Projects

Each project started from a question that kept coming up in real engineering work. It gets a page once it is documented well enough for someone else to use: what it does, why it is built the way it is, and what it doesn’t do yet.

Methods

A way of working, written down well enough to adopt, with an implementation you can install.

FeatureKit

How much process does a coding agent actually need?

Plain-Markdown skills that give Claude Code, Codex and OpenCode a way of working. The agent agrees a written plan before it writes code, builds in small vertical slices, and records evidence for every acceptance criterion before it calls the work done. Its own rules change only when an eval, a trial or a real repo shows they should.

How it works Docs

v0.14.1 · Method · AI-assisted engineering · 30 skills · Claude Code, Codex, OpenCode · No CLI, server or runtime

1Clarify
2Plan
3Slice
4Approve
5Build
6Prove
7Review
8Hand off

Baseline, Tokens, Layouts, Behaviors

How do you keep a product's CSS in step with a design system that keeps changing?

Baseline, Tokens, Layouts, Behaviors. Four cascade layers and a few strict rules for writing CSS against them. Semantic HTML is styled before it has a class, components use tokens that name a meaning rather than a value, space comes from named patterns, and state lives in attributes. A design change then lands in one place and flows through.

How it works Docs Source on GitHub

v1.1.0 · Method · CSS and design systems · Four layers on npm · W3C design tokens · Any framework

@layer tokens, baseline, layouts, behaviors;

Tools

Something you run, documented with what it does and what it doesn’t do yet.

None published yet. A tool gets a page once someone else can install and use it.

Experiments

A question, a method and a result.

None written up yet. An experiment gets a page once it has a result, whether or not it worked.