gpt-taste
gpt taste
design-everyday-things
Apply foundational design principles: affordances, signifiers, constraints, feedback, and conceptual models. Use when the user mentions "why is this confusing", "affordance", "error prevention", "discoverability", "human-centered design", "fault tolerance", "mental model", "mapping", or "seven st...
Full skill instructions
Foundational design principles for creating products that are intuitive, discoverable, and understandable. The "bible of UX" — applicable to physical products, software, and any human-designed system.
Good design is actually a lot harder to notice than poor design, in part because good designs fit our needs so well that the design is invisible. When something fails, users blame themselves — but the fault is almost always in the design. Great design bridges the gap between what people want to do and what the product allows: it is discoverable (you can figure out what to do) and understandable (you can figure out what happened).
Goal: 10/10. Score 2 points per satisfied row of the Quick Diagnostic (5 rows = discoverability, evaluation, error recovery, mapping, constraints). Bands: 9-10 = users act without instructions, understand every outcome, and recover from any error; 5-6 = one gulf or error path is broken; <=3 = users must consult a manual or routinely blame themselves. Report the current score and the diagnostic rows failing it.
Every interaction with a product requires bridging two gulfs:
USER PRODUCT
│ │
├──── Gulf of Execution ────────────────→│
│ "How do I do what I want?" │
│ │
│←──── Gulf of Evaluation ──────────────┤
│ "What happened? Did it work?" │
The gap between what users want to do and what the product lets them do. Users ask: What can I do here? Which control do I use?
Bridge with: clear signifiers, natural mappings, constraints, familiar conceptual models.
The gap between what the product did and what users understand happened. Users ask: What happened? Did it work? What state is the system in?
Bridge with: immediate visible feedback, clear system-state indicators, meaningful error messages, progress indicators.
Design goal: Make both gulfs as narrow as possible — action and understanding should be immediate.
See: references/two-gulfs.md for gulf analysis exercises.
Definition: Can users figure out what actions are possible and how to perform them? Its five components — affordances, signifiers, constraints, mappings, feedback — are detailed below.
Test: Put a new user in front of your product. If they can't figure out what to do within 10 seconds, discoverability is broken.
Anti-pattern: "The user manual explains it." If users need a manual, the design failed.
Definition: The relationship between an object's properties and a user's capabilities that determines how the object could be used.
Key insight: Affordances exist whether or not they are perceived — what matters for design is perceived affordance.
| Type | Definition | Example |
|---|---|---|
| Real | Physical capability exists | A button affords pressing |
| Perceived | User believes capability exists | A raised area looks clickable |
| Hidden | Exists but isn't obvious | Right-click context menu |
| False | Appears to afford action but doesn't | Decorative element that looks clickable |
| Anti-affordance | Prevents action | A barrier that blocks movement |
Digital applications:
| Element | Affordance | How to Signal |
|---|---|---|
| Button | Clicking/tapping | Raised, colored, shadow, hover state |
| Text field | Text input | Border, placeholder text, label |
| Scroll area | Scrolling | Scroll bar, fade at edge, partial content |
Common failures: flat design erasing perceived affordances (button or label?), too-small touch targets, interactive and decorative elements that look identical.
See: references/affordances.md for affordance design patterns.
Definition: Signals that communicate where the action should take place. Affordances determine what you CAN do; signifiers show you WHERE and HOW.
| Type | Definition | Example |
|---|---|---|
| Deliberate | Designed to communicate | "Push" label on door, placeholder text |
| Accidental | Unintentional but informative | Worn path in grass (people walk here) |
| Social | Other people's behavior | Line of people indicates entrance |
Digital signifiers:
| Signifier | What It Communicates | Example |
|---|---|---|
| Cursor change + hover state | This is interactive | Pointer → hand on links; button color change |
| Icons + labels | Function of the element | Magnifying glass = search; "Submit", "Cancel" |
| Color + position | Status, category, hierarchy | Red = error, green = success; close button top-right |
Design rule: When in doubt, add a signifier — better to over-communicate than leave users guessing.
See: references/signifiers.md when deciding which signifier to add to an unclear control.
