What the frontend-design skill actually gets done
It does not build the page for you. It changes the decisions made before any CSS is written. The skill's own description says it provides "guidance for distinctive, intentional visual design when building new UI or reshaping an existing one", with emphasis on aesthetic direction, typography, and choices that "don't read as templated defaults".
In practice that means four things change:
A design plan comes first. The model is told to commit to a small set of named colour values, the typefaces and the role each one plays, a layout concept, and the reasoning that makes this page different from a default. You see the plan before you see markup.
The brief is grounded first. If you have not said what the product is, it is told to identify a concrete subject, an audience and the page's primary job, and confirm those with you — rather than design for "a modern startup".
Typography gets rules, not vibes. At most two families, and they have to be clearly distinguishable rather than two similar-grotesques side by side; a deliberate type scale instead of six ad-hoc sizes; line length that stays readable; more line-height when the body text is serif.
A list of defaults to avoid. The skill names the visual tells it does not want — the warm beige background with a terracotta accent, the SaaS card grid, the all-caps micro-label on every block — and asks for a point of view instead.
When it triggers — and when it doesn't
It loads when the work is UI: a landing page, a marketing page, a dashboard, a settings screen, or a redesign of one of those. The trigger is visual direction, not code volume.
Reach for a different skill when:
- You already have a design system and want it followed. That is brand-guidelines — it hands the model your rules. frontend-design is there to invent a direction, which is the opposite job.
- You need the same colours and type on page two through ten. That is theme-factory, which produces a reusable token set so the second page does not drift away from the first.
- The output is an image, poster or diagram. That is canvas-design. Asking a UI skill for a print asset gets you a web layout with a poster's typography.
- You want a component library wired up. This skill has opinions about palette and type; it does not install your component system or match its API.
How to check it's working
Ask for a page whose subject and audience are specific — "a booking page for a climbing gym, aimed at people who have never climbed". If the first reply contains a short design plan naming hex values and typefaces before any markup, it loaded. If it starts writing CSS immediately and the result is a cream background with a serif display face, it did not.
A second check that costs nothing: ask "what is your design direction and why". A loaded skill answers with a named direction tied to the subject. An unloaded one answers with adjectives.
Common failures
The output still looks templated. The guidance is advisory, not enforced, and a vague brief gives it nothing to anchor on. Name the product, the audience, and the single thing the page has to accomplish.
Two typefaces that read as one. The skill asks for families that are clearly distinct. If your brief says "keep it simple", say "two clearly different families" instead — simple is what produces the defaults.
Styles cancelling each other out. The skill explicitly warns about CSS specificity getting in the way: a class-based rule and an element-based rule for the same block can cancel the intended padding. Look at the cascade before blaming the design advice.
Your brand gets overwritten. When the brief pins a direction the model is told to follow it exactly. If you have fixed colours or fonts, paste them and say they are fixed.
It designs instead of shipping. You get a plan and a strong opinion and want the other thing. Say "brief is fixed, skip the exploration" and it will move to markup.
How it compares with the other design skills
frontend-design supplies a point of view for one interface. theme-factory turns a direction into reusable colour and type tokens you can carry across a project. brand-guidelines enforces rules you already have rather than inventing new ones. canvas-design produces images rather than pages. web-artifacts-builder outputs a single self-contained HTML file you can open, share or drop into a chat.
If more than one of those describes your task, the order that works is: frontend-design for the direction, theme-factory to make it reusable, brand-guidelines if a real brand exists and must win.
Try a sample task
This walkthrough gives you a prompt and an acceptance check to run in your own client.
Build and review a responsive event page
Use an empty frontend project and provide the event details, venue and registration requirements.
Use frontend-design to create a responsive workshop landing page with one registration action and an accessible agenda. Check it at 375 px and 1440 px.
What to verify A working registration link, readable agenda, visible keyboard focus and no horizontal overflow at either width.
Best for Landing pages, product UI and interfaces that need a coherent visual direction.
How to use frontend-design
Specify the page purpose, relevant brand constraints, the main action, and what the design should convey before implementation.
Acceptance check Use a keyboard and a narrow viewport to check search, navigation and text contrast before approving the design.
Install for your agent
Run the command from your project directory. After installation, reload the client if it has already started. Tool access and execution features depend on the client.
Claude Code
npx skills add anthropics/skills --skill frontend-design -a claude-codeCodex
npx skills add anthropics/skills --skill frontend-design -a codexCursor
npx skills add anthropics/skills --skill frontend-design -a cursorGitHub Copilot
npx skills add anthropics/skills --skill frontend-design -a github-copilotAsk your agent to install the skill
Install frontend-design from anthropics/skills in this project. Show me the destination folder and report any dependencies the source lists.
For manual installation, copy the entire skill folder, including references and scripts, into a skill directory supported by your client. See the installation guide.
What the upstream guide covers
The upstream guide starts by choosing a design direction for the audience and task. It connects that choice to typography, color, motion and spatial composition in the implementation.
HavenSkills workflow check
No workflow check recorded yet. Use the sample task above to make your own check in your client.
Requirements, permissions and limits
- Runtime
- A writable project; browser or preview for visual checks
- File and tool access
- The client needs write access to the project for implementation. Browser previews and package installs depend on the host client’s permissions.
- Limits to check
- The skill provides design and implementation guidance. Accessibility, layout and browser behavior still need verification against the delivered interface.
Frequently asked questions
Is the frontend-design skill free?
It is open source. The repository README lists the skills as Apache 2.0 and the SKILL.md points at LICENSE.txt in the skill folder for the full terms.
Does it generate a whole website?
No. It changes how the model chooses palette, type and layout; the model still writes the code. It will not scaffold a project or pick your framework.
Will it override my existing design system?
Only if you let it. The skill is told to follow a pinned brief exactly, so paste your colours, fonts and spacing rules and mark them as fixed.
Does it need an image model or a screenshot?
No. It works from a written brief. That is also its weakness: the more specific the brief, the more it has to work with.
Does it work in Cursor, Codex or Gemini CLI?
Yes. A skill is a folder containing SKILL.md, and any client that reads that format can load it — one install command covers the terminal clients.