Primary question: Does your current AI-generated UI look the same on every project, and do you want a drop-in skill file to enforce better design taste?
RepoDaily adoption score
RepoDaily rates this as 89/100 (strong) for adoption: evidence, installation path, production risk, differentiation, license clarity, and AI/agent fit are scored from the article sources and adoption notes.
4 source(s) across 3 source category/categories, plus a RepoDaily-specific evidence module when available.
6 workflow step(s), 5 next-action step(s), and 1 command/install signal(s) were detected.
Trending momentum is +850 stars, with maintenance/release/issue signals counted when present.
Risk is marked medium, with 4 security note(s) and 3 explicit skip condition(s).
3 opportunity lens item(s), 4 alternative(s), and 3 type-specific section(s) support differentiation.
License source or license wording is present.
8 AI/agent-related signal(s) were detected in the article text and metadata.
Project overview
Taste Skill, hosted at Leonxlnx/taste-skill, calls itself the Anti-Slop Frontend Framework for AI Agents. The repository packages portable Agent Skills that upgrade AI-built interfaces by enforcing stronger layout, typography, motion, and spacing instead of the boilerplate-looking UIs that large language models tend to produce by default. It also ships image-generation skills for reference boards spanning web, mobile, and brand kits, which you pair with tools like ChatGPT Images and then hand the resulting frames to Codex, Cursor, or Claude Code for implementation.
The default `taste-skill` (install name `design-taste-frontend`) is now v2 (experimental), a substantial rewrite of the original v1. v1 is preserved as `taste-skill-v1` with install name `design-taste-frontend-v1`, so existing users can re-run the install command to upgrade without losing their old setup. v2 keeps the dial-driven philosophy from v1 through three named dials, `DESIGN_VARIANCE`, `MOTION_INTENSITY`, and `VISUAL_DENSITY`, and layers on structure, hard rules, and concrete implementation patterns that the agent can actually follow.
The project is MIT-licensed, sponsored by Emil Kowalski of animations.dev and by the Vercel Open Source Program, and compatible with the `vercel-labs/agent-skills` CLI. That CLI scans the `skills/` folder in the repository, so every skill, whether code or image-generation, installs the same way through `npx skills add https://github.com/Leonxlnx/taste-skill`. The maintainer is explicit that Taste Skill has no official token, coin, or crypto project, and that any token using his name, image, or project is unaffiliated.
Why it is trending now
- 850 period stars and a trending rank of 6 on 2026-07-06, driven by the v2 experimental rewrite becoming the default install.
- v2 introduces 14 numbered sections including Brief Inference, Design System Map, Dark Mode Protocol, Redesign Protocol, the Block Library contract, and a Final Pre-Flight Check.
- Section 9 ("AI Tells") hardens bans on the most common stylistic mistakes in AI-generated marketing pages, including a complete em-dash ban and bans on version footers, scroll cues, and fake product UI.
- One-command install via the vercel-labs/agent-skills CLI works with Codex, Cursor, and Claude Code, lowering the barrier to try it.
- Sponsorship from Emil Kowalski (animations.dev) and the Vercel Open Source Program signals credibility in the frontend tooling community.
Problem it solves
- AI coding agents produce frontends that look generic, over-symmetric, and full of the same boilerplate UI patterns on every project.
- Marketing pages generated by LLMs frequently include decorative noise: scroll cues, fake version footers, locale strips, decorative status dots, and div-based fake product UI.
- Design briefs often name a system like Material, Fluent, Carbon, Polaris, Atlassian, Primer, GOV.UK, USWDS, Bootstrap, Radix, shadcn, or Tailwind, but the agent has no rule for when to reach for the official package versus when to use web standards.
- Redesign work is risky because agents silently change URL structure, navigation labels, form field names, brand wordmarks, and legal copy.
- Dark mode is usually bolted on late, with no token strategy or contrast parity enforced from the start.
How it works
- Run `npx skills add https://github.com/Leonxlnx/taste-skill` to install all skills, or add `--skill "design-taste-frontend"` to install only the default skill.
- The CLI, from vercel-labs/agent-skills, scans the `skills/` folder in this repo and installs the skill files into your project.
- Alternatively, copy any `SKILL.md` into your project or paste it into a ChatGPT or Codex conversation.
- When you prompt your agent (Codex, Cursor, or Claude Code) on a frontend task, the skill's sections run first. Section 0, Brief Inference, makes the agent read the page kind, vibe words, references, audience, and constraints, then declare a one-line design read.
- Section 2 maps the brief to a design system. If the brief reads as Material, Fluent, Carbon, Polaris, Atlassian, Primer, GOV.UK, USWDS, Bootstrap, Radix, shadcn, or Tailwind, the agent reaches for the official package. If the brief is an aesthetic such as glassmorphism, bento, brutalism, editorial, dark tech, aurora, or kinetic typography, the agent uses web standards and labels the implementation honestly.
- The three dials (`DESIGN_VARIANCE`, `MOTION_INTENSITY`, `VISUAL_DENSITY`) tune output intensity. Section 9 enforces the AI Tells bans. Section 14, the Final Pre-Flight Check, runs as a hard checklist before shipping.
Product demo and interface preview
What v2 actually adds over v1
v2 (experimental) is the new default for `taste-skill` (install name `design-taste-frontend`). It keeps the dial-driven philosophy from v1 but adds fourteen numbered sections. New sections include Section 0 (Brief Inference), Section 2 (Brief to Design System Map), Section 8 (Dark Mode Protocol), Section 11 (Redesign Protocol with mode detection for Greenfield, Preserve, or Overhaul), Section 12 (the Block Library contract covering hero, feature, social-proof, pricing, cta, footer, portfolio, transition, and navigation blocks), Section 13 (Out of Scope), and Section 14 (Final Pre-Flight Check).
