The best Claude Code skills for front-end work

The best Claude Code skills for front-end work, chosen by what each covers: design, browser testing, React, shadcn/ui, Figma and feedback. With install steps.

Anthropic publishes a skill for Claude Code whose whole job is to stop the UI looking generated. Its description promises “choices that don't read as templated defaults.” That one line says a lot about where agents slip on front-end work, and why people go looking for the best Claude Code skills to install.

We picked seven for front-end work, each covering a point where an agent building UI tends to go wrong: design that looks generated, UI nobody checked in a browser, framework conventions, and what people said about the result. It's a list by coverage, not a ranking or a test report. Every description comes from the skill's own SKILL.md or README, opened on November 11, 2026.

What is a Claude Code skill?

A folder with a SKILL.md file in it. The file has a short description and then the instructions. Claude Code keeps the descriptions in context so it knows what's available, and according to Anthropic's documentation, “a skill's body loads only when it's used, so long reference material costs almost nothing until you need it.”

Where the folder sits decides who gets it. A personal skill lives in ~/.claude/skills/ and applies to all your projects on that machine. A project skill lives in .claude/skills/ inside the repository, and you commit it so your team gets it too. Plugins carry skills in a folder of their own.

The best Claude Code skills for front-end work, by what each covers

SkillPublisherWhat it covers
frontend-designAnthropicVisual direction, so new UI doesn't look templated
webapp-testingAnthropicChecking a local app in a real browser with Playwright
web-design-guidelinesVercelReviewing UI code against Vercel's Web Interface Guidelines
vercel-react-best-practicesVercelReact and Next.js performance rules
shadcnshadcn/uiYour project's shadcn/ui setup and component patterns
figma-design-to-codeFigmaBuilding a Figma design as code that matches it
The Skyelight plugin's skillSkyelightWorking through the feedback people left on the page

frontend-design, from Anthropic

The skill's description reads: “Guidance for distinctive, intentional visual design when building new UI or reshaping an existing one. Helps with aesthetic direction, typography, and making choices that don't read as templated defaults.” It tells Claude to work like a design lead whose client has already rejected proposals that felt templated.

It ships in Anthropic's skills repository. In Claude Code, run /plugin marketplace add anthropics/skills, then /plugin install example-skills@anthropic-agent-skills. That plugin bundles twelve skills, this one and the next included. Anthropic's official plugin marketplace also carries a plugin named frontend-design if you'd rather not take the other eleven. One caveat from the repository's README: the skills there are “provided for demonstration and educational purposes only.”

webapp-testing, from Anthropic

“Toolkit for interacting with and testing local web applications using Playwright,” says its SKILL.md. It supports “verifying frontend functionality, debugging UI behavior, capturing browser screenshots, and viewing browser logs.” In practice the skill has Claude write Python Playwright scripts, with a helper script that starts your dev server first. This is the one for the agent that says the page is done without having opened it. It installs with the same example-skills plugin.

web-design-guidelines, from Vercel

This skill reviews UI code for compliance with Vercel's Web Interface Guidelines. Its description says to use it when asked to “review my UI” or “check accessibility”. It fetches the current guidelines before each review and reports findings by file and line number, so the rules don't go stale inside the skill. Install it from Vercel's collection with npx skills add vercel-labs/agent-skills --skill web-design-guidelines -a claude-code.

vercel-react-best-practices, from Vercel

“React and Next.js performance optimization guidelines from Vercel Engineering.” The SKILL.md says it holds 70 rules across 8 categories, ordered by impact, with request waterfalls and bundle size at the top. The folder is called react-best-practices; the skill's own name carries the vercel- prefix. It comes from the same repository: npx skills add vercel-labs/agent-skills lets you choose it.

shadcn, from shadcn/ui

If your components come from shadcn/ui, this is the conventions skill for that library. Per shadcn's docs, it reads your project's components.json and gives the assistant “your framework, aliases, installed components, icon library, and base library”. It also points Claude at the shadcn CLI to look up a component's documentation before writing code. Install with npx skills add shadcn/ui.

figma-design-to-code, from Figma

Figma's skill is for when you ask Claude to “implement, build, port, or code up a Figma design as code.” It has the agent request a screenshot of the design and treat that render as the target the built page has to match. It needs Figma's MCP server, and both arrive together: Figma's README says its plugin “includes MCP server settings as well as Agent Skills for common workflows.” The command is claude plugin install figma@claude-plugins-official. For what designers get from this and the agents around it, see AI agents for designers.

