For non-designers who vibe-code the front endOpen source / MIT

Design with a
point of view.
Ship with evidence.

Your agent can generate UI. The hard part is knowing why it feels wrong—and how to fix it. This operating system routes the work through art direction, implementation, independent critique, accessibility, and rendered browser QA.

npx --yes github:Lrinvl1203/world-class-web-design-os install --agent codex

RUN / WEBSITESYSTEM READY
  1. 01Intentjob · audience · action
  2. 02DirectionPOV · anti-direction
  3. 03Buildsemantic · responsive
  4. 04Challengecritic · AI smell
  5. 05Provescreens · a11y · perf

Critical content survives without motion. Every flourish needs a communication job.

17specialist skills
100studied references
6QA viewports
5verified archetypes
Dailycontrolled learning

01 / First successful use

One command.
One new session.
One normal sentence.

No design vocabulary is required. Install the system, refresh skill discovery, and describe the website you need in plain language.

  1. 01 / TERMINAL

    Install the 17 skills globally.

    npx --yes github:Lrinvl1203/world-class-web-design-os install --agent codex

    Paste this into a terminal inside or outside your project. It does not modify the website yet.

  2. 02 / REFRESH

    Start a new Codex session.

    web-design-os doctor --agent codex

    The doctor should report 17/17 READY. A new session lets Codex discover the installed skills.

  3. 03 / ASK NORMALLY

    Give it the real product job.

    Use $web-design-orchestrator to redesign and implement this website for [audience], then verify it on mobile and desktop.

02 / See it work

Fifteen seconds.
Five different outcomes.

The same operating system produced five deliberately different website archetypes. The demo proves operation and range—not user validation or award status.

PUBLIC TEST / SAME INPUTS

Compare your normal workflow with Web Design OS using the same brief, assets, agent, model, and time budget. Publish the raw evidence—even when the OS loses.

Run the Same Brief Challenge ↗

03 / Rendered proof

Inspect the output,
not the adjective.

Five deliberately different archetypes exercise the same system. Scores are internal Design Quality evaluations, not field validation.

Nocturne Concierge editorial dark hospitality website shown at desktop width Open experiment ↗
CASE 01

Nocturne Concierge

Editorial hospitality · atmospheric restraint · intent-led motion

Internal score
92.8
Field evidence
Pending
Linehold Forge high-contrast industrial commerce website shown at desktop width Open experiment ↗
CASE 02

Linehold Forge

Industrial commerce · dense utility · responsive recomposition

Internal score
91.8
Status
Hold
Afterimage Atlas editorial cultural object study shown at desktop width Open experiment →
CASE 03

Afterimage Atlas

Editorial culture · object-led inspection · five-state narrative

Internal score
92.7
Field evidence
Pending
Sequence Desk operational SaaS interface shown at desktop width Open experiment →
CASE 04

Sequence Desk

Product UI · component states · accountable launch workflow

Internal score
92.8
Field evidence
Pending
Orbital Commons civic data and pseudo-3D heat comparison shown at desktop width Open experiment →
CASE 05

Orbital Commons

Civic data · purposeful pseudo-3D · semantic HTML fallback

Internal score
94.3
Field evidence
Pending

04 / Operating method

One orchestrator.
Specialists on demand.

The system activates only the skills needed for the job, keeps implementation and critique separate, then records reusable defects through a controlled proposal loop.

DESIGN DISCOVERY

Find the job before the style.

Define the audience, primary action, content hierarchy, constraints, and questions that must be true before visual work begins.

Exit gateA testable brief with one primary action

05 / The difference

A system, not a style vending machine.

Typical prompt pack

  • Style words before product intent
  • One implementation pass
  • Desktop screenshot as completion
  • Uncontrolled inspiration intake

World-Class Web Design OS

  • Business job and anti-direction first
  • Independent critique and AI-smell audit
  • Responsive, keyboard, axe, overflow, and console gates
  • Evidence-backed learning proposals—never silent mutation

06 / Install details

Install once.
Ask normally.

After installation, a plain-language website request automatically starts with the orchestrator.

01 / RUN
npx --yes github:Lrinvl1203/world-class-web-design-os install --agent codex
02 / ASK
“이 제품의 웹사이트를 디자인하고 구현해줘.”

The skill system is installed to your global Codex skills directory. Add --overwrite when updating an existing installation.

07 / Trust model

The gates are part
of the design.

A

Accessible by construction

Semantic structure, keyboard reachability, visible focus, target sizing, reduced motion, and axe checks are release gates.

B

Responsive by recomposition

Mobile is assigned a new hierarchy and interaction model instead of receiving a mechanical desktop stack.

C

Learning under review

Daily sources create dated proposals. Core skills change only after evidence, regression checks, and human adoption.

D

Claims labeled honestly

Internal scores, automated checks, and field evidence are separate signals. One never impersonates another.

08 / Before you install

Useful boundaries,
not bigger promises.

The OS is strongest as a governed workflow. It does not replace product judgment, user research, or a designer with real domain context.

Is this another component library or style prompt collection?

No. It can choose native HTML, custom components, registries, motion, or purposeful 3D after the user job and art direction are clear. The reusable product is the decision and verification workflow.

Is it universally better than UI prompt packs or design courses?

No universal claim is made. Its narrower advantage is connecting discovery, art direction, implementation, independent critique, responsive work, accessibility, browser QA, and controlled learning in one route.

Does “daily learning” silently rewrite the skills?

No. Trusted feeds and authorized APIs produce review artifacts. A recurring principle still needs three independent sources, counterexamples, regression scenarios, and an intentional reviewed change.

Why must I start a new session after installation?

Agent skill discovery happens when the session starts. Opening a fresh session prevents an old capability list from making the new skills appear missing.

Do I need animation or 3D for every site?

No. CSS or no motion is often the correct choice. 3D, WebGL, and heavier animation engines are used only when they communicate the product or brand idea and survive accessibility and performance review.

OPEN SOURCE / MIT

Make the next website
harder to ignore.

Star / view on GitHub ↗ Install the system ↑ Share a build or failure ↗