UI Skills by ibelick Review: Verdict, Safety and Alternatives
UI Skills by ibelick bundles seven instruction-only UI skills for coding agents, plus a registry. Our verdict is Recommended with caveats, 72/100, safety grade B. The motion skill has a measured gain from a third-party eval. The accessibility gain is small, and the metadata skill shows none. The baseline-ui skill hard-codes a React and Tailwind stack, so pick skills one by one.
- Skill
- UI Skills (ibelick/ui-skills)
- Publisher
- ibelick
- Verdict
- Recommended with caveats
- Caveat
- baseline-ui hard-codes Tailwind, motion/react and Base UI; metadata shows no uplift
- Score
- 72/100
- Confidence
- Moderate
- Safety grade
- B (skills are Markdown; router fetches third-party registry text)
- Best for
- React and Tailwind teams running motion and accessibility review passes
- Poor fit
- Projects not on Tailwind or React; distinctive visual design generation
- Reviewed version
- v0.2.3, commit ebf5f26 (2026-09-30), 7 skills
- Review date
- 2026-09-30
- Verified in
- none yet
- Claimed
- Claude Code, Cursor, Codex
- Since review
- 2 new skills (component-canvas, create-canvas) not reviewed
Get the skill: ui-skills repository on GitHub · SKILL.md at the reviewed commit · skills.sh listing
What are UI skills, and what is ibelick's UI Skills bundle?
UI skills are Agent Skills that teach coding agents interface rules, such as accessibility, motion and metadata. The ibelick UI Skills bundle holds seven reviewed instruction-only skills. It adds a registry, CLI and MCP server.
"UI skills" can also mean the job skills of a human UI designer. This page covers the other sense: Agent Skills that coding agents load. An Agent Skill is a folder with a SKILL.md file of plain instructions. The agent reads that file when a task matches the skill's description. Our guide to what a SKILL.md file holds explains the format in full.
ibelick is the GitHub handle of the publisher. The ibelick/ui-skills repository holds the skills, the ui-skills.com website, a command-line tool and an MCP server. The website lists about 200 third-party UI skills beside ibelick's own. We reviewed seven skills at commit ebf5f26 on 2026-09-30. Two more skill folders arrived in October, so the repository held nine on 2026-10-21.
| Skill | Job | Size or status | Measured result (Tessl) | Installs (skills.sh) |
|---|---|---|---|---|
| baseline-ui | Deslop constraints for agent-built UI | Mandates Tailwind, motion/react and Base UI | Quality 69, no eval | 21.3K |
| fixing-accessibility | Accessibility pass on forms, dialogs and icon buttons | Compact prioritised rule table | 99%, 1.06x uplift | 19.4K |
| fixing-motion-performance | Compositor-safe animation fixes | 5,565 bytes, about 1,391 tokens | 96%, 1.23x uplift | 23.9K |
| fixing-metadata | Titles, canonical, Open Graph and JSON-LD | Dense prioritised rules, 3.4 to 5.6 KB like the other fixing skills | 94%, 1.00x (no uplift) | 16.2K |
| improve-ui | Read-only UI audit that writes a plan | Plan template plus OpenAI agents manifest | Quality 75, no eval | 5.9K |
| create-design-md | DESIGN.md from a repo or public website | 16 KB; runs npx @google/design.md | Quality 70, no eval | 3.2K |
| ui-skills-root | Router to other UI skills | Loads at most three skills | Quality 60, no eval | 5.6K |
| component-canvas | New skill, added 2026-10-13 | Not reviewed | None | Not counted |
| create-canvas | New skill, added 2026-10-20 | Not reviewed | None | Not counted |
The seven reviewed SKILL.md files total about 43.6 KB. Each skill installs on its own, so a team takes one and skips the rest. Tessl is a third-party skill registry that runs scenario evals and quality reviews. Its "uplift" figure compares results with the skill against results without it. The skills.sh site is an open skills directory that counts installs. Together the seven skills had about 95.5K installs in September 2026.
Are ibelick's UI Skills worth installing?
Yes, ibelick's UI Skills are worth installing selectively. They score 72/100, Recommended with caveats, at Moderate confidence. Tessl measured a 1.23x motion performance uplift; accessibility gains are small, and metadata shows no uplift.
These are tight, opinionated UI rule sets with one measured win, on motion performance. Gains elsewhere are small or unmeasured. The repository's attention has moved to its registry website.
The main strength lies in two fixing skills. The fixing-motion-performance and fixing-accessibility skills are compact, prioritised rule tables. Tessl ran 3 scenarios for each one. Motion scored 96% with a 1.23x uplift. Accessibility scored 99% with a 1.06x uplift, so a strong base model already does most of that work.
