “every AI website looks the same” problem — has two more pieces, and they’re the two that separate a good-looking demo from something you’d actually ship.

Piece four is polish. Piece five is proof.

One teaches your builds how to move. The other makes your agent open its own work in a browser and check it before you ever see it. Neither is glamorous. Both are the difference between “Claude made a website” and “that website is ready.”

Here’s each one in detail — what it does, who it’s for, and the honest limits of what I can tell you about them.

Where these two sit in the stack

The stack fits together like this: the first piece supplies a real design system as a file, the second and third teach taste and a design vocabulary, the fourth is the animation polish layer, and the fifth lets Claude verify its own output in the browser.

That ordering matters. Polish without verification is decoration on an unchecked build. Verification without polish tells you a stiff, generic page works exactly as intended. You want both ends covered — the motion craft that makes a page feel designed, and the browser check that confirms the page actually behaves.

So let’s take them in order.

Emil Kowalski’s skills: the animation polish layer

Repository: github.com/emilkowalski/skills

Site: emilkowal.ski/skill

License: MIT

Stars: reported at roughly 47,000 when this was pulled in October 2026 — that’s a reported figure from the pull, not a fresh live count, so treat it as approximate.

What it is: a set of Claude skills focused mostly on animation and motion craft. That’s the whole pitch, and it’s more important than it sounds.

Static pages are the tell. When an AI builds a website, everything tends to arrive at once — no transitions, no easing, nothing that moves the way a designed thing moves. The layout can be right, the colors can be right, and it still feels like a screenshot instead of a product. Motion is where that feeling gets fixed: how elements enter, how states change, how the page responds when someone actually uses it.

That’s the layer these skills teach. Not a design system — the stack already has one of those. Not vocabulary — covered too. This is craft knowledge aimed at the specific weakness AI builds have: they don’t move right.

Who it’s for: anyone building sites or interfaces with Claude who keeps getting output that looks correct but feels dead. If your complaint about AI-built pages is “generic,” part of that complaint is layout and type — and part of it is that nothing on the page has any life in it. This is the piece aimed at the second half.

One honest note: the source material for this stack describes these skills by their focus — animation and motion craft — rather than walking through individual skills one by one. So I’m not going to invent a catalog for you. The repo is the catalog; go read what’s in it and take the pieces that fit your builds.

Playwright: the verification layer

Repository: github.com/microsoft/playwright

Site: playwright.dev

Stars: the source material for this stack doesn’t quote a star count, so I won’t either.

What it is: Microsoft’s browser automation project, and in this stack it has one specific job — it lets Claude open a browser and test and check its own work.

Read that again, because it’s the most important workflow idea in the whole stack. The first four pieces make the output better. This one makes the output checked. Claude builds the page, then actually opens it, looks at what rendered, and verifies it — instead of handing you code and a confident summary of code you’ve never seen run.

If you’ve followed my work for any length of time, you know this is a drum I beat constantly: never trust a report about the thing when you can look at the thing. Verification is the difference between an agent that says “done” and an agent that shows you done. Playwright is what closes that loop for web work.

It’s also worth knowing this project shows up in my library twice, in two different roles. It sits here in the design stack as the verification step, and it appears in my Claude Code plugins material in an MCP role — same project, two uses. When one tool keeps surfacing in different workflows, that’s usually a sign it’s load-bearing, not trendy.

Who it’s for: anyone letting an AI build web pages or web apps. The better your generation stack gets, the more verification matters — fast, polished, wrong is still wrong. A browser check catches the wrong part before your visitors do.

How I’d run the two together

Build with the system, the taste, and the vocabulary pieces. Apply the motion craft from Emil Kowalski’s skills where the page needs to feel alive — entrances, transitions, state changes. Then, before anything ships, run the verification pass: open the result in a real browser through Playwright and check what actually rendered.

Polish, then proof. In that order, every time.

The wider system these tools plug into — the one that turns repos like these into actual income workflows — is laid out in the pillar piece here: https://bigjermsays.com/how-to-make-money-with-ai-build-a-system-not-a-side-hustle/. Tools are cheap. Systems are the asset.

FAQ

Do I need all five pieces of the design stack?

No. Each piece fixes a different problem — system, taste, vocabulary, motion, verification. If your builds look generic, start with the design system. If they look fine but feel dead, that’s the motion layer. If you keep shipping broken pages, that’s verification. Take the piece that matches your actual problem.

Are these two free?

Emil Kowalski’s skills are MIT licensed. Playwright is an open-source Microsoft project. Neither asks you to buy a seat to use it the way this stack uses it.

Is the star count on Emil Kowalski’s skills exact?

No — roughly 47,000 was the figure reported when the stack was pulled in October 2026. Star counts move daily. I quote it as reported rather than dressing it up as a live-checked number.

What does Playwright actually check?

In this stack’s terms: it opens the built page in a browser so Claude can test and check its own work — what rendered, whether the page behaves — instead of trusting the code alone. It’s the verification step, not a design tool.

Start here

These two repos are finishing tools — they make the work better and the shipping safer. If you want the full system they plug into, start with the $7 Playbook: https://bigjermsays.com/product/the-ai-income-engine-free-playbook/ — or go straight to the complete build, the $47 Complete AI Income Engine: https://bigjermsays.com/product/complete-ai-income-engine-step-by-step-edition/