Living collection
Design and frontend | Skills for Claude Code, Codex and Coding Agents by Task
27 resourcesOrganized by use
Design and frontend
Design and frontend
9 resourcesFull profiles and sources
Impeccable
Skill
Design rules and review prompts for agent-built interfaces
Claude CodeCodex
Turns frontend design into decisions about hierarchy, color, typography, interaction and finish.
Best for: Reviewing and polishing interfaces generated by coding agents.
Why it belongs here. Gives an agent concrete design criteria for evaluating an interface instead of asking for vague visual polish.
Its limit. Rules and prompts do not establish that a screen serves its users; review the result in the actual browser and task context.
A set of instructions that helps AI design better and review its work. You can ask it to tidy up a screen, improve the lettering or remove distracting decoration. It helps polish a website without you having to explain every design detail from scratch.
Agents and setupWhere it works and what you need
Environments. The published review lists Cursor, Claude Code, GitHub Copilot, Gemini CLI, Codex CLI, Grok Build, Hermes Agent, OpenCode and other hosts.
Installation. Add it with npx impeccable install. Then /impeccable init defines product context and /impeccable audit reviews the interface.
Real cost and licenseSkill, model and external services
License. Apache-2.0 and free use according to the published review.
Detector. The detector does not need an API key or language model. The agent or model applying corrections has separate costs.
Use case and alternativeA specific task and another way to solve it
Example. Review an agent-generated campaign page and correct hierarchy, typography, color, interaction and generic visual patterns before publication.
Another route. For a one-off review, use design guidelines or a designer. Impeccable adds more when you want to repeat the same checks inside the agent workflow.
Sources and scopeWhat is documented and what was tested
Evidence. This entry uses the existing Marketing Ultra review and the project's official source. Author documentation is not described as a hands-on test by Dani.
Resource detailsEditorial profile and links
Claude Code /design
Claude Code design workflow
Going from an idea to several visual proposals helps you discuss what you want to build. Deciding is easier when you have something concrete in front of you instead of a description.
I would judge its value all the way through to the finished implementation. If the prototype looks great but has to be rebuilt from scratch, the savings disappear. I find it interesting for exploring and clarifying ideas, while watching the handoff to code closely.
Connect a local component library to Claude's design workflow
A Claude Code design workflow that syncs local HTML components and previews with a design-system project in claude.ai.
Best for: Reusing a project’s design-system components and references.
Why it belongs here. Connects existing visual components to an agent's interface generation work.
Its limit. The Claude-specific workflow does not establish an equivalent feature in Codex or another agent.
It gives Claude a reference for your project’s design so it can reuse the pieces you already have when creating another screen. That helps it keep the same buttons, colours and styles instead of inventing a different-looking website every time you ask for something.
Agents and setupWhere it works and what you need
Workflow. The described route uses /design-sync to share components and previews with a design-system project.
Requirements. The existing profile records a claude.ai session, a design_system project and a limit of 256 files per write operation.
Other agents. This Claude-specific workflow does not establish an equivalent feature in Codex. Keep compatibility claims tied to the documented route.
Real cost and licenseSkill, model and external services
Product. The existing profile identifies this as a proprietary Anthropic tool.
Access. Depends on a plan with access to the feature. Check current availability and plan terms before making a purchase decision.
Use case and alternativeA specific task and another way to solve it
Example. Share button and card HTML components with previews so subsequent screens start from the same library.
Another route. Keep components and visual rules in the repository and give them to the agent as context. This requires manual transfer and does not reproduce claude.ai synchronization.
Sources and scopeWhat is documented and what was tested
Evidence. This entry uses the existing Marketing Ultra review and the project's official source. Author documentation is not described as a hands-on test by Dani.
Resource detailsEditorial profile and links
UI UX Pro Max Skill
Resource
Organized references for styles, typography and design rules can give an agent some direction. I see a use for that in a first proposal.
