How do you redesign an existing app UI with Claude Code?

Most UI work is not a blank page; it is an existing app that looks dated. redesign-existing-projects, at 182,490 installs (skills.sh registry), is the skill built for exactly that job.

Key takeaways

• redesign-existing-projects has 182,490 installs and 60,436 GitHub stars (skills.sh registry, GitHub); the wider field is ranked on the best skills for frontend design pillar.
• Redesigns break accessibility more often than greenfield builds - pair the redesign pass with fixing-accessibility (14,306 installs, 3,598 stars).
• Its sibling design-taste-frontend (235,019 installs) sets the visual baseline the redesign converges toward.
• Work screen by screen with the existing component tree; wholesale rewrites lose behavior that took years to harden.
• If retention is the motive, improve-retention (2,898 installs) targets the UX changes that keep users, not just the paint.

When is a redesign worth doing at all?

Redesign when the UI blocks a measurable goal - activation, retention, accessibility compliance - and not merely because it looks old. The design-taste family's adoption tells the demand story: its three main skills together account for over 600,000 installs (235,019 + 185,533 + 182,490, skills.sh registry), and redesign-existing-projects alone is the third most-installed skill in the frontend design category family it belongs to. Developers install it after shipping features on top of a five-year-old interface, which is the common case. A useful test before starting: name the metric the redesign should move, and write it at the top of the branch description so every screen-level decision answers to it.

What is the best skill for redesigning an existing UI?

redesign-existing-projects by leonxlnx is the best pick at 182,490 installs and 60,436 GitHub stars (skills.sh registry, GitHub). Install it with:

npx skills add leonxlnx/taste-skill

Unlike greenfield taste skills, it reads the project that is already there - component structure, routing, styling conventions - and applies its design rules as a migration, not a rewrite. Strong companions: design-taste-frontend (235,019 installs) for the target aesthetic, and high-end-visual-design (185,533 installs) for the marketing surfaces you touch on the way. The complete ranking lives on the best skills for frontend design page.

How do you keep a redesign from breaking the app?

Constrain the blast radius to one screen or route per session, and make the agent verify behavior before styling. The skill's own rules require dependency verification against package.json before importing anything new, which prevents the classic redesign failure of dragging in a second component library. Keep state management and data fetching untouched in the first pass; a redesign that changes visuals and logic in the same diff is unreviewable. Google's engineering review guidance makes the same argument for small, single-purpose changes.

How do you handle accessibility during a redesign?

Treat accessibility as a gate, not a follow-up. Color and contrast changes are where redesigns most often regress, so run fixing-accessibility (14,306 installs, 3,598 stars, skills.sh registry, GitHub) after each screen and hold the result against WCAG criteria. The accessibility hub lists the wider tooling, from audit skills to compliance checks, if the project has formal requirements.

What does a redesign workflow look like in practice?

Four passes per screen, in order, each one small enough to review and revert on its own. First, an audit: ask Claude Code to describe the current screen's hierarchy and list what fights the visual goal. Second, structure: fix layout and spacing with the redesign skill's variance rules. Third, polish: typography, color, motion via the taste baseline. Fourth, verification: accessibility check plus a behavioral click-through. The UX design hub covers the research skills that tell you which screens deserve the effort first.

A complete starter stack

• redesign-existing-projects (182,490 installs) - the migration-style redesign engine.
• design-taste-frontend (235,019 installs) - the target aesthetic baseline.
• fixing-accessibility (14,306 installs) - the regression gate after every screen.
• improve-retention (2,898 installs) - ties the redesign to a retention outcome.

Common pitfalls

• Redesigning every screen in one branch - review becomes impossible and the rollback story disappears.
• Swapping the component library mid-redesign; restyle what exists before replacing it.
• Skipping the accessibility pass because "it worked before" - contrast regressions are invisible in a diff and show up in complaints.

Pick one screen, run the four passes, ship it, and repeat.

Common questions