Definition: The relationship between controls and their effects. Natural mapping means the spatial layout of controls matches the layout of what they control.
| Mapping Quality | Example | Why It Works/Fails |
|---|---|---|
| Natural | Volume slider (up = louder) | Matches mental model |
| Poor | Light switch panel | No spatial correspondence to lights |
| Poor | Stovetop knobs in a row | Layout doesn't match burner positions |
Digital principles: controls near what they affect, layout mirroring content, direction matching expectation (scroll down = content moves up), related controls grouped.
| Technique | How It Works | Example |
|---|---|---|
| Proximity | Control near target | Edit button next to content |
| Spatial | Layout mirrors real world | Map controls match compass directions |
| Cultural | Follows conventions | Red = stop/danger, green = go/safe |
| Sequential | Follows natural order | Steps 1, 2, 3 left to right (or top to bottom) |
See: references/mappings.md for mapping analysis exercises.
Definition: Limiting the possible actions to prevent errors.
| Type | Mechanism | Example |
|---|---|---|
| Physical | Shape/size prevents wrong action | USB plug only fits one way |
| Cultural | Social norms guide behavior | Red means stop, green means go |
| Semantic | Meaning restricts options | A rearview mirror only makes sense facing backward |
| Logical | Logic limits choices | Only one hole left for the last screw |
Digital constraints:
| Constraint | Implementation | Example |
|---|---|---|
| Input validation | Restrict what can be entered | Date picker vs. free text |
| Disabled states | Gray out unavailable options | "Submit" disabled until form valid |
| Forced sequence + undo | Steps in order; allow reversal | Wizard with locked steps; Gmail "Undo send" |
Design rule: Every constraint you add is one less error the user can make — make wrong actions impossible rather than punishing them.
See: references/constraints.md for constraint design patterns.
Definition: Communicating the results of an action back to the user. Feedback must be immediate (within 0.1s for direct manipulation), informative, appropriately dosed, and non-intrusive.
| Type | When to Use | Example |
|---|---|---|
| Visual | Most actions | Button press animation, color change, checkmark |
| Auditory | Important events, confirmations | Success chime, error sound |
| Haptic | Touch devices, confirmation | Vibration on key press |
| Progress | Long operations | Progress bar, spinner, skeleton screen |
Digital feedback patterns:
| Situation | Feedback Needed | Example |
|---|---|---|
| Form submission | Success/error message | "Saved!" toast or inline error |
| Loading | Progress indicator | Spinner, skeleton screen, percentage |
| Error | What went wrong + how to fix | "Invalid email. Please check format." |
Response times: 0.1s feels instantaneous; 1s is a noticeable delay (change cursor); 10s loses attention (show progress bar); over 10s users leave (show percentage, allow backgrounding).
Common failures: no feedback (did my click register?), delayed feedback (feels broken), unclear feedback, alert overload.
See: references/feedback.md when an action gives no clear result and you need the right feedback type and timing.
Definition: The user's mental model of how a product works.
| Model | Held By | Description |
|---|---|---|
| Design model | Designer | How the designer thinks it works |
| User's model | User | How the user thinks it works |
| System image | Product | What the product actually communicates |
Goal: The user's model should match the design model; the system image is the only bridge. Matching models let users predict outcomes and recover from errors; mismatches breed confusion, self-blame, and support calls.
Example (thermostat): design model — set a temperature, the system maintains it; common user model — higher setting heats faster (wrong), so users crank it to 90°F.
Build correct models with: familiar metaphors (desktop, trash), visible system state, clear feedback, consistent behavior, progressive disclosure.
See: references/conceptual-models.md when the user's model diverges from how the product works. For fully worked teardowns (door handles, thermostats, digital products), see references/case-studies.md.
Norman's key insight: there is no such thing as "human error" — only bad design. When someone errs, look for the design flaw, not the person's flaw.
Slips — correct intention, wrong action:
| Slip Type | Cause | Example | Design Fix |
|---|---|---|---|
| Action slip | Wrong action on right target | Click "Delete" instead of "Edit" | Separate destructive actions |
| Memory lapse | Forget step in sequence | Forget attachment after writing "attached" | Gmail's attachment reminder |
| Mode error | Right action, wrong mode | Type in caps lock | Show mode state clearly |
| Capture error | Habit overrides intention | Drive to old office on autopilot | Interrupt at decision points |
Mistakes — wrong intention, executed correctly:
| Mistake Type | Cause | Example | Design Fix |
|---|---|---|---|
| Rule-based | Apply wrong rule | Use formula for wrong situation | Provide context, confirm |
| Knowledge-based | Incomplete/wrong mental model | Misunderstand how system works | Better conceptual model |
| Memory lapse | Forget goal or plan | Forget why you opened the fridge | Reminders, history |
Prevent: constraints that make errors impossible, undo/redo everywhere, confirmation for destructive actions, sensible defaults, forgiving input. Recover: clear error messages, never erase the user's work, partial saves, easy reset to a known good state.
