Genjutsu - Setup Guide¶

Source: athevon/genjutsu via skills.sh - 18.1K combined installs across 20 indexed listings; first documented November 4, 2026 (morning sweep) GitHub: 741 stars, NOASSERTION licence (the repo root ships an MIT LICENSE file), last pushed 2026-11-03, 20 SKILL.md directories Category: Front-end craft for agents: layout systems, motion, component polish and design review in one skill library Quality Tier: 🟡 Yellow: paint draws Socket and Snyk Warns, and GitHub reports the licence as NOASSERTION even though the repository ships an MIT LICENSE file

Genjutsu is Adrien Thevon's creative coding library for Claude Code, claude.ai and Cowork. The 20 indexed skills cover web work (React, Vue, Svelte, Astro, vanilla CSS, Three.js and Canvas), Android (Jetpack Compose and Compose Multiplatform) and Apple (SwiftUI on iOS and macOS). It is one of the few packs in this catalog that treats motion, graphics and layout as one subject across three platforms.

Three listings are the entry points. cast casts an illusion on an existing UI: it scans the stack, proposes an interaction thesis and loads the sub-skills it needs. paint builds a visual universe from nothing: art direction, design system, implementation, audit. bunshin runs the whole pipeline for a site or app with parallel agents. Everything else is a sub-skill or an internal module that those three load on demand, which is why the lowest install counts here belong to the most specialised pieces.

Installation¶

The README documents the skills CLI route first, and it is specific about the URL:

npx skills add https://genjutsu.athevon.dev -g

After that, describe the task in Claude Code or type /genjutsu. The README asks for the -g flag on updates too, and warns against the repository route because it installs the skills without their modules:

npx skills update -g

The Claude Code plugin route exposes three named commands:

/plugin marketplace add AThevon/genjutsu
/plugin install genjutsu

That gives /genjutsu:cast, /genjutsu:paint and /genjutsu:bunshin inside a session. Project scope, the claude.ai and Cowork paths, and the extra pieces bunshin needs are documented on the project's install page. For Hermes, copy the skills/ tree into the profile skills directory and keep the _jutsu/ subdirectory intact, since cast and paint resolve their modules from there.

What It Provides¶

The suite splits into three entry points, platform sub-skills and internal modules.

Skill	Installs	Use For
paint	1,421	Art direction brainstorm, design system, implementation and audit in one anti-slop pipeline
cast	1,421	Scans the stack, proposes an interaction thesis, loads sub-skills and implements the motion
motion-principles	1,135	Timing, easing, enter and exit patterns, accessibility and performance budgets
css-native	1,056	Zero-dependency animation with scroll-driven CSS, View Transitions and @starting-style
threejs-r3f	1,026	Three.js and React Three Fiber scenes, shaders and postprocessing
design-audit	1,023	Checklist pass over motion gaps, accessibility, colour, responsive and performance
canvas-generative	1,001	Canvas 2D generative art: particles, flow fields, noise, fractals and L-systems
gsap	999	GSAP core, timelines, ScrollTrigger and plugins
framer-motion	991	AnimatePresence, layout animations, gestures and motion values
desktop-principles	984	Hover states, pointer precision, shortcuts, multi-window and focus management
mobile-principles	973	Touch targets, thumb zones, safe areas, gestures and mobile performance budgets
compose-multiplatform	949	expect/actual composables, density and font handling, iOS, Android and desktop interop
ui-ux-pro-max	948	Searchable design-system dataset of 84 styles and 192 palettes; loaded by cast and paint
compose-motion	943	animateAsState, AnimatedVisibility, Crossfade and shared element transitions
swiftui-motion	942	withAnimation, transitions, matchedGeometryEffect, PhaseAnimator and springs
swiftui-graphics	939	Metal shaders, visualEffect, Liquid Glass on iOS 26, Canvas and CRT effects
compose-graphics	936	Material 3 Expressive motion physics, AGSL shaders and DrawScope generative work
bunshin	380	Full site pipeline: real material, parallel builds, fix loops and a cold-eyes verdict
design-excellence	2	No directory in the current skills/ tree; the listing carries install history only
creative-excellence	1	The suite's former name per the changelog; no directory in the current tree

All 20 indexed listings appear in the table above.

Why This Matters for Hermes Agents¶

Most design skill packs are prose about taste. This one is organised by platform, so a Hermes agent working on a SwiftUI screen and one working on a Compose screen load different sub-skills that share the same vocabulary.