The caveat sits in baseline-ui. It hard-codes Tailwind, motion/react and Base UI preferences. The fixing-metadata skill showed no uplift at 1.00x. A WCAG 2.2 pull request (#37) closed without a maintainer reply. The skill files have not changed since July.
Best for React and Tailwind teams whose agents ship janky animation, inaccessible controls and generic AI-styled UI. Apply the skills as review passes on single files. Poor fit for projects not on Tailwind or React, where baseline-ui's stack mandates conflict with the codebase. Poor fit for teams that want distinctive visual design generation instead of restraint rules.
| Parameter | Score |
|---|---|
| Effectiveness | 74 |
| Reliability | 70 |
| Safety | 74 |
| Efficiency | 84 |
| Compatibility | 70 |
| Maintainability | 62 |
| Setup friction | 84 |
| Motion performance correctness | 82 |
| Accessibility coverage | 72 |
| Stack-mandate overreach (worst failure) | 60 |
| Instruction weight vs value | 85 |
| Cost of adoption | 82 |
| Registry routing trust | 58 |
The missing evidence is a test of baseline-ui and improve-ui on a non-Tailwind codebase. Such a test shows whether the stack mandates cause unwanted rewrites.
Our page on how verdicts are weighed gives the full rubric. The hub of UI and frontend design skills, ranked places this bundle among its peers.
How many tokens do UI Skills use?
The anchor UI skill, fixing-motion-performance, measures about 1,391 tokens per load. The other fixing skills run about 850 to 1,400. Our estimate for one component review is 6,000 input and 1,500 output tokens.
| Figure | Label | Value |
|---|---|---|
| fixing-motion-performance SKILL.md | Measured file size per load | 1,391 tokens (5,565 bytes) |
| Motion or accessibility review of one 300-line React component | Our estimate for a typical task | 6,000 input, 1,500 output |
| Same task with the anchor skill loaded | Our estimate, including the skill file | 8,391 input, 1,500 output |
The two labels answer different questions. The measured figure is the skill file alone. The estimate covers the whole task, including the component code. We assume about 5 reviews a week. The hidden cost sits in the router. Following ui-skills-root pulls up to three more third-party skills from the registry into context. The create-design-md skill downloads an npm package at runtime.
What changed since our review?
The seven UI Skills we reviewed have not changed since 2026-09-30. Two new skills, component-canvas and create-canvas, arrived in October without review. No new release followed v0.2.3, so the verdict stands.
We checked the repository through the GitHub API on 2026-10-21:
- 2026-10-13: component-canvas arrived in commits ffb4935 and 36f7f56. We have not reviewed it.
- 2026-10-16: ui-skills.com added Google and GitHub sign-in through Better Auth.
- 2026-10-16: PR #99 fixed broken registry entries on the site.
- 2026-10-20: create-canvas arrived in commit 32afd57. We have not reviewed it.
- 2026-10-20: the site added saved collections and a "deslop" collection.
- 2026-10-21: stars stood at 9,911, up from 9,292 at review.
The registry added Figma and HyperFrames publisher skills in the same period. The latest release is still v0.2.3, dated 2026-06-22.
Which UI skills are best for AI agents?
The best UI skills for AI agents depend on the job. Frontend Design (76) leads page generation, and ibelick's UI Skills (72) suit audit passes. Vercel's Web Design Guidelines (66) runs a file:line rule review.
The criterion here is our verdict score, with the UI job each skill does best.
| Skill | Verdict | Score | Grade | Best UI job |
|---|---|---|---|---|
| Emil Kowalski's design engineering review | Recommended | 76 | A | Animation and interaction review |
| Frontend Design verdict | Recommended | 76 | A | Generating a new page with a visual direction |
| UI Skills (this page) | Recommended with caveats | 72 | B | Motion and accessibility audit passes |
| Jakub Krehel's interface skills | Recommended with caveats | 70 | A | Named interface review rules |
| UI/UX Pro Max verdict | Conditional | 68 | B | Palette, font and UX-rule lookup |
| Vercel's Web Design Guidelines review | Conditional | 66 | C | File:line guideline review |
| Platform Design Skills review | Conditional | 63 | A | Platform convention checks |
Are UI Skills safe to install?
Yes, UI Skills are safe to install, with one caveat: they carry safety grade B. The seven skills hold no code. The router sends agents to fetch third-party skill text from ui-skills.com, and create-design-md runs an npm package.
Our review of the files at commit ebf5f26 found eight facts:
- Files: seven SKILL.md files total about 43.6 KB, and the anchor file is 5,565 bytes.
- Code: the skills contain no executable code.
- Extras: improve-ui adds one Markdown plan template and an OpenAI agents manifest that allows implicit invocation.
