Flaux
In developmentgame3dprototype

Waterway Rumble

A comic-book waterslide racer built entirely in Claude

Race seven rival swimmers down twenty waterslides, from a gentle straightaway to a genuine vertical loop-de-loop. Boost strips, shortcut jumps, a vessel shop, and onomatopoeia everywhere. A full 3D game that lives in a single shareable file. In active development, changing daily.

Three.jsTypeScriptVite

What it is

Waterway Rumble is a comic-book waterslide racing game. You race seven rival swimmers down big 3D slides, leaning through banked turns, hitting boost strips, jumping the gaps where the slide just stops, and occasionally getting flung round a genuine vertical loop-de-loop. When things go wrong, the screen tells you in giant comic letters: SPLASH, BONK, WIPEOUT.

There are around twenty slides, from a gentle straightaway up to an extreme tier with names like The Gauntlet and The Vortex, plus a World Tour that runs through Antarctica, Egypt, Paris, and a neon Tokyo midnight stage. Coins from racing buy new vessels in a shop, each with its own stats. The rivals have names and personalities: Torpedo Tess races very differently to Lazy Luke.

The unusual part

The whole game is one file, and it was built entirely inside Claude as an artifact. No repo, no dev server, no deploy pipeline. The published artifact link is the game.

That started as a constraint and turned into the point. A single HTML file holding a full Three.js racing game, twenty tracks, an economy, and a save system is a genuinely odd object, and Claude's artifact hosting means the shareable link and the build are the same thing. When I want someone to try it, I send the link. That's the entire distribution story.

Under the hood it's a proper project: TypeScript, Three.js for the 3D, Vite bundling it down to that single file, and Playwright tests that drive the slides and take screenshots to catch visual regressions. It just happens to live somewhere unusual.

What I learned building it

Racing games are feel, not features. The track list came fast. What took the time was the water: how much the current pushes you, how sticky the slow patches are, how hard you can pump for speed on a steep wall, whether a jump feels like a launch or a stumble. None of that is visible in a feature list and all of it is the game.

The comic presentation earned its keep too. Big skewed buttons, onomatopoeia popups, a countdown that fills the screen. It sets the expectation that this is silly and fast, which buys a lot of forgiveness while the physics is being tuned.

Status

In active development, and moving fast: most days it gets updated more than once, so what the link shows in the morning isn't what it shows at night. That's why there's no public play link here yet. Once the game settles it follows the MiPixArt path: wrap it with Capacitor, put it on a flaux.com.au subdomain, and take it to the stores. The one-file constraint means that move is mostly packaging.

The bigger lesson

I keep being surprised by where a real app can live now. This one skipped every piece of infrastructure I normally set up, and none of it was missed at the prototype stage. The lesson isn't that repos and pipelines don't matter. It's that they're a scaling decision, not an entry fee, and you can find out whether a game is fun before you pay it.