There are quite a few skills designed for Claude being shared, but they are scattered all over the place. This document gathers the eight most worthwhile ones in one place, detailing what each one is, how to install it, which prompts to use, and when to use it.
Even if you have never installed a skill before, you can easily follow along: The first section explains what a “skill” is and shows the easiest way to install one (drag-and-drop a single file). If you are already familiar, feel free to jump straight to the section you need using the table of contents below.
1. What is a Skill? Two Ways to Install for Beginners
Simply put, a Skill is an instruction package we install for Claude so it excels at a specific task. It’s just like installing an app on your phone: the device remains the same, but it gains new capabilities. Technically speaking, each skill is simply a text file named SKILL.md containing the rules Claude must follow.
The beauty of it is that Claude automatically knows when to use it. You don’t have to call the skill by name. Once installed, simply prompt, “build me a landing page to sell an online course,” and it will automatically pull up and apply the relevant design skill.
There are two places to install it, and these two places are different. Make sure you choose the right one, otherwise you won’t see any results after installing:
| Where to use | How to install | Best suited for |
|---|---|---|
| Claude on web / app (claude.ai) | Drag and drop the SKILL.md file into the chat window |
Non-technical users, easiest method |
| Claude Code (runs in the command line) | Type a single command: npx skills add ... |
Developers working on local codebases |
Three types of tools in this collection
The 8 items in this article are not all the same type. Knowing the right type will help you understand what we’re doing during installation and know where to look when things break.
| Type | What it is | In this article |
|---|---|---|
| Skill | A directory containing a SKILL.md file, which includes instructions to make Claude proficient in a new task.[11] |
Taste Skill, Emil Kowalski, Web Design Guidelines, Image to Code, Awesome Design HTML |
| Skill with bundled program | In addition to the instruction file, it also installs a command-line program. Only usable within Claude Code. | Impeccable, Playwright CLI |
| Reference file | Just a text file DESIGN.md that you feed to Claude. No installation required. |
Awesome Claude Design |
Why distinguish them: Once installed, Skills require restarting Claude Code to appear, verified using the /skills command. Command-line programs live outside Claude—you can run them directly in the Terminal. Reference files sit quietly in the project, and Claude only reads them when you mention them. When something breaks, just troubleshoot along that specific path.
Method 1: Drag-and-drop the file into chat (Easiest)
- Click the View on GitHub button for the skill you need, then open the
SKILL.mdfile. In the top right corner, click the Download raw file button (downward arrow icon) to download it to your machine. - Open Claude and drag-and-drop the downloaded file into the chat box (where you type messages).
- Send a prompt: “Please install this skill for me”. Claude will automatically add it.
- Done. From now on, simply tell it what you need done, and it will automatically use that skill.
If you prefer manual setup: Open Claude > go to Settings > Capabilities (or Skills) > add the downloaded file. If you don’t want to download the file, open SKILL.md on GitHub, click the Copy raw file button, and paste the content directly into the chat.
Method 2: One-line command for Claude Code
- Install Node.js first if you don’t have it yet; download it from nodejs.org and click Next all the way through.
- Open your command-line interface (Windows: Type cmd in the search bar; Mac: Open Terminal).
- Navigate to your project directory, then paste one line from the sections below and press Enter.
- Restart Claude Code. Type
/skillsto check if the skill appears in the list.
If you want the skill available across all projects instead of just this one, add --global to the end of the command.
General Syntax of the npx skills Command
Most skills in this article are installed using a tool called skills by the Vercel Labs team.[10] There’s no need to install it beforehand; the npx prefix automatically downloads and runs it. The first line in the box below checks if Node.js is already installed on your machine: If a version like v22.x.x appears, you’re all set; if it says command not found, download the LTS version from nodejs.org.[14]
By default, skills are copied into .claude/skills/ within the currently open project. Only that project can use them, but pushing them to Git makes them available to the entire team. Adding -g (or --global) installs the skills into ~/.claude/skills/, making them available to every project on your machine.[10]
In the /skills list, each skill appears under its display name. This name often differs from the repository name used during installation, so don’t assume the installation failed if you notice a discrepancy.
