Notes from the workshop.

// part one

Tearing down the gradient tool

The gradient tool got rebuilt from nothing. The old version had a flaw I had been ignoring for a while: every preset pulled from the same small set of colors, so they all looked like siblings. In the picker you genuinely could not tell them apart.

The new one is built on an open-source gradient library, ported with some care so it behaves like the real thing and not a rough copy of it. It got procedural textures, presets that actually look like different ideas, and a way to take your gradient with you, either as a single file or a snippet for a page you are already building. The picker shows real thumbnails now, each one rendered from the actual shader, so you choose with your eyes.

This was most of the day.

// part two

Halving the animation library, then doubling it

The text animation tool had a hundred animations and too many of them were filler. So every one went up for review, side by side with a hundred from another library, each one watched while it played, each one kept or cut. Fifty four got cut.

The thing I did not expect: most of the other library's animations were worth keeping, and the reason was subtle. They animate each letter on its own. Ours mostly moved the whole word at once. A fade that lands letter by letter and a fade that lands all at once look like the same animation in a list and are not the same animation at all. So both stayed. The catalogue is a hundred and forty six now, bigger than it started, and not padded.

// part three

A bug that had been wrong for a long time

While rebuilding the library I found out why the bouncy animations never felt right. Every bounce, every elastic, every rubber band had been using the wrong easing curve. A smoothing one. The keyframes were describing a bounce and the easing was flattening it the entire time.

Nobody had reported it, because it was not broken. It just felt slightly lifeless and there was no obvious reason why. Those are the bugs that last the longest.

// part four

Being wrong, on the record

Not everything today was a clean line. I was sure one of the glitch animations was mistimed and rewrote it to be faster. It got looked at, playing, and the original was simply better. I put mine in the bin. Twice I miscounted how many animations had survived the review and had to walk the number back, once in a document that disagreed with its own table.

I am writing that down on purpose. A devlog where every call lands perfectly is not a log, it is a brochure. The work was lumpy. It corrected itself a few times. That is what the work is actually like.

// part five

On doing this one Saturday at a time

Everything above is one Saturday. That is the whole project so far, and writing it out makes the pace honest. There is no sprint here and nothing is behind schedule, because there is no schedule. A tool gets the time it needs and ships when it is done.

The reward of building this way is that nothing has to be a minimum viable anything. The cost is that it is slow. That trade feels right. The log will grow one Saturday at a time, same as the workshop.

// part one

The tool that was almost a prompt dispenser

An old Notion doc surfaced. A six-stage prompt chain for spec-driven development, written a while ago and then forgotten. The obvious move was to ship it as written, a tool that hands you the prompts in order. I started down that road and stopped. Shipped literally it would have been a dated prompt dispenser, the exact kind of thin tool Cargo had already cut once.

So it got reframed. Not a chain of prompts to paste, but a tool that teaches one hard skill: spotting what is missing from a feature spec before you hand it to an AI coding agent. It teaches with causal chains, a gap in the spec leads to an assumption the agent makes leads to a concrete thing that breaks. Worked examples first, then a self-check so you practice the spotting yourself. That version earned its place. It became the tenth tool.

The document was raw material, not a product. The first idea was the wrong one, and the second only existed because the first got turned down on the way.

// part two

Renumbering, and a coupling that made no sense

The tools had been numbered in the order they were built, which is the order that matters least to someone arriving. They got renumbered into a deliberate presentation order, the visual and immediate tools first, the quieter references after them.

A small wrong thing fell out on the way. The moodboard tool used to stamp a tool catalogue number into the CSS you copy out of it. A number that means something inside this project and nothing at all in your file. It is gone from the export now. It had been there long enough that nobody questioned it, which is how that kind of thing tends to survive.

// part three

Making it work on a phone

The whole workshop was built desktop-first and, honestly, had no real mobile support. Not rough on a phone, broken in places. So every tool and every page got opened at phone width and worked through: a real mobile navigation menu instead of a desktop one squeezed thin, touch targets big enough for a thumb everywhere, no sideways scrolling, and one consistent set of breakpoints instead of the scatter of ad hoc widths that had piled up.

The UI Pattern Library was the hard one. It carries a lot of small embedded interactive demos and each had its own way of being slightly wrong on a phone. It took several passes on its own while the rest were already done.

// part four

A sweep for things that had quietly gone untrue

The last pass did not build anything. It read the copy against what actually shipped. A tool description still said a hundred text animations when the real number was a hundred and forty six. Another offered a phone mockup frame that does not exist. One advertised spring curves and a draggable demo for a tool that has neither. None of these were lies anyone told. They were true once, and the code moved while the words stayed still. They got corrected.

That is most of what Saturday Two was, looked at honestly. Not new things, but going back to finish: renumbering, mobile, the accuracy sweep. That is the unglamorous half of making something actually done, and it is the half that is easy to skip. The one good thing about doing it this way is that almost everything wrong got caught and fixed in the same pass that found it. The work stayed lumpy, same as Saturday One, and it kept correcting itself.

// the roadmap

The roadmap, loosely

A roadmap for a side project is a wishlist. Read it as one. No order, no dates.

The CSS Effect Lab got the same teardown the gradient tool got. It went from a single effect to a small playground of them, and it is better for it.

The dictionary of interface patterns that was circling got built. It shipped as the UI Pattern Library, a searchable reference of the small reusable pieces every web project rebuilds, with a position on when not to use each one.

The Component Prompt Builder was retired. It was a thin idea that did not earn its place next to the rest, so it came out rather than linger half believed in.

The Mockup Wrapper shipped, and went through the mobile pass with everything else. It could still use more frames, but it is no longer the one thing waiting. That clears everything that was mapped here, which is the honest state of a wishlist with no dates: the list it had has been built, and the next one is not written yet.

If you are reading this and that would help you, that is a good reason for it to jump the queue.