Taste Skill Review: Verdict, Safety and Alternatives

Taste Skill (taste-skill on GitHub) is Leonxlnx's anti-slop frontend design skill, installed as design-taste-frontend. We rated the September 2026 file Conditional, 62/100, with safety grade B. Taste Skill is an Agent Skill: a SKILL.md package that coding agents load before writing frontend code. Our verdict covers commit ce26fc2, and v2.0.0 has since replaced that file.

Get the skill: taste-skill folder on GitHub · SKILL.md at the reviewed commit · skills.sh listing

Changed since our review: on 2026-10-13, v2.0.0 replaced the reviewed file under the same install name. The new skill runs Node 22+ scripts and Chrome, and fetches images from the network. Our verdict and safety grade cover only commit ce26fc2 (2026-09-26), reviewed on 2026-09-30. We have not reviewed v2, now at v2.1.3, yet.

What is Taste Skill?

Taste Skill is Leonxlnx's anti-slop frontend design skill, installed as design-taste-frontend. It makes coding agents choose a deliberate design direction. It bans named AI-generated visual tells before any page ships.

Leonxlnx is the GitHub account of Leon Lin, who publishes the skill. The site tasteskill.dev credits Leon Lin and blueemi as its builders. Its install name, design-taste-frontend, is what the skills CLI and the plugin load.

"AI slop" here means the default look that models produce without direction. The reviewed file names its tells precisely. Three examples are equal three-card rows, eyebrow labels above headings and em dashes in copy. Read what goes in a SKILL.md for the file format behind every Agent Skill.

The file we reviewed was one long rulebook. It worked through four mechanisms:

• Design read: the agent states the design direction in one line before it writes code.
• Three dials: the agent sets three tunable dials before it designs the page.
• Routing map: the agent sends briefs that need a real design system to that system.
• Pre-flight checklist: the agent checks the page against a long list of banned AI tells.

The current v2 works differently. Its Node scripts draw design directions, inspect the page in a real browser and search open image collections.

Which Taste Skill version did we review?

We reviewed the Taste Skill file at commit ce26fc2, a single 87 KB rulebook. On 2026-10-13, v2.0.0 replaced it under the same install name. The old line now lives in taste-skill-classics.

The repository had no release tags at our review. The plugin manifest said 1.0.0, and the flagship SKILL.md called itself "v2 experimental". Three lines now share the Taste Skill name, and the table maps each one.

Taste-skill-classics is a sister repository that Leonxlnx created for the older skills. The documentation at tasteskill.dev/docs still describes the section-numbered rules of v2 experimental. That page therefore matches the file we reviewed more closely than the file you install today.

Is Taste Skill worth installing?

Yes, Taste Skill is worth installing for React landing pages, with review. We rated the September file 62/100, Conditional, at Moderate confidence. Its rules were specific, but its motion skeletons carried open bugs.

The main strength was concrete guidance. The one-line design read, the three dials, the routing map and the pre-flight checklist all named exact patterns. The main weakness was weight and code quality. The flagship file was about 87 KB and loaded whole, and two canonical GSAP skeletons carried reported bugs.

The best fit is a developer who builds React or Next.js landing pages and portfolios with a strong model. That developer gets a concrete ban list and a design-read step before code is written. Two groups fit poorly:

• Dashboard builders: the reviewed file declared dashboards, data tables and multi-step product UI out of scope.
• Vue, Svelte and vanilla users: the React-specific defaults and the 22K-token body cost more than they return.

These scores describe the reviewed file only. The table lists all 13 of them.

No published test compares design-taste-frontend alone against a plain model on the same briefs.

Our page on how we pin a verdict to a commit explains our method. The best Claude design skills, ranked shows where Taste Skill sits among its peers.

How many tokens does Taste Skill use?

The Taste Skill file we reviewed measured about 21,813 tokens per load. Our estimate for one landing page is 35,000 input and 12,000 output tokens. The new v2 SKILL.md is 38,458 bytes.

The first figure is the skill body alone. The second covers a whole page job, brief and generated code included. The v2 figure covers SKILL.md only. Its references and script output load on demand, so a real v2 session costs more than 9.6K tokens.

What changed since our review?

Taste Skill changed completely after our review. Version 2.0.0 made design-taste-frontend a new skill with Node scripts, a browser inspector and network image search. Our verdict covers the older file and stays unchanged.

The CHANGELOG says design-taste-frontend "is now v2, a new skill… not an edit of the earlier files." The previous default "is preserved as taste-skill-v2-experimental". The dated changes come from the CHANGELOG, the release list and GitHub PR states. The release notes are vendor-reported.