Ask Claude to install it for you: If you’d rather not type out commands, copy the command line for what you need, paste it directly into the Claude Code chat box, and message: “Run this command for me, ask if any permissions are needed.” Claude will ask for permission before running, and all you need to do is confirm.
Tip: Don’t install all 8 at once. Having too many skills related to design can easily cause “noise” for Claude, leading to worse results. Start with Taste Skill + Playwright CLI. These two complement each other without overlap: one takes care of aesthetics, while the other handles testing. Only add more when you feel something is missing.
2. Taste Skill: Injecting Good Taste into Claude
This is what we consider the most worthwhile part of the entire toolkit, because it doesn’t fix the code—it fixes Claude’s aesthetic mindset. Still the same AI, still the same prompt, but the returned results are completely different.
Specifically, Taste Skill is a very long list of “don’t do this, do that” rules, compiled from scrutinizing hundreds of AI-generated websites and documenting every recurring mistake. It directly eliminates the telltale signs of an “AI-made website”: purple-to-blue gradients, centered hero sections on dark backgrounds, identical rows of cards repeated section after section, and massive headlines paired with tiny subtext. In return, it pushes Claude toward asymmetrical layouts, clear typographic hierarchy, generous whitespace, and purposeful motion.
Three “Dials” to Tweak
The best part of Taste Skill is the three numbers at the top of the file, which you can adjust manually to get exactly what you want, on a scale from 1 to 10:
| Dial | Low Value Means | High Value Means |
|---|---|---|
DESIGN_VARIANCELayout unconventionality |
Centered, clean, safe | Asymmetric, off-grid, modern |
MOTION_INTENSITYMotion strength |
Color changes on hover only | Scroll-driven effects, magnetic buttons tracking the cursor |
VISUAL_DENSITYInformation amount per screen |
Spacious, generous whitespace | Dense, dashboard-style |
When building sales landing pages, we usually set VARIANCE 7, MOTION 5, DENSITY 3. For internal admin dashboards, it’s the opposite: VARIANCE 3, MOTION 2, DENSITY 8.
Installing Taste Skill
- The easy way: Open the repo below, go to
skills/taste-skill/SKILL.md, click Download raw file, then drag and drop that file into Claude chat and say “Install this skill for me.” - Via CLI (entire repo, including 13 variants): Paste
npx skills add https://github.com/Leonxlnx/taste-skill - Get only the standard version: Paste
npx skills add https://github.com/Leonxlnx/taste-skill --skill "design-taste-frontend" - To use it across all projects, add
--globalto the end.
Taste Skill (design-taste-frontend v2): Located in the Leonxlnx/taste-skill repo. This is v2, an almost complete rewrite of v1: Adds auto-inference of design intent from prompts, a design system map, GSAP boilerplate code, and a pre-output checklist. Open repo on GitHub · MIT · Claude · Cursor · Codex
Choosing the Right Flavor Among the 13 Skills
This repo contains 13 skills, each with its own personality.[1] Here are the ones people use the most:
| Skill Name | Best Suited For |
|---|---|
taste-skill (core version) |
Includes a real design system and a self-review step before exporting code |
soft-skill |
Pages needing a soft, elegant feel with plenty of breathing room |
minimalist-skill |
Minimalist style like Notion/Linear, with minimal clutter |
brutalist-skill |
Bold, high contrast, large typography blocks |
redesign-skill |
Revamping an existing page—reviews first, then makes changes |
brandkit, stitch-skill, output-skill |
Support tasks: Building brand identity kits, stitching screens together, standardizing output |
Installing the entire suite won’t confuse Claude. Each skill has its own description line; Claude reads the description and automatically selects the one that matches what you’re asking for. If you want to be completely sure, specify the style directly in your prompt, for example, “go with a minimalist approach.”
How to Use
The skill activates automatically whenever you ask for UI-related tasks. Just describe what you need without calling the skill by name:
3. Emil Kowalski: Smooth and “Eye-Pleasing” Motion
Taste Skill handles static beauty. This one takes care of beauty in motion. Emil Kowalski is a former engineer at Vercel and Linear, as well as the creator of the Sonner notification library. This skill set is years of his experience packaged together.[5]
The problem it solves is very specific: AI often picks the wrong ingredients for an effect. Choosing an ease-in curve for an opening menu (when it should actually be ease-out), or making a button expand from scale(0)—that is, out of thin air, when nothing in real life appears that way. Each small mistake might sound insignificant, but combined, they make the difference between an interface with “soul” and one that is… bland.
