A skill is a folder with a SKILL.md file that your AI agent loads automatically when a task matches it. Design skills carry taste: typography, spacing, motion, and hard bans on the defaults that make AI output look templated.
These twelve are the ones worth installing. Star counts, licences and update dates below are pulled from GitHub, not estimated.
| Your problem | Install | Why |
|---|---|---|
| My AI output looks generic | frontend-design | First-party, maintained, and aimed squarely at templated defaults. |
| I want one consistent system | impeccable | A full design language rather than a set of taste preferences. |
| Interactions feel stiff | apple-design | Treats motion as physics, so gestures stay interruptible. |
| I need to audit what exists | web-design-guidelines | Grades live UI and catches accessibility misses. |
Ordered by what to install first, not by star count. Stars and forks are repository totals, so skills sharing a repo share a number. All figures checked on 6 September 2026.
Anthropic anthropics/skills
Makes your agent commit to a real aesthetic direction instead of falling back on templated defaults.
Everything you generate looks the same
First-party, maintained, and aimed at the most common complaint about AI-built interfaces.
Paul Bakaus pbakaus/impeccable
A complete design language your agent applies to working code, rather than a loose set of preferences.
You want one system, not vibes
Built to be harness-agnostic, so the same rules follow you between tools.
Emil Kowalski emilkowalski/skills
Encodes the invisible details that separate software that works from software that feels finished.
It looks right but feels wrong
Taste from the person behind Sonner and animations.dev.
Leonxlnx Leonxlnx/taste-skill
A blunt filter that stops your AI producing boring, generic output.
You want the fastest possible fix
The quickest route out of the purple-gradient default.
Emil Kowalski emilkowalski/skills
Brings Apple's physical, gesture-driven approach to motion onto the web.
Interactions feel stiff or fake
The only skill here treating motion as physics rather than decoration.
Emil Kowalski emilkowalski/skills
Builds a single animation properly, making each decision in the order that decides whether it feels right.
You need to build one specific animation
Starts at whether the thing should move, not at which library to reach for.
Vercel vercel-labs/agent-skills
Grades interfaces you already have against the Web Interface Guidelines.
You need an audit, not new output
Catches accessibility misses that taste-driven skills sail straight past.
Emil Kowalski emilkowalski/skills
Surveys every animation in a codebase and hands back a prioritised plan of fixes.
You inherited messy motion
Gives you a roadmap instead of an unrequested rewrite.
Anthropic anthropics/skills
Applies one visual theme consistently across slides, docs, reports and pages.
Your deliverable is a set, not one screen
The fastest way to make a deck, a doc and a page look related.
Emil Kowalski emilkowalski/skills
Turns a vague description of a motion effect into the exact term for it.
You cannot name the effect you want
Naming an effect is what lets you ask for it precisely.
Anthropic anthropics/skills
Creates posters and static visual pieces as PNG or PDF, rather than interfaces.
The ask is a poster, not a screen
The one entry here that is not about interfaces at all.
h3nryprod01 h3nryprod01/design-taste
Merges three of the biggest design skills into a single install.
You would rather install one thing than three
A convenience bundle, and far smaller and younger than its sources.
Pick one primary taste skill
Choose a single skill to own your visual direction. Several overlapping taste skills will contradict each other and the output gets worse, not better.
Copy the folder in
Clone the repo and copy the folder containing SKILL.md into .claude/skills in your project, or the same path in your home directory to use it everywhere.
Restart your agent
Skills are read at startup. Restart Claude Code, Cursor, or whichever harness you use so the new skill registers.
Verify it triggers
Ask for a task it should match and confirm it loads. If it never fires, tune the description in its SKILL.md, because that is the part matched against your request.
You should not have to ask for taste twice.
A prompt applies to one message. Paste it, get your screen, and the next task starts from defaults again. A skill sits in the project and loads itself every time the work matches, so the standard holds on message one and on message fifty.
pasted once, per task
.claude/skills/name/SKILL.md
An open SKILL.md spec, portable across Claude Code, Cursor, Codex and more.
One taste skill for house style, plus an audit skill to grade what already exists.
Overlapping taste skills contradict each other. Install fewer than you think you need.
A skill is a folder holding a SKILL.md file, containing instructions an AI coding agent loads on demand when a task matches its description. A design skill carries taste: typography rules, spacing systems, motion principles, and explicit bans on the defaults that make AI output look templated. Unlike a prompt, which you paste once, a skill stays installed and fires automatically whenever relevant work comes up.
A prompt is a single message you copy and paste for one task. A skill is persistent, sitting in your project and loading itself whenever the agent recognises a matching task. Use prompts when you want a specific output right now, and skills when you want every future output to follow the same standards. Most people run both: a skill for the house style, prompts for individual screens.
No. SKILL.md is an open format, so these work across Claude Code, Cursor, Codex CLI, Gemini CLI and other harnesses that read the same structure. Anthropic publishes the spec in the anthropics/skills repository. A few skills call harness-specific scripts or sub-agents, and those parts degrade gracefully elsewhere, because the written guidance still applies even when the tooling does not.
Copy the skill's folder, the one containing SKILL.md, into .claude/skills in your project. Copy it into the same path in your home directory instead if you want it available everywhere. Restart your agent so it picks up the new skill. Some repositories also ship as installable plugins, so check the README of the specific repo, because install commands differ between them.
Start with Anthropic's frontend-design. It is first-party, actively maintained, and it fixes the most common problem, which is output that reads as a generic template. Add impeccable next for a fuller design system, then emil-design-eng once you care about how interactions feel rather than just how screens look. Installing all twelve at once is counterproductive, because overlapping skills give the agent contradictory instructions.
Yes. Skills load into context when their description matches a task, so several overlapping design skills can fire together and pull in different directions, with one demanding brutalist maximalism while another enforces restrained minimalism. Pick one primary taste skill, add specialists for distinct jobs like motion or accessibility auditing, and remove anything you are not actively using.
Every repository on this page is public and free to read. Licences differ, and each card shows the one its repository publishes, so check that before shipping a skill's contents inside a commercial product. A repository marked as having no licence means the author published no explicit terms, which means default copyright applies and you should ask before redistributing it.
Check the last update date, which every card on this page shows. Skills are instruction text rather than compiled code, so an older skill is not automatically broken, but a repository untouched for many months is worth reading carefully before you rely on it. Star counts tell you a skill was popular once, while the update date tells you somebody still cares.
Send the repo. If a link on this page has gone dead, tell us that too, because we would rather hear it from you than from Google.