- Packages: create-design-md tells the agent to run npx @google/design.md lint, export and diff.
- Router: ui-skills-root sends agents to npx ui-skills, which fetches registry text and prints it to stdout.
- Registry: about 200 third-party skills are served by raw URL, outside this publisher's control.
- Website: onedollarstats analytics and Cloudflare middleware run on the site, not inside the skills.
- Writes: improve-ui writes only to design-plans/, and create-design-md writes only DESIGN.md.
No hooks exist. GITHUB_TOKEN and GH_TOKEN are read only by maintainer scripts, not by the skills. Grade B reflects the router and the npm call, since routed third-party text becomes agent instructions. That makes the registry a prompt-injection surface. Read any routed skill before you let an agent follow it. Our checklist on how to vet a skill and its sources covers that review step by step.
What do users and maintainers report about UI Skills?
Users and maintainers report measured gains for two UI Skills. Tessl found 1.23x for motion and 1.06x for accessibility. Issue #40 found 18 of 176 registry URLs dead, later fixed in PR #56.
These are the 14 dated reports we weighed, 5 of them strong, from GitHub, Hacker News, Tessl, skills.sh and the web. Strong items come first.
| Claim | Source | Date | Strength |
|---|---|---|---|
| fixing-motion-performance scored 96% with a 1.23x uplift over 3 scenarios. | Tessl | 2026-09 | Strong |
| fixing-accessibility had quality 91, 99% across 3 scenarios, and 1.06x uplift. | Tessl | 2026-09 | Strong |
| fixing-metadata had quality 84 and 94%, with 1.00x and no uplift. | Tessl | 2026-09 | Strong |
| A WCAG 2.2 coverage PR closed unmerged without a maintainer reply. | PR #37 | 2026-07 | Strong |
| An audit found 18 of 176 registry URLs returning 404. | Issue #40 | 2026-07 | Strong |
| baseline-ui had quality 69, passed a Snyk scan, and has no eval. | Tessl | 2026-09 | Weak |
| create-design-md had quality 70 with low-risk Snyk findings. | Tessl | 2026-09 | Weak |
| ui-skills-root had quality 60 and improve-ui 75, neither with an eval. | Tessl | 2026-09 | Weak |
| The maintainer added a health check and fixed or removed 24 entries. | PR #56 | 2026-08 | Weak |
| A user found it hard to tell what a skill's output looks like. | Issue #22 | 2026-06 | Weak |
| A before-and-after screenshot PR was declined. | PR #39 | 2026-07 | Weak |
| A community registry submission closed unmerged after three months. | PR #21 | 2026-06 | Weak |
| A Hacker News submission drew 5 points and no discussion. | Hacker News | 2026-07 | Weak |
| The bundle had about 95.5K installs in total. | skills.sh | 2026-09 | Weak |
We found no independent bug reports about the skills themselves.
What are UI Skills' known failures?
At v0.2.3, UI Skills have four known failures. The baseline-ui skill mandates Tailwind and motion/react without checking the project. The fixing-metadata skill adds no measured uplift, WCAG 2.2 is missing, and registry content is unaudited.
- Stack mandates: baseline-ui says MUST use Tailwind defaults, motion/react and
cn, with no check that the project uses Tailwind. - Flat metadata gain: fixing-metadata scored 94% but showed 1.00x, so it adds nothing over the base model.
- Missing WCAG 2.2: PR #37 added target size and focus-appearance criteria, then closed unmerged without a reply.
- Unaudited routing: the registry once had 18 dead URLs (#40), and routed content stays third-party text.
Our review names the worst case in these words:
The agent applies stack-specific mandates (Tailwind defaults, no gradients, Base UI) to a codebase or brand that intentionally differs, producing churn or regressions presented as fixes.
The baseline-ui skill does include "unless explicitly requested" escapes and prefers existing primitives. Maintenance is the slower risk. The skill files last changed in July 2026, while the website kept moving. Community PRs #21, #43, #44 and #68 went unanswered.
Do UI Skills work in Claude Code, Codex and Cursor?
No agent is verified for UI Skills yet. The publisher claims UI Skills work in Claude Code, Cursor and Codex. The improve-ui skill ships an OpenAI agents manifest, and Antigravity reads the format untested.
| Agent | Status | Evidence |
|---|---|---|
| Claude Code | Claimed | ui-skills-root SKILL.md |
| Cursor | Claimed | ui-skills-root SKILL.md |
| Codex | Claimed | ui-skills-root SKILL.md; improve-ui OpenAI agents manifest |
| Antigravity | Reads the same format; no published run | Plain SKILL.md format |
| OpenAI, Copilot, Gemini, DeepSeek, OpenClaw | Claimed by the publisher | ui-skills.com home page |
Plain SKILL.md files work in any skills-aware agent. The baseline-ui skill still assumes React plus Tailwind, whatever the agent. Verified means a dated run on that agent, and none exists for this bundle.