A large number of options does not guarantee a website with personality. You need to choose what fits the brand and check that the page does its job. I would use it as starting material without letting the catalog decide for me.
For giving an agent design criteria before it builds or reviews an interface. Turns any coding agent into a design-aware assistant: 192 industry-specific rules, 79 searchable styles, 74 typographic pairings, and automatic generation of complete design systems before any code is written.
Best for: Preparing design criteria and a design system before writing the interface.
Why it belongs here. For giving an agent design criteria before it builds or reviews an interface. Turns any coding agent into a design-aware assistant: 192 industry-specific rules, 79 searchable styles, 74 typographic pairings, and automatic generation of complete design systems before any code is written.
Its limit. No MCP integration: works as a skill installed in the assistant's directory, not as a server external tools can connect to. The free version generates no real visual assets and the premium tier has no published price.
It helps AI decide what a website should look like before building it. Tell it who the site is for and what you want to achieve, and it guides choices about colours, lettering and layout. That gives it more direction than “make me a nice website”.
Agents and setupWhere it works and what you need
Compatibility. Works with more than 20 assistants: Claude Code, Cursor, Windsurf, Kiro, GitHub Copilot, Codex CLI, Gemini CLI, Warp, Augment, and others. Supports 22 tech stacks (React, Next.js, Vue, Svelte, Flutter, SwiftUI, Tailwind, etc.). Requires Python 3.x and Node.js/npm to install the CLI.
Before using it. Provide the industry, content and brand constraints. Check suggested rules and combinations on a real screen before making them the product’s system.
Real cost and licenseSkill, model and external services
Cost. Basic version free under MIT license. Premium version with brand identity, AI-powered visual asset generation, and dedicated support; price not published on the official website as of the review date (September 2026).
What to budget for. Check what the basic version delivers and what premium advertises. Access to design guidance does not necessarily include visual assets or their production.
Use case and alternativeA specific task and another way to solve it
Example. Install the CLI with npm install -g ui-ux-pro-max-cli, run uipro init --ai claude in your project, and ask "Build a landing page for my SaaS product". The skill generates a complete design system (palette, typography, pattern, antipatterns) before the agent writes any code.
Alternative. For injecting real visual references instead of abstract rules, Refero uses web screenshots via MCP. For ready-made components, conventional UI kits (shadcn/ui, Tailwind UI) solve the problem without an intermediate skill.
Sources and scopeWhat is documented and what was tested
Review. Reviewed from the repository README and the official website (uupm.cc) in September 2026. No hands-on test run; conclusions based on public documentation.
Resource detailsEditorial profile and links
Frontend Design de Anthropic
Resource
I find it pretty basic. It is fine to get started: it gives the agent some design guidelines and moves you beyond asking for a modern website without specifying anything else. But compared with Impeccable and other options, it falls short for me.
I give it 79 because it is useful as an entry point. At this stage I expect more from a design skill, and better alternatives exist.
For giving an agent design criteria before it builds or reviews an interface. Official Anthropic skill for giving coding agents art direction. Free, compatible with Claude Code, Claude.ai, and the Claude API (and with Codex and Cursor per the Agent Skills standard). Forces the agent to plan palette, typography, and composition before writing code.
Best for: Defining a visual direction before generating an interface with AI.
Why it belongs here. For giving an agent design criteria before it builds or reviews an interface. Official Anthropic skill for giving coding agents art direction. Free, compatible with Claude Code, Claude.ai, and the Claude API (and with Codex and Cursor per the Agent Skills standard). Forces the agent to plan palette, typography, and composition before writing code.
Its limit. Does not generate original design or replace a proper design system. Results depend on the brief, references, and the model. The license is proprietary (Anthropic Commercial Terms), not open source.
These instructions ask AI to think through a website’s design before building it: how to arrange the page, which colours to use and what lettering fits. The aim is to give it a clear direction so the result does not look like every other website.
Agents and setupWhere it works and what you need
Compatibility. Claude Code, Claude.ai, Claude API and, per the Agent Skills standard, also Codex and Cursor. Requires an agent compatible with SKILL.md.