The Skyelight plugin's skill, from us

Every skill above teaches the agent something it can check for itself. None of them can tell it what a person thought. On November 6, 2026, Claude Code rebuilt our homepage hero and checked it in screenshots at three sizes. It looked right. On the live site the animated dots behind the headline were huge and soft, and the agent's check had missed it because its headless browser couldn't draw WebGPU. A person saw it in seconds. (The longer version is in Claude in Chrome and website feedback.)

Skyelight is a website feedback tool, and its plugin for Claude Code is how that kind of comment reaches the agent. Per our setup docs, the plugin connects Claude Code to Skyelight and “adds a skill for working through feedback, plus two commands.” /skyelight:whats-new catches you up on feedback since you last looked. /skyelight:fix-feedback picks up an open item, fixes it and replies on the thread, and leaves resolving to you. To install it, type /plugin, open Discover and search for Skyelight.

It only earns its place if people are reviewing your pages in Skyelight. With no threads, the skill has nothing to work through. You also need an owner, admin or collaborator role in a workspace, and on the Free plan the agent gets 25 agent threads a month.

The one input a skill can't write for itself

Reviewers press C, click the element they mean and write the comment. With the Skyelight plugin installed, Claude Code reads that thread, makes the fix and replies where the reviewer left it.

How do you install a Claude Code skill?

There are three routes, and the list above uses all of them.

  • Copy the folder. Put the skill's folder in ~/.claude/skills/ or your repository's .claude/skills/. Nothing else is required.
  • Install a plugin. Anthropic's plugin documentation says typing /plugin opens a panel on its Discover tab, where you search, open a plugin and choose a scope: for you everywhere, for everyone on the repository, or for you in this repository only. Anthropic's official marketplace is added the first time you start an interactive terminal session. Any other marketplace needs /plugin marketplace add first.
  • Use the skills CLI. npx skills add comes from Vercel Labs. Its README says it needs Node.js 22.20 or later, installs to the current project by default, and takes -g to make a skill available across projects.

Read before you install. A skill is text an agent will follow, and the same plugin documentation warns that “a plugin can run hooks and MCP servers, so read the pane before you install.”

How can you tell if a skill is loading?

Ask. The troubleshooting steps in Anthropic's skills documentation include checking that the skill appears when you type What skills are available? You can also type / and look for it in the menu. Skills from a plugin show under the plugin's name, which is why Skyelight's commands start with /skyelight:.

Two things can hide a skill that is installed. An author can set user-invocable: false, which keeps it out of the menu so only Claude can call it. The shadcn skill does this, so asking is the way to check that one. And if you install a great many, descriptions start to disappear: the listing of names and descriptions gets a budget of 1% of the model's context window, and when it overflows Claude Code drops descriptions starting with the skills you invoke least. Without its description, a skill has nothing for Claude to match your request against. /doctor shows an estimate of what the listing costs.

If a skill is listed and still doesn't fire, the documented fix is to rephrase the request closer to its description, or to call it by name.

Is a skill the same as an MCP server?

No. A skill is instructions. An MCP server is a connection that gives the agent tools and data it didn't have: a Figma file, a browser, a backlog, a feedback thread. Anthropic's documentation lists them as separate parts of a plugin, which “adds its skills, agents, hooks, and MCP servers.”

The two often travel together. Figma's plugin and ours each install a server for access and a skill for how to use it. If you're choosing servers, we compared the main ones in which MCP server does what for front-end work, and how to send website feedback to a coding agent covers the Skyelight side in full.

Frequently asked questions

What is a Claude Code skill?

A folder with a SKILL.md file of instructions that Claude Code loads when the task calls for it. Anthropic's documentation says Claude uses skills when relevant, and that you can also run one directly by typing a slash and its name.

Where do Claude Code skills live?

In ~/.claude/skills/ for personal skills, which apply to all your projects on that machine, and in .claude/skills/ inside a repository for project skills, which you commit so your team gets them. Plugins carry skills in their own folder.

Is a skill the same as an MCP server?

No. A skill is written instructions, and an MCP server is a connection that gives the agent tools and data. A plugin can install both at once, which is how the Figma and Skyelight plugins work.

Do skills use up Claude's context?

A little. Claude Code keeps a listing of skill names and descriptions in context and loads a skill's full text only when it's used. Anthropic's documentation says that listing gets a budget of 1% of the model's context window, and descriptions are dropped when it overflows.

Seven is already a lot. Install the ones that match the mistakes you keep correcting, and leave the rest on GitHub.

Give your agent the feedback people left on the page

No card required.