Your AI agent writes flawless code. Now teach it taste.

DesignOS is the open-source operating system that gives coding agents design intelligence — 60+ opinionated modules, a self-scoring review engine, and a memory that never forgets a decision.

Install in one command See what it builds →

MIT · markdown-native · works with Claude Code, Cursor & any agent

you › Design a Stripe-level SaaS landing page.

designos boot · kernel v1.2
 routed: landing-pages · hero · persuasion
 sector: industries/saas.md
 memory: brand.md · design.md loaded
 design loop: research → wireframe → ui …
 review: contrast 4.2:1 on muted text — fixing
 re-scored: UI 96 · UX 97 · A11y 96 · Perf 97
 shipped · decisions → memory/design.md
60+knowledge modules
24industry playbooks
9specialist agents
95the score threshold
1command to install
The system

Five layers between a prompt and a shipped screen

Not a prompt pack — an operating system the agent boots into, routing only the modules each task needs.

01 · KNOWLEDGE

Opinionated rules, not vibes

Typography scales, spacing grids, 21 component specs, persuasion psychology, motion tokens — every rule with its reason.

foundations/ components/ psychology/ motion/ patterns/ native/
02 · LOOPS

Process that catches what talent misses

Research → wireframe → UI → review → accessibility → performance → SEO → refactor. No stage skipped, ever.

loops/ workflows/ checklists/
03 · REVIEW ENGINE

An adversarial critic with veto power

Six dimensions, hard 95 threshold, instant-fail caps for contrast and fake data. Under 95 goes back — before you ever see it.

scoring/ validators/ evals/
04 · MEMORY

Decision #40 agrees with decision #4

Seven files per project — brand, design, pages, todos, bugs. Every choice recorded with its reason, forever.

memory/
05 · INDUSTRY INTELLIGENCE

Fintech isn't gaming isn't healthcare

24 sector playbooks: buyer psychology, visual language, trust requirements, and what each industry instantly punishes.

industries/
+ · AGENT STUDIO

Nine specialists, one veto

Creative director, UX researcher, a11y specialist, copywriter, reviewer — real Claude Code subagents, one flag away.

agents/
The review engine

Nothing ships on vibes

Every deliverable is graded across six dimensions against a written rubric. Accessibility failures cap the score at 60 — no exceptions, no negotiation. Under threshold, the work returns to the loop with specific objections.

Read the rubric →

UI Craft96 / 100
UX & Flow97 / 100
Accessibility96 / 100
Performance97 / 100
Modernity96 / 100
Conversion95 / 100
Gateall ≥ 95 → SHIP
Proof, not promises

Four pages, one token system, zero hand-tuning

A complete fictional product designed by the system — view source on any of them: every value resolves to a token, every decision traces to a module.

Install

One command. Any agent.

$ npx github:ardamoustafa1/DesignOS init --agents
 DesignOS copied to ./DesignOS
 CLAUDE.md wired
 9 subagents registered

Then ask for anything — "design a pricing page for a fintech app, dark theme" — and watch it route modules, run the loop, and refuse to hand you anything under 95.

Open source · MIT

Make your agent dangerous

Star DesignOS on GitHub

150+ files · 8,600+ lines of design judgment · yours