Webgpu

A focused skill for building WebGPU applications, from device initialization and buffer layouts to render pipelines and compute passes. You get guidance on WGSL shader authoring, workgroup sizing, and synchronization patterns. The skill pushes you toward modular pipeline design and includes practical advice on fallback strategies when WebGPU isn't available. It covers common use cases like grid simulations and GPU compute workloads, with specific attention to avoiding performance pitfalls like full GPU readbacks in hot paths. The included reference materials and examples give you working code to adapt rather than starting from scratch every time.

skills.sh installs: 873
GitHub stars: 35
last updated: Aug 21, 2026

View on GitHub

Install

$ npx -y skills add cazala/webgpu-skill --skill webgpu --agent claude-code

Installs into .claude/skills of the current project.

Files

WebGPU

Build the smallest observable GPU result first, then scale it. Treat every WebGPU feature as five explicit contracts:

Capability contract: baseline, optional features and limits, fallback.
Data contract: WGSL types, byte offsets, strides, usage flags.
Pass contract: reads, writes, ordering, ping-pong ownership.
Presentation contract: canvas format, physical size, alpha and color.
Lifecycle contract: creation, resize, errors, loss and destruction.

Copy the full Vite starter for a new browser project. It contains a typed compute-and-render particle field with ping-pong buffers, compilation diagnostics, resize handling, pointer interaction, device-loss reporting, and cleanup.

Work from a resource and pass graph

Before writing WGSL, list persistent resources and pass ownership:

CPU seed ─writeBuffer─> state A
                         │
state A ─compute step─> state B ─render─> canvas
state B ─compute step─> state A ─render─> canvas

For every pass, record:

SKILL.md on GitHub

Details

Repository: github.com/cazala/webgpu-skill
Categories: Git & Pull Requests, Code Review & Quality, Debugging, Automation & Workflows, Design & UI/UX
First seen: May 16, 2026