How do you install UI Skills?
You install ibelick's UI Skills one at a time with the open skills CLI. Run npx skills add https://github.com/ibelick/ui-skills --skill fixing-motion-performance, for example. The ui-skills CLI prints skill text without writing files.
The bundle has no Claude Code plugin. The repository root had no .claude-plugin folder on 2026-10-20. Skills install as plain folders instead, and no credentials are needed.
How do you install UI Skills in Claude Code?
- Run
npx skills add https://github.com/ibelick/ui-skills --skill fixing-motion-performancefor the anchor skill. - Swap the
--skillvalue for fixing-accessibility, fixing-metadata or another skill name. - Copy one skill folder into
~/.claude/skills/as the manual alternative.
Our walkthrough on installing skills in Claude Code explains personal and project scopes.
How do you use the ui-skills CLI and MCP server?
The CLI browses the registry and prints skill text. The README lists these commands:
- Run
npx ui-skills startto begin. - Run
npx ui-skills categoriesto see the registry categories. - Run
npx ui-skills list --category motionto list skills in one category. - Run
npx ui-skills get baseline-uito print one skill.
The MCP endpoint is https://www.ui-skills.com/mcp, with two tools: list_skills and get_skill. Both paths serve third-party skills beside ibelick's. Treat their output as untrusted instructions until you read it. We assessed the MCP server only as far as the router finding. Our explainer on skills and MCP compared covers the general split.
How do you use UI Skills on a project?
Use ibelick's UI Skills on a project as review passes, one file at a time. Run fixing-motion-performance on an animated component, then check behaviour is kept. Skip baseline-ui's stack mandates on projects without Tailwind.
- Pick one fixing skill for the job in front of you.
- Run it on one file, such as an animated component.
- Read the diff line by line.
- Confirm the component behaves as before.
- Repeat the pass on the next file.
Our review lists six use cases. Two are primary, and baseline deslop is the most-installed secondary one:
- Motion review: rewrite a component to compositor-only motion, weekly during UI work (Tessl eval, 1.23x).
- Accessibility pass: check forms, dialogs and icon buttons, once per feature (Tessl eval, 1.06x).
- Baseline deslop: constrain agent-generated UI, once per session, on Tailwind and React only (no eval, 21.3K installs).
The improve-ui plans and DESIGN.md generation rest on the publisher's SKILL.md files only. Nobody has reported results for them.
Test it yourself on one component first:
- Run /fixing-motion-performance on one existing animated component.
- Check that its fixes keep the component's behaviour.
- Confirm the fixes remove layout-property animation.
What are the alternatives to ibelick's UI Skills?
The main alternatives to ibelick's UI Skills are Frontend Design and Jakub Krehel's skills. Frontend Design generates pages; Krehel's skills apply named review rules. Platform Design Skills audits native and web screens against platform conventions.
- Frontend Design: Recommended, 76, for generating a page with a clear visual direction.
- Jakub Krehel's Interface Skills: Recommended with caveats, 70, for named interface review rules.
- Platform Design Skills: Conditional, 63, for platform convention checks.
- Web Design Guidelines: Conditional, 66, for a file:line guideline review.
Frontend Design lives in anthropics/skills, and Web Design Guidelines lives in vercel-labs/agent-skills. The ui.sh and AccessLint projects publish UI skills, and we have not reviewed either. The design hub lists every UI skill we rated, with grades and scores.
Where are UI Skills on GitHub?
UI Skills are on GitHub in the ibelick/ui-skills repository, MIT-licensed, beside the registry website, CLI and MCP code. We reviewed release v0.2.3, commit ebf5f26, on 2026-09-30. The skill files last changed in July.
- Repository: ibelick/ui-skills on GitHub, under the MIT licence.
- Size: 257 files at review, website code included.
- Popularity: 9,292 stars and 423 forks at review, with 7 contributors.
- Releases: 8 in 12 months, the latest v0.2.3 on 2026-06-22.
- Version: package.json reads 0.2.4, ahead of the v0.2.3 tag.
- Commit: ebf5f26, pushed 2026-09-30 and reviewed the same day.
Common questions
Do UI Skills generate designs?
No. UI Skills constrain and audit interfaces, and Frontend Design is the generator in our ratings.
Are UI Skills free?
Yes. The bundle is MIT-licensed and free to install.
Can I install only one UI skill?
Yes, each skill installs alone with the --skill flag. Our guide on when a skill does not load helps after install.
Do UI Skills work without Tailwind?
Yes, the fixing skills do. The baseline-ui skill assumes Tailwind, motion/react and Base UI.
