← All open-source news
Open sourceProject

Taste Skill, a free agent skill that gives Claude Code and Codex an art direction and a browser check, gains about 6,000 GitHub stars in a week

· 1 min read

Leon Lin's MIT-licensed Taste Skill makes coding agents plan a page's look, use real credited photos and check the result in a browser before they finish. It has about 112,000 GitHub stars and gained about 6,000 in a week.

Taste Skill

Taste Skill is a free, MIT-licensed agent skill by Leon Lin (GitHub: Leonxlnx) for AI coding agents such as Claude Code and Codex. Its aim is to stop them building the same generic, template-looking pages. The README describes it as giving agents "art direction, reference images and a browser check instead of generic AI pages". The repository has about 112,000 stars and gained about 6,000 in a week on GitHub's trending list.

What it does: before writing markup, the skill classifies the business (software, gaming, tech, product, service or place) and plans the page. It deals a colour palette, a font pairing and one of eleven hero layouts, and uses real photographs from a vetted library plus real brand and interface icons. Shaders and scroll effects are kept to software, tech and gaming pages. When the page is built, it is checked in headless Chrome at several screen widths, with up to three rounds of fixes. explainx.ai says the website workflow picks from 20 researched art directions, and that photos come from Wikimedia Commons, Openverse and the Met under public-domain or Creative Commons licences, credited on the page.

How to try it: with the skills CLI, npx skills add https://github.com/Leonxlnx/taste-skill --skill "design-taste-frontend". In Claude Code: /plugin marketplace add Leonxlnx/taste-skill, then /plugin install taste-skill@taste-skill. In Codex: codex plugin marketplace add Leonxlnx/taste-skill, then codex plugin add taste-skill@taste-skill. It needs Node 22 or newer and Chrome, Edge or Chromium; no API keys are required.

Limits: the README names the current version as tasteskill 2.5.1, and the main repository now holds a single skill, design-taste-frontend; older styles moved to separate repositories. Extra photos, icons and Google Fonts need network access. explainx.ai, citing the README, says a build takes roughly 30 to 45 minutes and a few hundred thousand tokens. Cursor is listed in the repository's topics, but there are no Cursor install steps. Taste Skill has no Hacker News discussion of note so far.

More news

llama-modchip
Project

llama-modchip, a free llama.cpp fork, says it runs DeepSeek's 284B-parameter model on a PC with 64GB of RAM and no GPU; 402 points on Hacker News

The open-source project squeezes DeepSeek-V4-Flash into a 56.56 GiB file and reports about 9–15 tokens per second on an AMD Ryzen 9 laptop chip. The figures are the project's own, it needs its patched runtime, and memory headroom is tight.

OpenMontage
Project

OpenMontage, a free tool that turns Claude Code, Cursor or Codex into a video studio, gains about 4,400 GitHub stars in a week

OpenMontage lets an AI coding assistant research, script, gather footage, edit and render a whole video from a plain-language brief. It runs with no API keys using free tools, is AGPL-3.0 licensed and has about 78,600 GitHub stars, but has no tagged releases yet.

Humanizer
Project

Humanizer, a free agent skill that makes Claude Code or Codex rewrite text so it stops sounding like AI, gains about 3,000 GitHub stars in a week

Humanizer, an MIT-licensed skill by developer blader, checks text for 26 patterns from Wikipedia's "Signs of AI writing" guide and rewrites them without adding facts. It works in Claude Code, Codex and other agents, has about 66,300 GitHub stars, and is back on GitHub trending.