Before using it. Provide the brand, real content and references with a clear purpose. Then review the rendered interface: design instructions alone do not check readability or mobile use.
Real cost and licenseSkill, model and external services
Cost. Free. The skill has no cost, but the agent usage (Claude Code, API) has its own pricing depending on your plan.
What to budget for. Spending depends on the agent and the design iterations you request. Reusing approved components can reduce later corrections.
Use case and alternativeA specific task and another way to solve it
Example. In Claude Code, register the repository with /plugin marketplace add anthropics/skills and install with /plugin install example-skills@anthropic-agent-skills. Also available on Claude.ai (paid plans) and via the API. Ask the agent to create a landing page and the skill activates visual planning.
Alternative. Manual style prompts, UI UX Pro Max Skill (third-party), or the native Claude Code /design functionality (proprietary, a different product).
Sources and scopeWhat is documented and what was tested
Review. Review published on Marketing Ultra (September 2026), based on official documentation and the repository README. Does not include a locally executed test.
Resource detailsEditorial profile and links
Taste Skill
Resource
Being explicit about how much motion, density or variety you want can save a lot of back-and-forth with the agent. It is a useful way to pin down decisions that otherwise become requests to make something prettier.
Still, adjusting those controls does not resolve taste. Composition, messaging and the page's purpose continue to need judgment. I see value in directing a proposal more clearly, within that limit.
For giving an agent design criteria before it builds or reviews an interface. It turns three common design decisions—variety, motion and density—into explicit controls a SKILL.md-compatible agent can follow.
Best for: Adjusting hierarchy, density and motion in AI-generated interfaces.
Why it belongs here. For giving an agent design criteria before it builds or reviews an interface. It turns three common design decisions—variety, motion and density—into explicit controls a SKILL.md-compatible agent can follow.
Its limit. The default version 2 is still labelled experimental, and there is no independent evidence showing better results or real adoption.
It helps you tell AI what kind of design you want: simple or adventurous, calm or animated, sparse or full of information. It gives you a clearer way to steer the result than saying “this doesn’t feel right”.
Agents and setupWhere it works and what you need
Compatibility. The documentation states support for tools that accept SKILL.md files and names Codex, Claude Code, Cursor, Gemini CLI, v0, Lovable and OpenCode.
Before using it. Define the audience and content before adjusting variety or density. Review the complete page so style does not get in the way of the task.
Real cost and licenseSkill, model and external services
Cost. Taste Skill is open source under the MIT licence. Any costs for the agent, subscription or API used sit outside the tool itself.
What to budget for. Most work goes into iteration and interface verification. Use a representative screen to evaluate the guidance before applying it across the product.
Use case and alternativeA specific task and another way to solve it
Example. It can guide a B2B landing page with moderate variety, restrained motion and enough density to explain the offer without turning it into a dashboard.
Alternative. The verified alternative is manually reviewing every generated interface and requesting hierarchy, spacing and animation corrections one by one.
Sources and scopeWhat is documented and what was tested
Review. Official sources reviewed on 23 September 2026. Compatibility is stated by the project and does not amount to local testing.
Resource detailsEditorial profile and links
Emil Design Engineering (Emil Kowalski Skills)
Resource
I like that it turns motion and interactions into specific decisions: what purpose they serve, when they help and when they are unnecessary. That contributes far more than filling a page with animations because they look impressive.
I see it as a good complement for refining how an interface behaves. You still need to check how it feels to use, including on mobile. The author's judgment helps; the decision about your product remains yours.
For giving an agent design criteria before it builds or reviews an interface. It provides concrete criteria for agents to decide when to animate, for how long and how to handle microinteractions without adding motion for its own sake.
Best for: Reviewing microinteractions, animation and interface behavior.
Why it belongs here. For giving an agent design criteria before it builds or reviews an interface. It provides concrete criteria for agents to decide when to animate, for how long and how to handle microinteractions without adding motion for its own sake.