The entry points force a decision before code. cast scans the stack and proposes an interaction thesis, and the internal tells module names the reflexes a model falls into when nothing asked for them: invented data, decorative filler, converged layouts, hollow copy. That is a review an agent cannot run on its own taste, and it is the part worth borrowing even if you never install the rest.

Platform depth is unusual for a library this size. Android and Apple each get their own motion skill, graphics skill and platform-principles skill, rather than a single mobile afterthought.

ui-ux-pro-max is worth having for its data alone: 84 styles, 192 palettes, 74 font pairings, 25 charts and 99 UX guidelines across 22 stacks, shipped as a searchable dataset plus a CLI. It is an internal module, loaded by cast and paint rather than invoked directly.

The last useful signal is honesty about results. The README publishes per-run receipts with cost, turn counts and what each run left unverified, and it says plainly that measured gains over a page built without genjutsu are modest. Treat the suite as craft scaffolding, not a guarantee.

Usage¶

You say	What happens
"Make this dashboard feel alive without new dependencies"	cast scans the stack, states an interaction thesis, then applies css-native timing and transitions
"Give this product a look of its own"	paint runs art direction, builds the design system, implements it and audits the result
"Build the whole landing site from our brand material"	bunshin harvests the material, builds in parallel and loops fixes to a verdict
"Why does this page feel generic?"	tells names the specific tells, and design-audit returns the checklist gaps
"Add a shader background to the hero"	threejs-r3f on web, swiftui-graphics on Apple
"Our Compose screen drops frames on scroll"	compose-motion and mobile-principles work the animation and the performance budget
"Check the motion for accessibility"	motion-principles covers reduced motion and the audit checks the same path

Verification¶

Confirm the skills landed in the agent skills directory:

ls ~/.claude/skills | sort

Then fetch one raw file before installing anything. This path is skills/_jutsu/motion-principles/SKILL.md on branch main and returned 200 when this guide was written:

curl -fsSL https://raw.githubusercontent.com/athevon/genjutsu/main/skills/_jutsu/motion-principles/SKILL.md | head -20

Frontmatter with a name and a description confirms the indexed listing is backed by a real file. Repeat with skills/paint/SKILL.md and skills/bunshin/SKILL.md, which sit outside the _jutsu directory. The install index itself is at https://www.skills.sh/athevon/genjutsu.

Security¶

skills.sh verdicts for sampled skills (verified November 4, 2026) - re-check the security pages on skills.sh before production use:

Skill	Gen Agent Trust Hub	Socket	Snyk
paint	Pass	Warn	Warn
design-audit	Pass	Pass	Pass
mobile-principles	Pass	Pass	Pass
swiftui-graphics	Pass	Pass	Pass
creative-excellence	Pass	Pass	Pass

paint is the only flagged skill in the sample and carries two Warns. It is also the skill that writes most freely, since its job is to create a design system and implement it across a project, so scope it to a branch you can discard.

Integration with CorpusIQ¶

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects

Limitations¶

Snapshot data, verified November 4, 2026: 18,070 combined installs across 20 indexed listings; 741 GitHub stars; NOASSERTION licence; last pushed 2026-11-03. Counts drift over time.
GitHub reports the licence as NOASSERTION even though the root LICENSE file is the MIT text (Copyright (c) 2026 Adrien Thevon). Read the file before redistributing; do not rely on the badge alone.
The install index and the tree do not fully agree. design-excellence and creative-excellence carry install counts but have no directory in the current skills/ tree, while orchestration and tells are in the tree and carry no indexed listing. The README says creative-excellence was the suite's name up to version 2.
bunshin is web only for now, per the README, and it depends on the modules the site-URL install provides; a repository-route install leaves it weaker.
Several listings are internal modules. ui-ux-pro-max, tells and orchestration are loaded by cast, paint or bunshin, so installing them on their own buys little.
Install counts cluster tightly: 17 of the 20 listings sit between 936 and 1,421 installs, so the order inside the suite carries little signal.

Pro Tips¶

Install from the site URL with -g, not from the repository name. The README warns that the repository route drops the modules.
Load one entry point, not the whole set. cast and paint pull what they need; preloading all 20 spends context on platforms you are not touching.
Ask for the interaction thesis in writing before implementation. It is the cheapest place to catch a direction you did not want.
Run design-audit at the end of a build, not the start. It is a checklist over finished output, not a plan.
Keep the eval receipts in view: the author reports modest gains over an unskilled control, so judge the page rather than the claim.

← Skills Catalog | Skills Marketplace → Powered by CorpusIQ