A few specific rules it taught Claude
- Never use
ease-infor UI. It starts slowly right when the human eye is looking the closest, making the website feel laggy even if the duration is identical. - UI transitions must be under 300ms. A 180ms menu feels noticeably faster than a 400ms menu.
- Buttons must shrink when pressed:
transform: scale(0.97). It makes users feel like the device actually “heard” them. - Keyboard-driven actions must have zero transitions. Things you press hundreds of times a day become sluggish if they have animations. Raycast doesn’t have open/close transitions, and that is the right call.
- Only animate
transformandopacity. These two run on the GPU. Touchingheight,width, ormarginforces the browser to recalculate the layout, causing stutter. - Use custom curves, not defaults. Default CSS easings are too soft and lack punch. For example, use
cubic-bezier(0.23, 1, 0.32, 1).
The Repo Has 15 Skills, Not Just One
When installing via the command line, you get all 15. Here are the 9 most notable ones:
| Skill | When to Use |
|---|---|
emil-design-eng |
Core skill: General philosophy, primarily focused on motion, along with design advice. Install this first. |
animate |
Need to build a new animation from scratch; it automatically chooses the right curves, duration, and properties. |
review-animations |
Already have animations and want it to inspect where things feel off. |
improve-animations |
Scans all animations across the project and returns a prioritized improvement plan. |
find-animation-opportunities |
Finds where motion should be added. More importantly, points out where it shouldn’t. |
apple-design |
Apple design and motion principles, distilled from WWDC sessions and translated for the web. |
animation-vocabulary |
Teaches you how to name the exact effect you want, so the AI gets it right on the first try. |
pick-ui-library |
Picks the right library for the job, instead of letting AI hand-code a toast component or install an abandoned package. |
prototype |
Builds multiple variations of the same UI section, allowing you to click back and forth to compare. |
Installing Emil Kowalski’s Skill Set
- The easy way: Open the repo in the box below, go to
skills/emil-design-eng/SKILL.md, click Download raw file, then drag that file into the Claude chat and say “Help me install this skill.” - The CLI way (get all 15 skills): Paste
npx skills@latest add emilkowalski/skills - After installing, try prompting: “Review all the animations in this project for me.” It will return a Before / After / Why table for each issue.
Emil Kowalski: Design Engineering: The core emil-design-eng skill in the emilkowalski/skills repo. Inside, you’ll find a decision matrix on “whether to animate,” duration tables for each component type, recommended easing curves, tips for clip-path and blur, drag-and-drop interaction rules, and a final checklist. Open the repo on GitHub · Motion and Details
4. Impeccable: A single command to clean up your layout
Instead of fumbling around and second-guessing which font to use or what spacing looks right, with Impeccable you just need to throw in a single command. It automatically cleans up and realigns your entire layout.
Impeccable originated from Anthropic’s own frontend-design skill, then expanded into 24 commands and 59 automated lint rules. These checks run without making AI calls, so it’s both fast and saves tokens.[6]
Most Frequently Used Commands
| Command | What it does |
|---|---|
/impeccable init |
Must be run first. It asks about your product (who uses it, what it’s used for, what the brand tone is) and writes it to the PRODUCT.md file. Subsequent commands read this file so they don’t give generic answers. |
/impeccable layout |
Fixes layout, spacing, and visual rhythm. This is the “cleanup” command you will need the most. |
/impeccable typeset |
Fixes fonts, font sizes, and typographic hierarchy. |
/impeccable critique |
Reviews the design like a designer: Hierarchy, clarity, and user emotion. |
/impeccable polish |
Final pass before deploying the site. Re-synchronizes the design system and checks for any missed details. |
/impeccable bolder · /impeccable quieter |
An opposing pair. If the site is too bland, use bolder; if it’s too noisy, use quieter. |
/impeccable distill |
Strips everything away, keeping only the core essentials. Use when the page is overcrowded with too many things. |
/impeccable harden |
Handles edge cases: Errors, text overflow, multi-language support, and empty states. |
Append a section name to keep it focused: /impeccable audit blog, /impeccable polish settings, or speak directly in plain language: /impeccable redo this hero section. Pin frequently used commands with /impeccable pin audit, so from then on you only need to type /audit.