Its limit. It codifies Emil Kowalski’s prescriptive vision, but does not replace an accessibility audit, testing on real devices or the judgment of a senior designer.
It gives AI guidance on how a website should respond when you use it: menus opening properly, buttons reacting clearly and movement that does not get in the way. It helps polish the small details that make a page feel well made.
Agents and setupWhere it works and what you need
Compatibility. The README documents installation through npx skills@latest add emilkowalski/skills. The attached evidence does not confirm specific CLI destinations or a local installation for this review.
Before using it. Give the agent the component and the behavior you want to improve. Check the result with keyboard input, reduced motion and repeated interaction.
Real cost and licenseSkill, model and external services
Cost. The repository is published as open source under the MIT license and does not document a purchase price. Costs for the coding assistant or infrastructure used fall outside the available evidence.
What to budget for. Downloading the skill does not replace implementing and checking each adjustment. A focused review makes it easier to judge whether the interaction improves.
Use case and alternativeA specific task and another way to solve it
Example. Install the package and request a review of a dropdown with an ease-in curve, 400 ms duration and a centered origin, asking for a Before/After table.
Alternative. Ask a senior designer to manually review animations and interaction details in their real-world context.
Sources and scopeWhat is documented and what was tested
Review. Profile reviewed on September 25, 2026 using the repository, official page, README, the SKILL.md pinned to a commit and license information. No local test was run.
Resource detailsEditorial profile and links
Web Design Guidelines de Vercel
Resource
I like concrete checks that help catch problems before delivering an interface. A form, keyboard focus or awkward behavior can matter much more than another visual effect.
These rules help review work and prevent oversights. Visual identity and testing the actual website remain necessary. As a complement to the process, I see considerable value.
For giving an agent design criteria before it builds or reviews an interface. Provides a systematic review of interface code against more than 100 accessibility, performance and UX rules.
Best for: Reviewing interface accessibility, performance and usability from its code.
Why it belongs here. For giving an agent design criteria before it builds or reviews an interface. Provides a systematic review of interface code against more than 100 accessibility, performance and UX rules.
Its limit. It does not create a visual identity or certify legal or WCAG compliance on its own.
A checklist that helps AI look for usability problems on your website, such as a confusing form or a button you cannot reach with the keyboard. It is useful for spotting issues that a good-looking screenshot will never show you.
Agents and setupWhere it works and what you need
Compatibility. It is distributed through the Skills CLI. The public package page lists agents including Claude Code, Codex and Cursor.
Before using it. Ask for findings with file, element and rationale, then check them in the rendered interface. Rules support review; they do not certify the result.
Real cost and licenseSkill, model and external services
Cost. It is an open-source resource with an MIT licence declared in the README. The cost of the agent or its subscription is not part of the skill.
What to budget for. Include time to fix and verify findings as well as agent use. Applying a rule list does not replace usability and accessibility checks.
Use case and alternativeA specific task and another way to solve it
Example. Audit a landing page before launching campaigns and review it against rules involving forms, focus, images, navigation and interaction.
Alternative. Carry out a manual UI review with an accessibility and UX checklist, especially when testing with assistive technology or human interpretation is needed.
Sources and scopeWhat is documented and what was tested
Review. Profile checked on 25 September 2026 against the official README and public package page.
Resource detailsEditorial profile and links
Stitch Design Skills
Resource
Turning a project's visual rules into a reusable document seems like a good idea. If the agent preserves those decisions between deliveries, there is less to explain again.
What I want to see is that continuity reaching the result. If the handoff forces me to fix the same things again, the document alone has not saved work. I judge it by what it preserves and simplifies in a specific workflow.
For giving an agent design criteria before it builds or reviews an interface. It turns the visual rules of an existing front end into a reusable DESIGN.md and connects that document to other design and build workflows in the Stitch ecosystem.
Best for: Documenting and reusing design decisions in a Stitch workflow.