Error message checklist:
See: references/human-error.md for error prevention patterns.
Norman's model for how humans interact with products:
1. GOAL → "I want to adjust the temperature"
2. PLAN → "I'll use the thermostat"
3. SPECIFY → "I'll press the up arrow"
4. PERFORM → (presses button)
─── Gulf of Execution ───
5. PERCEIVE → (sees display change)
6. INTERPRET → "The number went up"
7. COMPARE → "Is this what I wanted?"
─── Gulf of Evaluation ───
Design implications: support stages 1-3 with signifiers, mappings, and constraints; stage 4 with good affordances; stages 5-7 with feedback and visible state. Walk any interaction through each stage to find where users get stuck.
See: references/seven-stages.md for stage-by-stage analysis.
Observation → Idea Generation → Prototyping → Testing → (iterate)
Two specifics that change how you run this loop: in Observation, don't ask users what they want (they don't know) — watch for workarounds and frustrations in real contexts. In Testing, use real users not designers — 5 reveal ~85% of problems, so observe behavior over opinions and iterate.
| Mistake | Why It Fails | Fix |
|---|---|---|
| No signifiers | Users can't find features | Add visual cues for every interactive element |
| No feedback | Users don't know if action worked | Respond to every action within 0.1s |
| Blaming users | Ignores design flaws | Look for design cause of every "user error" |
| Feature creep | Complexity overwhelms | Apply constraints, progressive disclosure |
| Inconsistency | Breaks conceptual model | Same action = same result everywhere |
| Ignoring context | Designed for ideal conditions | Observe real usage environments |
Audit any design:
| Question | If No | Action |
|---|---|---|
| Can users figure out what to do? | Poor discoverability | Add signifiers, improve affordances |
| Do users understand what happened? | Gulf of evaluation too wide | Add feedback, show system state |
| Can users recover from errors? | No error tolerance | Add undo, confirmation, clear messages |
| Does the control layout match the output? | Poor mapping | Reorganize controls to match spatial layout |
| Are impossible/irrelevant options hidden? | Missing constraints | Disable, hide, or remove invalid options |
For the complete framework:
Don Norman, PhD is co-founder of the Nielsen Norman Group, director of The Design Lab at UC San Diego, and a former VP of Advanced Technology at Apple, where he coined the term "user experience." The Design of Everyday Things (1988, revised 2013) is widely considered the most influential design book ever written and is required reading in design programs worldwide.
gpt taste
Premium brand-kit image generation for high-end identity systems, logo concepts, and visual-world presentations.
Premium website design-to-code skill that generates visual references first, then builds faithful frontend implementations.
Premium frontend image direction for landing pages and marketing sites, generating one distinct horizontal image per section.
Premium mobile app screen concepts and flows with clean hierarchy, strong imagery, and distinctive art direction.
Emil Kowalski's philosophy on UI polish, component design, and the invisible details that make software feel good.
Frontend design and iteration for production-grade interfaces, from concept to ship.
Reverse-engineer design tokens from public websites into starter token files for your project.
Builds an animation from scratch by making the decisions in the order that determines whether it feels right.
Create beautiful visual art in .png and .pdf documents using design philosophy. You should use this skill when the user asks to create a poster, piece of art, design, or other static piece. Create original visual designs, never copying existing artists' work to avoid copyright violations.
Applies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply.
Toolkit for styling artifacts with a theme. These artifacts can be slides, docs, reportings, HTML landing pages, etc. There are 10 pre-set themes with colors/fonts that you can apply to any artifact that has been creating, or can generate a new theme on-the-fly.
Chat-to-edit AI studio for product photos, portraits, branding, and video.
Generate your dream room in seconds.
Redesign your interior in seconds using AI
Transform rooms with AI-generated designs.
Transform Your Space with AI-Powered Japanese Interior Designs by Jife
AI-powered wireframes for non-designers.
Revolutionize Property Viewing with AI-Generated Floor Plans and Virtual Tours
Transform sketches into stunning architectural renders with AI.