Skela

Three.js gives you a full API and no opinion about how a scene is structured, so every project re-invents its own setup, its own scene graph conventions, and its own animation plumbing, and none of it travels to the next project. Skela is an attempt to write that shared structure once: a framework for composing objects and scenes through a clean class hierarchy, with no build step between the source and the browser.

A BaseObject class with a lifecycle interface, build, update, render, and dispose, plus a perform() action dispatcher that treats position, rotation, scale, and visibility as first-class actions. A base.jinja scene template uses Jinja block inheritance to generate the surrounding setup: renderer, camera, lighting, and orbit controls. A reference Human subclass shows how a skeletal hierarchy composes with a multi-channel animation and action system.

The framework is plain JavaScript with zero build tooling, so a scene is a folder of files the browser can load directly. The template layer, in Jinja2 served by Flask, fills in the boilerplate that would otherwise be copy-pasted per scene. The reference human demonstrates the whole model at once, and a companion Next.js site built on React Three Fiber wraps the same object model in an interactive demo: walk, run, kick, jump, wave, punch, dance, nod, bow, and spin.

The dispatcher and the multi-channel action system were designed separately and the seams show. I would define the action channel format first and make the dispatcher just one consumer of it.