List of Things It Prohibits
This part is worth reading, as it serves as a checklist of “AI-generated website hallmarks” that you should avoid even if you don’t install the skill:
- Overused fonts: Arial, Inter, default system fonts
- Gray text on a colored background
- Pure black or pure gray—always mix in a slight tint of color
- Wrapping everything in cards, then nesting cards inside cards
- Bouncy/elastic easing effects that look dated
- Purple-to-blue gradients, glowing light effects on dark backgrounds
Installing Impeccable (Claude Code Only)
- Open your terminal and navigate to your project root directory.
- Paste:
npx impeccable installand hit Enter. It automatically detects whether you are using Claude Code or Cursor and installs it in the right place. - Reopen Claude Code, run just once:
/impeccable init, and answer a few questions about your product. - From then on, simply type
/impeccableto see the full list of commands.
An alternative method for Claude Code only: /plugin marketplace add pbakaus/impeccable, then open /plugin and select Impeccable. For future updates: npx impeccable update.
Warning: Impeccable is not a drag-and-drop SKILL.md file like other skills. It comes with a small binary and a “hook”—a trigger that runs automatically every time Claude edits a UI file. Therefore: (1) It only works with Claude Code / Cursor / Codex, it cannot be used on claude.ai; (2) This hook runs independently of whether you approve individual commands, so only install it in projects you trust. If you want to disable all hooks for a single run, add --no-hooks during installation. The live mode also only works locally on your machine—do not attempt to run it on a production website.
Impeccable: 24 design commands: Impeccable is not distributed as a standalone SKILL.md file. The installer automatically scaffolds the skill, hook, and linter right inside your project. All 24 commands are listed in the table above, with full documentation at impeccable.style. View repo on GitHub · Apache 2.0 · Install via CLI
5. Web Design Guidelines: The Tough Judge
The four skills above take care of the creation. This one handles the grading. It acts like a tough judge, scrutinizing every line of code based on Vercel’s design guidelines: over 100 rules on accessibility, performance, and user experience.[2]
The smartest point: the skill file is very short because it contains no rules at all. Every time it runs, it fetches the latest rule set from Vercel’s repo before evaluating. This means the guidelines are always up to date—you install it once and use it forever without worrying about it becoming obsolete.
What it inspects
- Accessibility: Missing
aria-labeltags, using semantically incorrect HTML tags, buttons inaccessible via keyboard - Focus states: Is the focus ring visible, is
focus-visibleused correctly - Forms: Presence of
autocomplete, how errors are reported, how data is validated - Motion: Does it respect
prefers-reduced-motion(essential for users with motion sickness), does it use GPU-friendly properties - Typography: Curly quotes, proper ellipsis characters, tabular figures for aligned numbers
- Images: Dimensions declared (omitting them causes layout shifts during loading), lazy loading, alt text
- Dark mode, touch, internationalization and a few more categories
Installing Web Design Guidelines
- The easy way: Open the repository in the box below, go to
skills/web-design-guidelines/SKILL.md, click Download raw file, then drag that file into the Claude chat and message: “Help us install this skill”. - The CLI way (fetches Vercel’s entire skill set): Paste
npx skills add vercel-labs/agent-skills - Trigger it by directly messaging prompts like: “Review our interface”, “Check accessibility”, “Audit this page’s design”.
This skill requires internet access to fetch the latest guidelines. If Claude reports it cannot access the web, make sure to enable web search permissions for it.
How to Use
If you only want the linting skill without installing Vercel’s entire suite, install it with this command:
When asking it to inspect, clearly specify the file or directory you want to review, otherwise it will ask you what you want to check.[2] The ruleset it downloads is located in the web-interface-guidelines repository.[12]
Enforce audit first, fix later: Do not ask it to audit and fix in the same prompt. Have it list the errors first, review them yourself, and then decide which ones are worth fixing. Many rules in the guideline are matters of opinion, not actual bugs in your project.
