The renamed repository contains 6 brushes, not a code editor
GitHub redirects blakeshao/p5-playground to blakeshao/kool-brushez. The current project is a drawing application, not a place to type and execute arbitrary p5 sketches. Its six brushes produce animated music scores, schematic diagrams, airbrushed patches, subway networks, crowds, and model-kit sprues. All of them draw into the same piece of paper, so you can switch styles without discarding earlier marks.
The narrower purpose works in its favor. One brush picker, one settings panel, and one canvas are easier to understand than a general creative-coding environment. Mouse, pen, and touch share the same input path. Desktop panels collapse, while small screens turn them into side drawers. The UI also includes erasing, auto-fill, new variations, clear, aspect-ratio choices, and keyboard shortcuts. It feels like a specific art tool rather than a framework demo.
A brush needs 3 files and a small lifecycle
Each brush has a registry entry, an implementation module, and a preview image. The registry declares its name, controls, description, and lazy loader. The factory then implements setup, resize, drawing, pointer handling, finishing, clearing, auto-fill, randomization, and settings methods. The host owns the canvas, history, export, and coordinates, which prevents every brush from creating a separate p5 world.
The interface has useful constraints. A brush cannot create or resize the main canvas, install global callbacks, or save the file itself. Switching brushes finishes pending animation, commits pixels to the common artwork, and clears temporary layers. Modules load on demand and are initialized once. Those rules are specific enough to stop state from leaking across tools, yet adding a seventh brush does not require modifying the studio internals.
What happened when we ran it
We ran commit 0cc5c7f in a fresh unprivileged Debian container with 3 CPUs, 8 GB of RAM, Node 22, and no secrets. Installation took 6 seconds, added 23 packages, and used 48 MB on disk. The checkout contained 62 files, roughly 8,583 lines of source, and occupied 1.9 MB before dependencies. The Vite build completed successfully in 8 seconds.
The default test command finished in 5 seconds with 23 passed and 0 failed under node:test. Npm audit reported 0 known vulnerabilities across every severity. The repository has a tests directory, no Dockerfile, and no CI workflow. This was a clean run for the supplied Node target. It did not include the separate Playwright browser command, so we did not independently verify touch, export dimensions, Retina behavior, or cross-brush compositing in Chromium.
History stops at 16 steps, and reload removes the drawing
Undo and redo keep raster snapshots of the whole paper, with a limit of 16 editing steps. That includes strokes, auto-fill, and clear. The history restores pixels rather than procedural brush state, so a restored pattern may begin animating from fresh internal state. The approach is simple and predictable for a small canvas, but each checkpoint copies image data and the stack is intentionally bounded.
The harder limitation is persistence. Artwork lives only in the current tab's memory. Reloading or closing the page loses it unless you exported first. PNG export settles pending animation and saves the full-resolution paper without the controls or cursor. There is no documented JSON project file, layer model, SVG path export, or cloud save. For a quick visual study, that is fine. For work you expect to revise tomorrow, it is a deal breaker.
The paper favors fitted studies over detailed navigation
Fixed aspect ratios include 1:1, 4:3, 3:4, 16:9, and 9:16. The long edge stays at 1,200 logical units, and rendering density can rise to 3 times the logical size for a dense screen. Existing artwork is fitted uniformly when the window or ratio changes, without cropping. The app retains source pixels between resizes until another edit is committed.
There are no zoom controls and no canvas scrolling. That makes the entire paper visible and keeps pointer mapping straightforward, but it rules out a common illustration habit: zooming into one corner for precise cleanup. The eraser is shared across every brush and acts on committed raster pixels. If you need editable objects, selectable strokes, or an infinite work area, tldraw or a vector editor is a better starting point.
The September 8 push is recent, but the project is still young
The current repository was created on September 3, 2026, and its last default-branch push was September 8. GitHub showed 186 stars and 3 open issues and pull requests on September 29; all 3 were pull requests, and there were no open issues in the API response. One proposes transparent PNG output, while another adds a new brush. There is no tagged release.
MIT licensing is clear, and the bundled font carries its own SIL Open Font License notice. The project also documents Node versions, the brush contract, shortcuts, query parameters, build commands, and browser testing. Those details make it easier to fork than its age suggests. Still, the absent CI workflow and release history leave maintenance expectations undefined. Our 23 passing tests support the current snapshot, not a promise about future compatibility.