Four big changes followed our review. The install name now delivers a new skill. Scripts and network access arrived. The scope widened to dashboards, admin and analytics views, settings and inboxes. Fixes for two reviewed defects merged. We have not reviewed v2 yet.

Is Taste Skill safe to install?

Yes, the Taste Skill file we reviewed was safe to install, with safety grade B. It was text only, with no network calls or hooks. The new v2 runs Node scripts and Chrome, and we have not graded it.

Grade B reflects one repo-level detail. The plugin installed the whole repository, which carried local README asset scripts that no skill invoked. Our reading of commit ce26fc2 found eight things:

• Flagship file: skills/taste-skill/SKILL.md is 87,253 bytes, about 21.8K tokens, with no references or scripts.
• Plugin scope: the manifest uses source ./, so all 13 SKILL.md files (about 302 KB) install together.
• Repo scripts: four Node .mjs files process README images with hardcoded Windows paths (issue #113); no skill calls them.
• Lookup script: skill.sh is a bash lookup table using declare -A, which needs bash 4+.
• Flagship behaviour: no hooks, MCP configs, env vars or network calls appear in the flagship skill.
• Install prompts: the skill tells the agent to check package.json and print install commands before importing libraries.
• Third-party audits: skills.sh reports Pass from Gen Agent Trust Hub, Socket and Snyk; Tessl reports a passed Snyk scan.
• Prompt injection: no handling is needed or present, because the skill reads user briefs and project files only.

Gen Agent Trust Hub, Socket and Snyk are security scanners that skills.sh runs on listed Agent Skills. Tessl is a registry that scores skill quality.

New since review (not graded): the v2 scripts need Node 22+. The inspector launches Chrome, Edge or Chromium over the Chrome DevTools Protocol. assets.mjs searches Openverse, Wikimedia Commons, the Art Institute of Chicago and the Met, without API keys. Inspector runs keep browser profiles in a temp folder, and issue #157 shows one run deleting another's profile.

Read how to check a skill's safety before you install the v2 scripts on a work machine.

What does the community report about Taste Skill?

The community reports useful structure from Taste Skill. Issue #86 praises its brief inference and pre-flight checklist. Issue #106 says agents still return generic output, and Tessl scored its quality 57.

These are the 17 dated reports we weighed, from GitHub, Hacker News, Tessl and the web. Six count as strong, and strong items come first.

The tastegate result measures a merged derivative, not Taste Skill alone. It counts as indirect support only. No source measured how often agents follow the checklist.

What are Taste Skill's known failures?

At commit ce26fc2, Taste Skill had four known failures. Two motion skeletons broke layout and hid content from reduced-motion users. Section 7 contradicted its own transition rule, and folder names broke plugin lookups.

The worst failure starts with a copied skeleton. The agent pastes a canonical GSAP pattern that hides content from reduced-motion users or breaks scroll layout. The page looks polished and ships inaccessible.

The fixes landed in the old line on 2026-10-11, before v2.0.0 replaced it. We have not checked whether the edited taste-skill-classics copy carries them.

A separate load failure arrived after our review. Version 2.1.0 shipped an unquoted YAML description, and Claude Code reported "Unknown skill" (#151). Version 2.1.2 fixed it. Our guide to fix a skill that will not load covers that error and its usual causes.

Does Taste Skill work in Codex, Cursor and other agents?

Yes, by the publisher's claim: Taste Skill names Codex and Cursor. We verified it only in Claude Code. The site tasteskill.dev names Gemini CLI, v0, Lovable and OpenCode, with no documented run outside Claude Code.

"Verified" means our evidence shows the skill loading and running in that agent. "Claimed" means the publisher says so and no source confirms it.

Gemini CLI is Google's terminal coding agent. The v0 and Lovable tools are hosted app builders that generate frontends from prompts. None has a documented Taste Skill run. Our list of top Claude Code skills by verdict covers the agent we verified.

How do you install Taste Skill?

You install Taste Skill with one command: npx skills add https://github.com/Leonxlnx/taste-skill --skill design-taste-frontend. In Claude Code, the Leonxlnx plugin marketplace works as well. Both paths now install the new v2, not the file we reviewed.

The skills CLI is the npx skills command from skills.sh, a public directory of Agent Skills. It copies a skill folder into the agent's skills directory. The Claude Code plugin marketplace is the /plugin system that installs packaged skills from a GitHub repository.

In Claude Code

Run one of the two install paths below: the skills CLI or the plugin marketplace.

npx skills add https://github.com/Leonxlnx/taste-skill --skill "design-taste-frontend"

/plugin marketplace add Leonxlnx/taste-skill
/plugin install taste-skill@taste-skill

The v2 scripts need Node 22+ and an installed Chrome, Edge or Chromium. Since v2.1.1, the plugin carries only design-taste-frontend. Our guide to installing skills in Claude Code covers skill folders, scopes and restarts.

