genjutsu

Overview

What genjutsu is, how its pieces fit together, and the shortest path from installing it to seeing something move.

All documentation

genjutsu is a skill: a body of instructions Claude loads and follows. It is not a library you import, not a component kit, and not a code generator with templates. Nothing from genjutsu ships in your bundle. What lands in your project is ordinary code written in the idiom your project already uses.

Its subject is the part of interface work that models are worst at: motion, visual identity, and the discipline to hold a position instead of producing something competent and forgettable.

The shape of it

Two skills are the entry points. Fifteen modules sit behind them and load only when they apply.

Piece What it is
/genjutsu:cast Gives an existing interface motion. Scans, states a thesis, implements, audits.
/genjutsu:paint Builds a visual universe from nothing. Brainstorm, two theses, design system, implement, audit.
Fifteen modules Focused references on motion foundations, per-stack APIs, platform UX and auditing. Loaded on demand, never invoked directly.

The split matters. cast adapts to what you already have. paint replaces the visual layer and writes a design system into your repository so the direction survives past the session that created it.

Where it runs

Both surfaces, from one installation source.

  • Claude Code, as a plugin, invoked with slash commands.
  • claude.ai on web and in the apps, as an uploaded skill, invoked by name or just described in plain language.

And three runtimes underneath: web, Apple (SwiftUI on iOS and macOS), and Android (Jetpack Compose, including Compose Multiplatform). Which one you get is detected from your project, not asked.

The shortest path

  1. Install it. Two slash commands in Claude Code, or one zip upload on claude.ai.
  2. Open a project with an interface in it.
  3. Ask for something concrete. “Make the hero feel alive on scroll” is enough to start; the skill will ask before it commits.
  4. Read the thesis it proposes. That single sentence is your decision point, and it exists so you never discover the direction after the code is written.
  5. Say yes, or say what feels wrong. Rejecting a thesis costs a sentence. Rejecting an implementation costs an afternoon.

What it refuses to do

Worth knowing up front, because these are deliberate and they are not configurable away.

  • It will not write code before a thesis is validated.
  • It will not install a dependency without asking.
  • It will not replace an animation library you already chose.
  • It will not skip the accessibility and performance audit, including when the work looks finished and everybody is happy.

The reasoning behind each of those is in Principles.