world-class-web-design-os
Health Warn
- License — License: MIT
- Description — Repository has a description
- Active repo — Last push 0 days ago
- Low visibility — Only 5 GitHub stars
Code Pass
- Code scan — Scanned 12 files during light audit, no dangerous patterns found
Permissions Pass
- Permissions — No dangerous permissions requested
No AI report is available for this listing yet.
17 installable agent skills for distinctive web design: art direction, implementation, critique, accessibility, and visual QA for Codex, Claude Code, and Cursor.
World-Class Web Design OS
Design with a point of view. Ship with evidence.
An open operating system for distinctive, usable, accessible, and verifiable agent-made websites.
한국어 · Live overview · 60-second start · Same Brief Challenge · Contribute
18 seconds: one operating system, two deliberately different outputs. Explore the live proof.
World-Class Web Design OS installs 17 routed specialist skills into Codex, Claude Code, Cursor, GitHub Copilot, OpenCode, or the shared .agents/skills convention. It guides a coding agent through discovery, art direction, implementation, independent critique, accessibility, performance, and rendered visual QA.
It is not a component library or a list of style prompts. The output can look completely different from project to project because the system governs decisions and gates—not a house style.
Start in 60 seconds
- Install the skills:
npx --yes github:Lrinvl1203/world-class-web-design-os install --agent codex
Or use the universal Agent Skills CLI for another compatible agent:
npx --yes skills add Lrinvl1203/world-class-web-design-os -g
Start a new agent session so skill discovery refreshes.
Ask in one sentence:
Design and implement a distinctive website for this product using the installed Web Design OS, then verify it at mobile and desktop widths.
Broad website work routes through web-design-orchestrator; specialist skills load only when their job is relevant. To update an installation, append --overwrite. To install every declared target, use --agent all.
npx --yes github:Lrinvl1203/world-class-web-design-os doctor --agent codex
The doctor command confirms the expected 17 skills are discoverable. Both installers read this public GitHub repository directly; the universal CLI currently discovers all 17 skills without a duplicate skills/ tree.
What is actually in the repository
| Included | Why it matters |
|---|---|
| 1 orchestrator + 16 specialist skills | Keeps the active context focused while preserving a complete design workflow |
| 100 evidence-labeled references | Makes art-direction and technical choices searchable instead of relying on style adjectives |
| 2 runnable, contrasting website experiments | Shows that the system does not force every output into one house style |
| Playwright, axe, responsive, overflow, console, and reduced-motion gates | Turns “looks finished” into inspectable release evidence |
| Cross-agent installer and doctor | Gives Codex, Claude Code, Cursor, Copilot, OpenCode, and .agents users one maintained source |
What changes in practice
flowchart LR
A[Intent<br/>audience · job · action] --> B[Direction<br/>POV · anti-direction]
B --> C[System<br/>tokens · states · sources]
C --> D[Build<br/>semantic · responsive]
D --> E[Challenge<br/>critic · AI smell]
E --> F[Prove<br/>screens · a11y · perf]
F --> G[Learn<br/>reviewed proposals]
- The business job and primary action are resolved before styling.
- References are decomposed into principles instead of copied wholesale.
- Three materially different art directions are considered when the direction is open.
- Motion and component sources are chosen by communication intent, not fashion.
- Mobile receives a deliberate hierarchy and composition.
- Implementation and critique are separate passes.
- Completion requires screenshots, keyboard and accessibility checks, overflow checks, and a clean console.
- Daily learning produces evidence-backed proposals; it never silently rewrites core skills.
Proof you can open
![]() |
![]() |
| Nocturne Concierge Editorial hospitality · atmospheric restraint |
Linehold Forge Industrial commerce · dense utility |
| Experiment | Archetype | Internal score | Field evidence | Status |
|---|---|---|---|---|
| Nocturne Concierge | Editorial hospitality | 92.8 | Pending | Internal target passed |
| Linehold Forge | Industrial commerce | 91.8 | Pending | Hold; below 92 target |
Design Quality scores are internal rubric evaluations. They are not user research, conversion data, award results, or external expert validation. See public benchmark for the evidence model.
Test the claim yourself
Run The Same Brief Challenge: build once with your normal agent workflow, then in a clean session with Web Design OS using the same brief, assets, model, and time budget. Publish the raw browser checks, reviewer rationale, manual corrections, and limitations. Negative and partial results are welcome.
Submit a reproducible result or show what you built.
Public launch trail
These links prove where the project was introduced; they are not endorsements or external quality validation.
| Channel | Status | Link |
|---|---|---|
| Product Hunt | Launched Aug 11, 2026 | Launch page |
| X | Published | Announcement |
| Threads | Published | Launch post |
| Published | Architecture discussion | |
| GitHub | Open for feedback | Start-here discussion |
The machine-readable source of truth is config/distribution.json; CI rejects stale or contradictory launch records.
The 17-skill stack
The orchestrator keeps the active set small and activates specialists only when they have a job.
| Layer | Skills |
|---|---|
| Frame | design-discovery, reference-forensics |
| Direct | art-direction, design-system, component-source-router |
| Experience | interaction-design, motion-engine, creative-web |
| Build | frontend-implementation, responsive-recomposition |
| Prove | a11y-performance, visual-qa, design-critic, ai-smell-detector, visual-polish |
| Improve | continuous-learning |
| Route | web-design-orchestrator |
Read the architecture or inspect any skill in .codex/skills.
CLI
web-design-os route "editorial product site with scroll storytelling"
web-design-os search "typography motion" --limit 8
web-design-os setup .
web-design-os install --agent codex
web-design-os doctor --agent codex
web-design-os evolve --offline
web-design-os eval
routereturns at most three active specialists in addition to the orchestrator.searchqueries the 100-reference evidence atlas.setupcreates.web-design-os/project-context.mdwithout overwriting an existing brief.evolveruns the controlled learning pipeline;--offlineuses cached and manual sources.
Daily evolution without silent drift
The scheduled workflow collects official APIs, RSS feeds, and reviewed manual signals into a dated report. A proposal needs three independent sources and survives a 30-day cache policy before it can be considered. The automation may write evidence and proposals; it may not modify core skill instructions.
See evolution model, source registry, and learning records.
Every Monday at 09:10 Asia/Seoul, weekly-evidence-review combines the latest repository traffic, star/fork signals, distribution state, external curation PR status, validation results, and controlled-learning output into a private 90-day artifact. It creates a review queue but never edits skills, posts, commits, opens PRs, or self-merges.
Run locally
Requirements: Node.js 20+ and Python 3 for the full validation suite.
npm install
npm run validate
npm run eval:skills
npm run qa:install
npm run qa:launch
Preview the launch site:
node scripts/serve-static.mjs . 4173
Open http://127.0.0.1:4173/site/.
Boundaries
- World-Class Web Design OS does not guarantee awards, conversion lifts, or “world-class” outcomes.
- External libraries, references, and community signals remain subject to their own licenses and terms.
- Automated checks are necessary evidence, not a replacement for representative-user research.
- High-cost 3D, WebGL, smooth scrolling, or overlapping animation libraries require an explicit communication and performance case.
Third-party names identify sources only and do not imply affiliation. Experiment imagery and claims are fictional concept material. See notices and the IP/provenance audit.
Contributing and security
Start with CONTRIBUTING.md. Use the design-quality issue form for evidence-backed improvements. Please report security concerns through the process in SECURITY.md, not a public issue.
If this makes your next website more deliberate, star the repository so other builders can find it—and share the artifact or failure that should improve the system next.
MIT © 2026 Lrinvl1203
Reviews (0)
Sign in to leave a review.
Leave a reviewNo results found