Vercel Labs Web Design Guidelines: Located in vercel-labs/agent-skills. The file is very short because it only serves as an “anchor”; the actual rules reside in the web-interface-guidelines repo and are fetched fresh on each run. Open repo on GitHub · MIT · Official Vercel
6. Awesome Design: Design frameworks from major brands
Simply put: It loads entire design frameworks from major tech giants into Claude. Color palettes, typography, border radiuses, spacings—everything is pre-standardized, so all you have to do is name the brand.
There are two versions, sharing the same concept but implemented differently, and in this set of 8 skills, they are counted as two. You should understand the differences to choose the right one:
| Awesome Design HTML | Awesome Claude Design | |
|---|---|---|
| File format | Over 110 real, runnable HTML pages | 68 text-based DESIGN.md files |
| Used with | Claude Code (installed as a skill) | Claude Design on claude.ai, or placed inside a Claude Code project |
| Pros | Claude opens the file and sees actual pixels directly, no imagination needed | Lightweight, generates a complete design system along with a UI kit |
| Includes | About 90 web pages and 22 iOS app templates | 68 brands, categorized by industry |
The HTML version is based on an argument we found quite convincing: A text line saying “rounded blue button” forces the reader to imagine it, whereas an HTML file is the result itself. Claude reads that file and sees the exact same pixels that you see. A 300-line HTML file conveys color, font size, motion, and interaction in 30 seconds. Describing that in text would take hours while still losing precision.[3]
How to Call It: Just Mention the Brand Name
No special commands needed. Once installed, simply prompt naturally—just mention the brand name, and it will automatically pull up the reference file:
Available out of the box: Stripe, Linear, Notion, Apple, Vercel, Figma, Airbnb, Cursor, Claude, Mistral, Spotify, Tesla, Ferrari, BMW, Nike, Shopify, Supabase, Sentry… along with 22 iOS app templates.
Easiest Way to Install Awesome Design HTML
- Open Claude Code, paste this exact prompt into the chat:
Please install this Claude Code skill for me: https://github.com/yzfly/awesome-design-html - Claude will automatically download the entire repo into its skills directory. Done.
- If you prefer doing it manually, run:
git clone https://github.com/yzfly/awesome-design-html.git ~/.claude/skills/awesome-design-html - Don’t want to install anything: Head over to the online library, open any page you like, and press Ctrl+S to save it. Each page is a standalone HTML file ready to be used as a template right away.
Using Claude Design on the web is slightly different: Visit awesome-claude-design, download the DESIGN.md file for your preferred brand, attach it to the chat input at claude.ai/design, and message: “Create a design system from this DESIGN.md”.
Using DESIGN.md Directly in Claude Code
Even without using Claude Design on the web, you can still take advantage of this file format. The VoltAgent/awesome-design-md repository contains 73 DESIGN.md files extracted from major brand websites, each accompanied by two preview.html pages (light and dark) for previewing.[13]
- Open the repo, navigate to the
design-md/directory, select your preferred brand, openDESIGN.md, and copy its content. If you want the entire repo, click the green Code button and select Download ZIP. - Copy the file into your project’s root directory, keeping the filename as
DESIGN.md. Placing it right at the root makes it easy for Claude to find and ensures the whole team shares a single version. - Explicitly mention the filename in your prompt; don’t let Claude guess. A sample prompt is provided in the box below.
Warning: This skill requires the entire directory of over 110 HTML files to function properly. The SKILL.md file alone is merely a guide that maps out which brand corresponds to which file. If you only download that single file, Claude will know the list but won’t have the reference files to inspect. To actually use it, install it using one of the three methods above. Additionally, these files serve as learning references, not official design systems of the respective companies. Color palettes, logos, and proprietary fonts remain their intellectual property. Use them as inspiration to craft your own system; do not copy them 1:1 for commercial use.
Awesome Design HTML: Repo yzfly/awesome-design-html. The guide file includes a complete list of brands alongside file codes, a “design archetypes” table to look up suitable brands when you can’t recall specific names, and quality standards for building new pages. Open repo on GitHub · MIT · Over 110 reference pages
7. Image to Code: Drop an image in, get code out
This is super convenient: You drop in a screenshot or a design file, and Claude translates it directly into code, capturing every single detail.
