Skip to content

rgranet/designlint-skill

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

17 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

DesignLint

Force distinctive, human-feeling UI/UX design before any AI agent writes code. Audit and improve existing interfaces.

By Ruben Granet · v1.2.0 · MIT License


The Problem

AI coding agents converge. Left unconstrained, they always produce the same thing:

UI: Inter font · Purple/blue gradients · White background · rounded-xl cards · Opacity fade-ins

UX: Tab bar (5 icons) · List → Push detail · Modal for everything · Green toast on success · 3-screen onboarding carousel

Voice: "Submit" · "An error occurred" · "No items found" · "Success!" · (spinner)

This isn't a style preference — it's a structural failure. Every AI-generated interface looks, feels, and sounds like every other AI-generated interface.

DesignLint breaks that convergence with two modes:

  • Create — Understand the user, make 9 deliberate design decisions, resolve a creative tension, and critique the result before shipping.
  • Audit — Diagnose an existing interface against the same standards, score it, and produce a concrete improvement plan.

Two Modes

Create Mode

"Build me a task management app"

The agent stops before writing code. It identifies the user and their emotional state, picks a creative tension, selects an archetype as a departure point (not a template), makes 9 design decisions covering aesthetics through voice, and runs a human critique before shipping.

Audit Mode

"What's wrong with this UI?" / "Improve this interface" / "Review this design"

The agent documents what exists, runs the full convergence scan (UI, UX, and human), scores the design 1–5, produces an improvement brief with the top 3 highest-impact changes, and optionally executes the improvements.


Before / After

Without this skill, an AI agent asked to "build a task management app" produces:

UI:  Inter · #7C3AED primary · white bg · rounded-xl cards · fade-in
UX:  Tab bar (Home, Tasks, Calendar, Search, Profile) → List → Modal form → Toast
Copy: "Submit" · "No tasks found" · "Task created successfully"

With this skill, the same prompt first produces a design brief:

User:       Freelance designer juggling clients — overwhelmed, needs calm control
Tension:    Dense BUT calm
Archetype:  Things 3 — Keep: today-first, Discard: system blue, Add: warm editorial type
Palette:    Chalk & Ink (#F5F0E8 bg / #1A1209 text / #C85A38 accent)
Type:       Cormorant Garamond display / Karla body
UX:         Today-First nav · Inline editing · Optimistic UI · Progressive onboarding
Voice:      Warm + brief. CTA="Done" · Empty="A clear day — nice." · Error="Lost that one. Undo?"
Departure:  Priority indicated by type weight (light→bold), not color badges

The code that follows is intentional, empathetic, and distinctive.


What's Inside

designlint/
├── SKILL.md                        # Core protocol (Create + Audit modes)
├── AGENTS.md                       # Drop-in for OpenAI Codex
├── .cursorrules                    # Drop-in for Cursor
└── references/
    ├── archetypes.md               # Design vocabulary library
    ├── ux.md                       # UX paradigm library
    ├── web.md                      # React / Next.js / HTML patterns
    ├── ios.md                      # SwiftUI patterns
    ├── compose.md                  # Android / Jetpack Compose patterns
    ├── react-native.md             # React Native + Reanimated patterns
    ├── dashboards.md               # Data UI & dashboard patterns
    └── persuasion.md               # CRO, copy conversion, psychology principles

Install

Claude Code (native)

claude skill install designlint.skill

Cursor

cp designlint/.cursorrules ./
cp -r designlint/ ./designlint/

OpenAI Codex

cp designlint/AGENTS.md ./
cp -r designlint/ ./designlint/

Manual (any agent)

Add the designlint/ folder to your project and reference SKILL.md in whatever instruction mechanism your agent supports.


How It Works

Create Mode — 5 phases, 9 decisions

Phase 0 — Understand the human (user context, creative tension, archetype as departure point)

Phase 1 — 9 decisions:

# Decision What it breaks
1 Aesthetic Axis Generic "Modern SaaS" tone
2 Typography Contract Inter, Roboto, system-ui
3 Color System Purple gradients, safe blues
4 Spatial Grammar Centered column, uniform padding
5 Motion Character Uniform 300ms opacity fades
6 Signature Detail Forgettable, invisible interfaces
7 UX Paradigm Tab bar, modals, global toasts
8 Voice & Tone "Submit", "An error occurred", robot-speak
9 The Departure Reproducing archetypes instead of designing

Phase 2 — Document the design brief (comment block, auditable)

Phase 3 — Build with platform patterns (adapt, don't copy)

Phase 4 — Anti-convergence audit (UI + UX + accessibility)

Phase 5 — Human critique (empathy, craft, honesty — iterate if flat)

Audit Mode — 5 steps

Step What happens
A1 Capture what exists (fonts, colors, layout, UX, voice)
A2 Convergence diagnosis (UI + UX + human scans)
A3 Score 1–5 (Generic → Human)
A4 Improvement brief (top 3 changes + UI/UX/voice fixes + Departure)
A5 Execute if asked (brief → Create Mode Phase 3)

Platform Coverage

Platform Reference file Status
Web (React/Next.js/HTML) references/web.md ✅ Full
iOS (SwiftUI) references/ios.md ✅ Full
React Native references/react-native.md ✅ Full
Android (Jetpack Compose) references/compose.md ✅ Core
Dashboards / Data UIs references/dashboards.md ✅ Full
Persuasion & Conversion references/persuasion.md ✅ Full

Agent Compatibility

Agent Install method Status
Claude Code Native .skill install
Cursor .cursorrules
OpenAI Codex AGENTS.md
Windsurf .windsurfrules (same as .cursorrules)
Other agents Manual system prompt

Contributing

Issues and PRs welcome. The most impactful contributions are:

  • New archetypes in references/archetypes.md
  • New UX paradigms in references/ux.md
  • Platform references (improvements to existing or new platforms)
  • Before/after examples showing the skill's impact

License

MIT — see LICENSE

About

Stop AI-generated interfaces from all looking the same. DesignLint forces AI agents to understand the user, make 9 deliberate design decisions, and critique the result before writing code. Two modes: Create from scratch, or Audit and fix existing UIs.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors