taste-skill: How to Validate Boring AI-Generated Landing Pages in a Real Browser

Leonxlnx/taste-skill

·2026.09.03 20:22

This agent skill prevents the 'generic AI design' commonly seen when coding agents build web pages. It goes beyond simple code generation by applying art direction and including a process to verify results in a real browser. It applies a 6-step build process developed from research on 247 curated sites.

1 / 6

It finds and places copyright-free real photos and videos from open archives such as Wikimedia Commons, Openverse, and the Met. It automatically credits all images and works without being tied to specific frameworks like React, Vue, Svelte, or plain HTML. Starting from v2.1, it also generates dashboards and app screens based on rules derived from analyzing real product screens.

It renders the completed page in a real browser at three different widths to check for 'slop'. It features a self-verification loop where the agent analyzes its own screenshots and fixes the top 3 issues up to 3 times. It also includes layout and motion rules applied from an analysis of the top 100 Awwwards Sites of the Day.

It is compatible with major coding agents such as Claude Code, Codex, and Cursor, and can be used without an API key if you have Node 22 or higher and a Chrome-based browser. It is suitable for developers who want to generate web pages or app screens with brand identity, at a cost of approximately 30–45 minutes and hundreds of thousands of tokens.

GitHub
GitHub repository

Leonxlnx/taste-skill

Agent skills that give coding agents taste: art direction, reference images and a browser check instead of generic AI pages

JavaScript

This introduction was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report errors, attribution issues, or removal requests via Contact.