Why it belongs here. For giving an agent design criteria before it builds or reviews an interface. It turns the visual rules of an existing front end into a reusable DESIGN.md and connects that document to other design and build workflows in the Stitch ecosystem.
Its limit. The README requires Stitch MCP to be configured and running for the package. The repository licence does not remove access requirements or guarantee official support.
It records how your website is designed and prepares a guide to its colours, lettering and recurring elements. AI can consult that guide when creating new screens, so you do not have to keep reminding it that the pages should look like they belong together.
Agents and setupWhere it works and what you need
Compatibility. The README lists Codex, Antigravity, Gemini CLI, Claude Code, Cursor and OpenCode. OpenCode needs manual installation and adaptation of some names.
Before using it. Check the generated DESIGN.md against the project’s actual components. For agents requiring manual adaptation, verify names and paths before reusing the workflow.
Real cost and licenseSkill, model and external services
Cost. The repository code uses an Apache-2.0 licence. No single euro price has been verified for Stitch, the agents or the required external services.
What to budget for. Separate the code from accounts required for Stitch and the agent. Generating documentation and turning it into an interface can take several iterations.
Use case and alternativeA specific task and another way to solve it
Example. Analyse the /src folder and generate .stitch/DESIGN.md with detected colours, typography and components.
Alternative. The listing does not establish a direct substitute. Taste Skill and Anthropic's Frontend Design are related resources for managing visual judgement, not verified equivalents.
Sources and scopeWhat is documented and what was tested
Review. Documentation and repository structure checked on 27 September 2026; no personal testing or production results are claimed.
Official repository · Official Stitch MCP documentation · Stitch MCP setup
Resource detailsEditorial profile and links
UI Skills de ibelick
Resource
I like being able to pick specific instructions to solve an interface problem. Choosing what I need and adapting it fits how I work.
I would not install everything just to have more skills. If Emil or Impeccable already covers a rule, adding another version can create more noise than value. The point is to select the pieces that actually improve the work.
For giving an agent design criteria before it builds or reviews an interface. It lets you browse and download design engineering skills through a website, a CLI and an MCP connection.
Best for: Finding specialized instructions to review an interface problem.
Why it belongs here. For giving an agent design criteria before it builds or reviews an interface. It lets you browse and download design engineering skills through a website, a CLI and an MCP connection.
Its limit. The catalogue brings together resources from different authors, and verified sources do not demonstrate that every skill runs fully in every agent or certify WCAG compliance.
A catalogue of design instructions you can give your AI assistant. Find the task you need help with, choose a suitable set and add it to the assistant. It saves you starting from scratch when you want to improve a particular part of your website.
Agents and setupWhere it works and what you need
Compatibility. The website organises the catalogue by agents, including Claude Code, Cursor, Codex and GitHub Copilot. This does not prove that every skill has been tested end to end in all of them.
Before using it. Check the chosen skill’s author and content before installing it. An agent appearing in the catalog does not replace testing it with your project.
Real cost and licenseSkill, model and external services
Cost. The repository releases its code under the MIT licence. The assistant, APIs and other external services retain their own terms, and no single price has been verified.
What to budget for. Choose by the task solved, not by the number of installed skills. The agent consumes resources to apply them and changes need review.
Use case and alternativeA specific task and another way to solve it
Example. Browse the catalogue, choose a skill suited to the problem, and check its attribution and scope before applying changes.
Alternative. Write a focused project instruction for the problem you need to review. A catalog helps discover approaches; your own rule lets you keep only what applies to your interface.
Sources and scopeWhat is documented and what was tested
Review. Reviewed on 27 September 2026 using the repository, its README, the licence and the official website.









I like Impeccable because it turns many design decisions into something you can discuss and work through with the agent. Hierarchy, composition, typography, polish. There is judgment there that you can bring into the process.
But passing a checklist does not guarantee a good website. You need to look at the result and judge it. It remains a useful reference for me, with that clear limit, and I am keeping its 88.