Skip to content

plugin87/ux-ui-agent-skills / .claude/skills/design-screen

design-screen

Verified

Design or build a page, screen, landing page, website, app or UI from a request - the whole job, end to end. Use this for "design a landing page", "build a settings screen", "make me a website", "design the onboarding", "ออกแบบหน้าเว็บ", "ทำหน้า landing", "ออกแบบ UI", "ทำแอป". It runs the pipeline no single skill covers: the house doctrine, a written Brief Inference so the result is not the statistical mean, a committed visual direction, tokens, build, the objective gates, then an adversarial critic. Hands off to data-dashboard for dense data screens, redesign for existing UI, and image-to-code when a reference image is given.

Skill: Design a screen

Do this first, before anything else

Write the Brief Inference block into your reply now. Not after reading the rest of this file, not in your head, not "once the layout is clearer" - now, before any markup, any file, any tool call.

Domain:           fintech | editorial | dev-tool | healthcare | consumer | ...
Audience & tone:  expert vs first-time, calm vs energetic, premium vs utilitarian
Mood:             the one adjective the result must earn - "expensive", "precise", "warm", "brutal"
Motion depth:     none | subtle feedback | expressive choreography
Layout family:    the section sequence you will use, named, varied per section
Reference anchor: an archetype, or a named system - see step 3

Fill in every line with a decision. Mood: clean is not a decision; Mood: expensive is. If you cannot name the mood and the layout family, you have not decided anything yet and whatever you generate will regress to the mean.

The stop condition: if you are about to write markup, create a file, or call a tool that produces UI, and this block is not already in your reply, stop and write it first. Measured on 2026-10-07, one run in three skipped it - the skill loaded, the block did not get written, and the result was a screen designed by nobody. This section is at the top of the file because it was previously below twenty-nine lines of background, and background is not an instruction.

Everything below is how to execute the decisions above.


Requirements

This skill needs the full UX/UI Agent Skills kit: the gates, the references and the design-system library. Install it with npx ux-ui-agent-skills@3.1.0 init or as the Claude Code plugin. npx skills@1.7.6 add copies the skill folder and nothing else. Installed that way, this skill runs in guidance-only mode: no gate runs, so it gives no verdict.

Step 0: is the kit here? Check once: ls ${CLAUDE_SKILL_DIR}/../../../tokens >/dev/null 2>&1 && echo KIT_OK || echo KIT_MISSING On KIT_MISSING, open your reply with: "Guidance-only mode: the UX/UI Agent Skills kit is not installed here, so no gate ran and nothing below is measured." Then work from this file and general practice, under three rules. Do not guess or cite the contents of a kit file you could not open. State no measured number (no contrast ratio, no N/N, no "passes WCAG"): write "not verified" where one would go. End with a line Skipped (kit not installed): that names every gate script this file tells you to run, then the install command above.

Hand off instead, when the work is actually

The requestUse
A dashboard, terminal, console, or any screen whose job is many numbers at oncedata-dashboard
An existing site or app to improveredesign
A screenshot, mockup or reference image to matchimage-to-code
One component's spec, not a screendesign-component
A brand foundation from scratch, before any screenbrandkit

Otherwise, continue.


1. Taste preflight

The block at the top of this file. If it is not in your reply yet, go back and write it before reading further - the rest of this skill executes decisions it assumes you have already made.

Full version and the reasoning: ${CLAUDE_SKILL_DIR}/../../../taste/design-taste.md.

Banned defaults: break each one deliberately

These are the tells that make UI read as generated. They are defects, not starting points.

  • Three or four identical equal-weight cards. One thing leads: more size, more weight, or its own row. A grid of equals gives the eye nowhere to land.
  • Everything centered. Use asymmetry and a real grid. Center sparingly.
  • The same left-text / right-image row repeated. Vary composition per section.
  • Display type that is just bold body text. The largest heading is at least 2.5x the body size, short, on a wide measure (18-24ch).
  • Generic drop shadow on every box. Most things are flat; elevation is layered and rare.
  • Pure #000 on #fff. Off-black on warm-white, from the theme.
  • A rainbow of accents. One primary, one accent at most; neutrals carry it.
  • Emoji anywhere - icons, bullets, status dots, labels. Use lucide inline SVG with currentColor, or plain words. This is a hard gate, not a preference.
  • Em-dashes in UI copy, marketing filler ("elevate", "seamless", "unlock"), hollow triads ("powerful, intuitive, beautiful"), fake labels ("SECTION 01", lorem ipsum). Say the concrete thing or say nothing.
  • A colored border on one side only of a card, alert, toast or callout. The tinted 3-4px strip is a generated-UI cliche. Full hairline border or surface separation, and status carried by a real icon plus text, never colour alone.
  • Cramped vertical rhythm. Generous macro-whitespace between sections.

2. Doctrine

Load design-doctrine and follow it. The parts that decide this job: token by intent, one shared theme, the eight states on every interactive element, one thing leads, zero emoji, and never state a number you did not measure.

3. Direction into tokens

Pick the reference anchor from ${CLAUDE_SKILL_DIR}/../../../taste/aesthetic-systems.md or a named system (node ${CLAUDE_SKILL_DIR}/../../../scripts/library.mjs get <name> fetches its spec and prints where it is), then resolve it into tokens rather than styling screens directly. One CSS-variable layer, imported once. If a brand colour fails contrast, adjust it: taste serves tier 4 and never overrides accessibility.

Reference output to match for quality: ${CLAUDE_SKILL_DIR}/../../../examples/showcase/ and ${CLAUDE_SKILL_DIR}/../../../examples/templates/.

4. Build

Mobile is the base, not an afterthought: write the phone layout as the base styles and layer wider screens on with min-width. max-width is fine for a narrow refinement, never as the way mobile gets supported. Every value a token. Every interactive element ships all eight states. Content first, then the layout that serves it.

5. Gate: run it, never claim it

node ${CLAUDE_SKILL_DIR}/../../../scripts/accuracy_report.mjs

Report the real N/N line. For a single file, at minimum:

node ${CLAUDE_SKILL_DIR}/../../../scripts/gate.mjs verify_states.mjs <file> [--dark]
node ${CLAUDE_SKILL_DIR}/../../../scripts/gate.mjs verify_responsive.mjs <file> --scale=1.25
DS_REQUIRE_BROWSER=1 node ${CLAUDE_SKILL_DIR}/../../../scripts/slop_tells.mjs --strict <file>
DS_REQUIRE_BROWSER=1 node ${CLAUDE_SKILL_DIR}/../../../scripts/taste_audit.mjs --strict <file>

A render gate with no browser prints SKIPPED and exits 0. That is not a pass. Set DS_REQUIRE_BROWSER=1 when you run one on its own and intend to trust it.

6. Render and look

Gates pass while a screen is still visibly broken. Screenshot it, look at every state, and click each control to confirm something actually changed.

7. Critic

Run /critique. It is adversarial on purpose and it sees the work, not your reasoning. A passing gate is never evidence of taste.


What you deliver

The Brief Inference block, the files, the real gate output, and the critic's verdict. If a gate was not run, say which and why - an unmeasured result is reported as unmeasured, never as a pass.

writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deploymentStars: 318kUpdated Claude CodeCodexVerifiedInstall
xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved.Stars: 196kUpdated Claude CodeCodexVerifiedInstall
shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy.Stars: 118kUpdated Claude CodeCodexVerifiedInstall
childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release.Stars: 117kUpdated Claude CodeCodexVerifiedInstall