In Codex, Cursor and other agents

Codex has two documented paths, and no published run confirms either one.

npx skills add Leonxlnx/taste-skill -a codex

codex plugin marketplace add Leonxlnx/taste-skill
codex plugin add taste-skill@taste-skill

Cursor users run the same npx skills add command, per the README. Treat every path outside Claude Code as untested.

How do you install the version we reviewed?

The reviewed line now lives in Leonxlnx/taste-skill-classics, under the install name design-taste-frontend-v2-experimental. The CHANGELOG for 2.1.3 gives the plugin route.

/plugin marketplace add Leonxlnx/taste-skill-classics

The skills CLI accepts the same repository with --skill design-taste-frontend-v2-experimental. That file has changed since our review. It is 89,426 bytes with sha256 2298b5a2…, against 87,253 bytes and aa194351… for commit ce26fc2. Our verdict therefore describes it closely, not exactly.

How do you use Taste Skill?

You use Taste Skill by asking for a landing page with a written brief. The current v2 deals three directions, you choose one, then it builds and checks the page in Chrome. Review any copied scroll animation.

The current v2 runs ten steps, according to its SKILL.md. The six below carry the work, and all of them are vendor-reported:

1. Inventory the client's own material first.
2. Run draw.mjs and pick one of three design directions.
3. Write the idea for the page.
4. Build the page with full, real content.
5. Gate the build with lint.mjs and inspect.mjs in a real browser.
6. Critique the screenshots, fix the top three problems, and repeat for up to three rounds.

The v2 study behind these directions covered 247 curated sites and produced 20 directions, per the publisher.

The reviewed file ran a shorter loop with no scripts:

1. State the design read in one line.
2. Set the three dials for the page.
3. Follow the routing map to a design system when the brief needs one.
4. Pass the pre-flight checklist before the page ships.

The skill targets three jobs. Landing pages with a non-default visual direction are the primary use, backed by Hacker News and GitHub reports. Personal and designer portfolios are a publisher claim only. Audit-first redesigns of an existing site are a publisher claim only.

Test it yourself before you adopt it:

1. Install only design-taste-frontend, not the whole bundle.
2. Give the agent the same landing-page brief with the skill and without it.
3. Compare the two pages for the named tells.
4. Turn on reduced motion and check that every section stays visible.

What are the alternatives to Taste Skill?

The closest alternatives to Taste Skill are Frontend Design and Impeccable. Frontend Design scores 76, Recommended, and runs no scripts at all. Impeccable scores 67, Conditional, and focuses on critique and polish.

Six design skills we have reviewed compare directly. The table is sorted by score, and each row names the use case that favours it.

Frontend Design is Anthropic's own design skill. Impeccable, by Paul Bakaus, adds critique, audit and polish commands. The ui-skills.com directory lists 27 skills tagged "taste" by other authors. Only Leonxlnx's is the skill on this page. Our full list of every design skill we rated ranks the rest.

Where is Taste Skill on GitHub?

Taste Skill is on GitHub in the Leonxlnx/taste-skill repository under skills/taste-skill, MIT-licensed. We reviewed commit ce26fc2 on 2026-09-30. The repo published its first release tags in October 2026, from v2.0.0 to v2.1.3.

• Repository: github.com/Leonxlnx/taste-skill, folder skills/taste-skill.- Licence: MIT.
• Stars: 91,572 at review and 99,798 on 2026-10-21.
• Forks: 6,224 at review.
• Contributors: 7 at review.
• Installs: skills.sh counted 543.6K installs of design-taste-frontend at review.
• Tags: none at review; v2.0.0 to v2.1.3 since 2026-10-13.
• CI: a frontmatter check since v2.1.2.
• Sister repositories: Leonxlnx/taste-skill-classics for the older skills and Leonxlnx/imagegen-skills for image skills.

Common questions

Is Taste Skill free?

Yes, Taste Skill is free and MIT-licensed. You pay only for the model tokens it uses.

Does Taste Skill work with Vue or Svelte?

No, not fully: the reviewed file hardcoded React and Next.js defaults, so Vue and Svelte users discarded parts. The v2 description names any stack, which is vendor-reported.

Does Taste Skill need Node, Chrome or network now?

Yes, the v2 scripts need Node 22+ and Chrome, Edge or Chromium. assets.mjs needs network access for image search.

Does Taste Skill work in ChatGPT?

No. No source names ChatGPT or GPT as a supported agent. The publisher claims Codex, OpenAI's coding agent.

Is tasteskill.dev the official Taste Skill site?

Yes, tasteskill.dev is the publisher's official site. Its homepage still says 13 skills, though the repo became one skill on 2026-10-19.