The changelog calls v2 a pre-release. It is the new default install because, per the maintainer, it is genuinely better than v1, but refinements may land in any v2 experimental release. The install name, dial names, and section structure will stabilize at v2.0.0 stable. v1 is preserved as `taste-skill-v1` (install name `design-taste-frontend-v1`).
How to try it in under 15 minutes
- Install all skills: `npx skills add https://github.com/Leonxlnx/taste-skill`.
- Install a single skill by its install name (the `name:` field inside the SKILL frontmatter, not the folder name): `npx skills add https://github.com/Leonxlnx/taste-skill --skill "design-taste-frontend"`.
- If you have v1 installed already, re-run the install command to upgrade to v2.
- If you do not want the CLI, copy any `SKILL.md` file directly into your project, or paste it into a ChatGPT or Codex conversation.
- Open Codex, Cursor, or Claude Code and prompt it on a real frontend task. Compare the output against the same prompt without the skill installed.
- Check the generated page against Section 9 (AI Tells) and Section 14 (Final Pre-Flight Check) to see which bans and checks fired.
Maintenance and stability risk
v2 is explicitly experimental. The changelog states that refinements may land in any v2 experimental release and that the API (install name, dial names, section structure) will stabilize at v2.0.0 stable. That means section numbering, ban lists, and protocol details may shift between experimental releases.
The project is maintained by Leonxlnx with contact channels for feedback: GitHub Pull Requests or Issues, X direct messages to @lexnlin or @blueemi99, and email at hello@tasteskill.dev. The repository follows SemVer-ish discipline: experimental pre-releases iterate freely, and stable releases lock the API.
Who should pay attention?
Good fit if
- You use Codex, Cursor, or Claude Code for frontend work and your output looks the same on every project.
- You produce marketing pages, hero sections, pricing pages, or portfolio sites where layout, typography, motion, and spacing are visible selling points.
- You want a drop-in skill file instead of a build tool, runtime library, or design system package.
- You need dark mode to be enforced with token strategy and contrast parity from the first prompt, not as a late retrofit.
Skip for now if
- You build dashboards, data tables, multi-step forms, code editors, native mobile apps, or realtime collaboration UIs. Section 13 lists these as out of scope.
- You need a stable, locked API. v2 is experimental and iterating until v2.0.0 stable.
- You do not use an AI coding agent. Taste Skill is a skill file, not a standalone library or SaaS product.
Risks and cautions
v2 is explicitly experimental and may change before v2.0.0 stable, but the one-command install, MIT license, and v1 fallback keep the practical risk low.
- v2 is a pre-release. The changelog states refinements may land in any v2 experimental release.
- Section numbering, ban lists, and protocol details may shift between experimental releases until v2.0.0 stable locks the API.
- Section 13 explicitly excludes dashboards, data tables, multi-step forms, code editors, native mobile, and realtime collaboration UIs.
- The maintainer is an individual, not a funded company, which means bus-factor risk even with sponsor support from animations.dev and Vercel.
- MIT License, Copyright (c) 2026 Leonxlnx.
- Skills are markdown files (SKILL.md) installed into your project, not runtime dependencies or services that execute code on install.
- The CLI used to install skills is `npx skills add` from the vercel-labs/agent-skills repository, not a private binary.
- The maintainer publicly disclaims any official token, coin, or crypto project, and warns that any token using his name, image, or project is unaffiliated and not endorsed.
Alternatives to compare
| Approach | When to use | Trade-off |
|---|---|---|
shadcn/ui | You want a concrete React component library built on Radix primitives and Tailwind, rather than a skill file that teaches your agent how to design. | Free, MIT-licensed. |
vercel-labs/agent-skills | You want the underlying CLI and skill format that Taste Skill builds on, without Taste Skill's design-specific rules and bans. | Free. |
| You want a utility-first CSS framework that Section 2 of Taste Skill can direct your agent to use when the brief reads as Tailwind. | Free, MIT-licensed. | |
Hand-written design system docs | You already have an in-house design system and want full control over every rule, at the cost of writing and maintaining it yourself. | Free, but requires ongoing writing effort. |
What this trend reveals
Ship a v2.0.0 stable readiness checklist
The changelog promises API stability at v2.0.0 stable but does not publish a public checklist of what must be true before that release. A readable checklist would reduce upgrade anxiety for teams considering v2 in production.
Open a GitHub Issue on Leonxlnx/taste-skill asking for a v2.0.0 stable exit criteria document. If the maintainer confirms one exists or is planned, that becomes the basis for adoption planning.
Add dashboard and data-table coverage
Section 13 explicitly lists dashboards, data tables, multi-step forms, code editors, native mobile, and realtime collaboration UIs as out of scope. These are high-volume use cases for AI coding agents.
Search the repository Issues for dashboard or data-table skill requests. If none exist, file one and gauge maintainer interest before building a companion skill.
Publish before-and-after comparison galleries
The repository ships image-generation skills for reference boards and lists Floria Top and Floria Bottom as README examples, but there is no published side-by-side gallery of v1 versus v2 output on the same prompt.
Reproduce three real prompts with v1 (`design-taste-frontend-v1`) and v2 (`design-taste-frontend`) installed, screenshot the results, and submit a Pull Request adding them to the examples folder.
RepoDaily verdict
Taste Skill is one of the few open-source projects that directly targets the visual sameness problem in AI-generated frontends. v2 is experimental and explicitly out of scope for dashboards and data-heavy UIs, but for marketing pages, hero sections, portfolios, and brand sites, it is a credible drop-in skill file that is worth the 15 minutes it takes to evaluate.