What few people know, however, is that this skill also works in reverse—and that’s the best part. It forces Claude to generate the design image first, and only then write code based on that image. The mandatory sequence is: Generate image -> Deeply analyze image -> Write code. The reason is simple: If you let the AI code freely from the start, it falls back on generic, familiar patterns. But when it has to draw it first and then replicate it with code, it has a concrete “blueprint” to stay faithful to.
Notable rules in this skill
- One section, one image. If you request 8 sections, it generates 8 separate images instead of cramming them into a single long image with tiny, unreadable text.
- No cropping old images. If you need a close-up of a specific section, generate a new image—cropping ruins all spacing and proportions.
- Don’t be “lazy” with the image count. More clear images are better than fewer compressed, tiny ones.
- Hero headlines: maximum 3 lines. If it’s longer, trim the copy; do not add extra line breaks.
- Preventing “drift” during coding. The most common issue is an image that looks great, but turns into a generic page once coded. The skill has an entire dedicated rule prohibiting this.
- Banned buzzwords: “unleash”, “elevate”, “revolutionize”, “next-gen”, “seamless”, along with fake brand names like Acme, Nexus, and NovaCore.
Image generation tool required: The image generation step only works if your current software supports generating images. Standard Claude Code cannot generate images natively yet, so either attach an image-generating MCP or create images elsewhere and drop them into the chat.[1] Generated images are only meant for locking down layout and aesthetic vibes—do not crop and paste them into the page. Text in AI images often has Vietnamese spelling errors, and images are much heavier than code.
Tip: This skill only follows the proper workflow when you explicitly state the order in the prompt. Append this phrase: “follow the skill: generate images, then analyze, then code”. Without this reminder, it tends to jump straight to coding. If you already have design mockups or screenshots, simply drop the image into chat and send: “Analyze this image thoroughly and code it identically: Colors, fonts, spacing, border-radius, button styles”. Skip the image generation step.
Installing Image to Code
- The easy way: Open the repo in the box below, navigate to
skills/image-to-code-skill/SKILL.md, click Download raw file, then drag that file into the Claude chat and prompt: “Please install this skill for me.” - The CLI way: Paste
npx skills add https://github.com/Leonxlnx/taste-skill --skill "image-to-code" - It is located in the same repository as Taste Skill, so if you already installed the entire repo in section 2, you already have it—no need to reinstall.
Image to Code: Located at skills/image-to-code-skill/SKILL.md. Contains 38 rules, the most notable of which are Rule 29 (the “AI smell” checklist to avoid), Rule 14 (hero section rules), and Rule 27 (preventing drift during image-to-code conversion). Open repo on GitHub · MIT · Same repo as Taste Skill
8. Playwright CLI: Giving Claude a Pair of Eyes
This is the ultimate game changer, and it’s straight from Microsoft. The seven skills above only teach Claude to know what looks good. This one lets it actually see the results of what we’ve built.
Specifically: Claude opens the browser on its own, navigates to the page it just generated, takes screenshots or inspects the page structure, and checks for broken layouts, text overflows, or misaligned buttons—then fixes them automatically before you even notice. No more painful loops of “you open the preview → spot a bug → take a screenshot → describe it to Claude → it fixes it → you preview it again.”
Commands You Should Know
| Command | What it does |
|---|---|
playwright-cli open |
Opens a new browser. Runs headless by default; add --headed to make the window visible so you can watch along. |
playwright-cli goto <url> |
Navigates to a page |
playwright-cli snapshot |
Reads the page structure as text. This is the most frequently used command. It is lighter than an image and more accurate when locating elements. |
playwright-cli screenshot |
Takes a screenshot. Add --filename=home-page.png to name the file. |
playwright-cli resize 390 844 |
Resizes the window to test mobile responsive layouts |
playwright-cli open --mobile |
Emulates a full mobile device |
playwright-cli console |
Reads error logs in the browser console |
playwright-cli show --annotate |
An often-overlooked gem. It opens the page so you can highlight areas and make notes directly, then sends it back to Claude along with the image. Give design feedback by drawing, not by describing. |
playwright-cli close |
Closes the browser |
Installing Playwright CLI Specifically for Claude Code
- Install Node.js 18 or higher first if you don’t have it yet, download at nodejs.org.
