Image: GitHub
Why it mattersA team using coding agents every day keeps getting the same generic page. An installed skill with a built-in design check is one way to stop that.
Coding agents produce competent code and dull pages. Many teams have noticed. One of the most-starred open-source answers to that problem is Taste Skill, by developer Leonxlnx. The GitHub API reported 104,673 stars for Leonxlnx/taste-skill on 29 October, with 7,078 forks. The repository was created on 19 February 2026.
Trendshift ranked the repository 12th on its daily list on 29 October. Taste Skill was the author's second repository to reach the AI News, after Earshot in October and Unlazy in September.
What the skill does
The README calls Taste Skill an Agent Skill. It pushes coding agents to design websites and app screens. The alternative, the README says, is assembling the average landing page. The skill installs with npx skills add or as a plugin inside Claude Code, Codex and Cursor. It uses no API keys. It needs Node 22 or newer and an installed Chrome, Edge or Chromium. It uses the network to find pictures.
The README says v2 is built from a study of 247 curated sites. A build runs in six steps. The skill reads the brief and lists what only this client has. A script deals three of 20 researched directions. The agent picks one. The agent then writes one sentence tying the client's own material to that direction. The build uses real pictures and footage from open collections, credited on the page. The skill checks the page in a real browser at three widths and fails it on slop. It looks at its own screenshots and fixes the top three problems. It will try up to three rounds.
The README says a build takes roughly 30 to 45 minutes and a few hundred thousand tokens.
What v2.1 added
Dashboards and app screens have their own route. The skill tells a page from an app screen, the README says. For app screens, it deals one of the styles distilled from real product screens. It builds the shell, the data, the charts and the tables to those numbers. It checks the result with app rules: quiet panels, aligned figures, colour spent on data, a phone layout where the navigation folds away. A product's landing page builds the app shown in its screenshots the same way. The README says this makes the screenshots look like a working product.
v2.1 also checks the frame of every page. The README lists four rules: a clear title on the first screen, equal margins, no section pushed to one side, and no random blank stretches. Movement rules were measured on the latest 100 Awwwards Sites of the Day, the README says.
What leaves the project
Pictures come from open collections such as Wikimedia Commons, Openverse and the Met, the README says. Only public domain, CC0, CC BY and CC BY-SA files are offered. Credits are placed on the page. A client is expected to replace the pictures with its own.
The repository is MIT licensed. The author lists sponsors including Vercel's Open Source Program and Kimi. Older skills (high-end-visual-design, minimalist-ui, industrial-brutalist-ui, gpt-taste, stitch-design-taste) moved to a separate repository, taste-skill-classics.
Why it has traction
Taste Skill does not promise a model that writes better code. It names one failure of coding agents: the generic landing page. It gives the agent a checklist and a real browser visit to catch that failure. Some teams write code with Claude Code or Codex every day. For them, the installed skill is a cheap step to add. The star count is one signal that other developers hit the same problem and found this answer. The look of the output is the one a reader has to judge.
Source
- Leonxlnx/taste-skill on GitHub, the repository README
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