- Open your terminal, paste:
npm install -g @playwright/cli@latestand press Enter. Wait for it to finish downloading. - Verify the installation with:
playwright-cli --help. If you see the list of commands, you’re good to go. - Navigate to your project directory and run:
playwright-cli install --skills. The skill file will be placed in.claude/skills/playwright-cli. - If you want to use it across all projects, add
-g:playwright-cli install --skills -g
Alternative method: npx -y skills add microsoft/playwright-cli --skill playwright-cli --agent claude-code
How to use
You rarely have to type these commands manually—Claude will handle that. Your job is simply to clearly state what you need to check:
How it differs from Playwright MCP: Both come from Microsoft and both control browsers. The CLI version is lightweight and fast, ideal for taking screenshots and quick checks without stuffing the entire web page into Claude’s context window. The MCP version maintains sessions and inspects deeper, making it better suited for long-term tasks like running test suites.
Note: This skill requires the playwright-cli program to be installed on your machine to work. Simply dragging the SKILL.md file into claude.ai will not work, as Claude on the web cannot execute commands on your local machine. You must follow the installation box above. Installing the skill isn’t strictly required either: you can simply prompt, “Use playwright-cli to inspect the UI of this page, run playwright-cli –help first to see available commands.” Having the skill just makes the execution much more accurate.
Microsoft’s Playwright CLI: Repo microsoft/playwright-cli. The skill file is a comprehensive command cheat sheet: navigation, interactions, keyboard, mouse, tabs, cookies, network throttling, log reading, screenshots, video recording, and how to target elements using the e15 ref ID extracted from snapshots. View repo on GitHub · Official Microsoft project · Requires local software installation
9. Putting it all together: Building a website in 6 steps
If you’ve installed all 8 skills but don’t know which order to use them in, you’ll still end up with messy results. Here is the exact workflow we are using. Just follow it step-by-step, with each step being a single prompt sent to Claude.
Step 1: Declare the context once
Run /impeccable init. It asks what your product is, who uses it, what the tone of voice is, and then writes it to PRODUCT.md. This step takes 3 minutes but saves a lot of time, because every subsequent command reads this file, so it no longer gives generic answers.
Step 2: Finalize the design direction before coding
Prompt something like: “Create an English course landing page, Linear-style but warmer. Show us the design direction before writing code.” Awesome Design handles brand references, while Taste Skill takes care of layout and visual rhythm. Do not let it write code at this step.
Step 3: Generate design images, then start coding
Message: “follow the skill: generate images, then analyze, then code”. Image to Code will generate images for each section, analyze them, and then build the code (requires an image generation tool, see section 7). If you already have design images ready, simply drag them in and skip the generation step.
Step 4: Clean up layout and typography
Run /impeccable layout and then /impeccable typeset. These two commands fix spacing, visual rhythm, fonts, and font sizes—the things most prone to misalignment when coding quickly.
Step 5: Add motion, then audit it
Prompt: “Add animations to the necessary parts, then audit everything according to Emil’s rules.” Emil Kowalski’s skill builds and evaluates at the same time. It will return a Before / After / Why table: Where it’s ease-in, change it to ease-out; where it’s scale(0), change it to scale(0.95).
Step 6: Self-Inspect, Self-Fix, Then Deliver
Prompt: “Use playwright-cli to open this page, take screenshots on both desktop and mobile, detect visual bugs, and fix them automatically. Then inspect the UI against Vercel standards.” Playwright inspects, Web Design Guidelines scores. This is the step most people skip, yet it is the most effective one.
Tip: Save the snippet below into a file named CLAUDE.md in your project root. Claude will automatically read it every time it starts and follow the workflow without needing any reminders.
10. Frequently Asked Questions & Troubleshooting (FAQ)
A compilation of answers to common questions and troubleshooting tips when applying the 8 Agent Skills to Claude Code:
Click on the error you’re encountering to see how to fix it.
Two things to be cautious about: Skills are text files written by strangers, so before installing, open SKILL.md and skim through it to see what instructions it gives Claude. Playwright can open any page, so don’t let it navigate unfamiliar sites on its own and follow arbitrary instructions found there. Everything here modifies files in your project, so make a commit beforehand to be safe before running batches.
Read next: Headless WordPress with Next.js: REST API, Image Handling, and SEO Optimization

Chưa có bình luận nào. Hãy là người đầu tiên chia sẻ góc nhìn!