[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$f3dxh0kyyv5myu":3},{"code":4,"message":5,"data":6,"timestamp":38},10000,"success",{"full_name":7,"position":8,"description":9,"language":10,"stars":11,"forks":12,"stars_gained":13,"age_days":14,"hn_threads":15,"one_liner":16,"summary_md":17,"topic":18,"github_topics":21,"license":34,"homepage":35,"updated_at":36,"history":37},"juliobsz\u002Fskills",10,"An agent skill that builds award-caliber websites from one line","JavaScript",938,37,71,13,[],"An agent skill that turns a one-line brief into a scoring, self-reviewing website build for Claude Code and Codex users.","If you already pay for a capable coding agent and want a portfolio or launch page with a visual point of view, this skill is worth testing on one real project. It trades the blank-page problem for an opinionated build loop that reviews its own output.\n\n## The problem\n\nGetting a good-looking marketing site out of a coding agent today usually means several rounds of the same loop: ask for a hero, get generic gradients, ask for motion, get something that breaks on mobile. The user's goal is a finished site with a coherent identity, but the friction is art direction — the agent defaults to what an average model produces. This skill changes that step: it forces the agent to write a concept, art direction and scoring rubric before writing markup, then screenshot the result and fix it. The user ends up with a built, audited site rather than a single HTML draft.\n\n## What's different\n\nMost of the value is in the process, not a template. The README describes six phases: it infers the brief and records assumptions instead of asking a questionnaire, draws three concepts from different territories and picks one, then writes palette, type pairing and motion language into `DESIGN.md`. A \"genericness test\" is meant to kill the average-model look before the build. After building, a QA loop runs a production build, a crawling audit for SEO, metadata, accessibility, weight and placeholder copy, plus real-browser screenshots on desktop, mobile, reduced motion and mid-transition. It measures LCP and CLS on a throttled phone and runs axe. A separate critic subagent that never saw the plan scores the site against the rubric, and the agent iterates until every category reaches 8\u002F10.\n\nThe trade-off is that this is a long, multi-step run with real browser calls and helper scripts, not a quick generation. It also imposes taste: three concepts, a curated free-font library and a target score decide what you get. If you have a specific brand system or an existing design language, the skill's art direction may fight it rather than help. And the \"award-caliber\" framing is the author's claim; the shipped site is judged by a skill-authored rubric, not by an external jury.\n\n## Who should try it\n\nA good fit is a developer or designer who can already run an agent and wants a convincing personal portfolio, studio landing page or product launch page where the visual direction matters and they are comfortable steering the concept. The README's own examples — a Lisbon architecture studio, a mechanical keyboard launch — show the intended small, story-driven site. A poor fit is someone who needs a page inside an established design system, or who wants a quick throwaway draft: the install and QA loop add overhead that a one-off page does not justify. The optional `\u002Fjury` skill also scores any existing site against the same rubric, so designers evaluating a competitor can use that half without building.\n\n## Requirements and limitations\n\nNode 22.20+ is required for the skills CLI, while helper scripts need Node 20+. Playwright and axe-core must be present in the site's project, though the skill installs them as dev dependencies when needed; `\u002Fjury` installs them once into `~\u002F.cache\u002Fjury` without touching the reviewed project. A model that can read images is required so the QA loop can inspect its own screenshots — a text-only model breaks the core feedback loop.\n\nOngoing upkeep matters here: the skill must be updated with `npx skills update -g`, agents only pick up skills on a new session, and installing via both the skills CLI and the Claude Code plugin results in duplicates. The README also states that skills are instructions your agent follows and these run builds and scripts on your machine, so read `SKILL.md` before installing. A Vercel, Netlify or Cloudflare account is only needed for deployment, and nothing deploys without confirmation. See the [install and troubleshooting notes](https:\u002F\u002Fgithub.com\u002Fjuliobsz\u002Fskills\u002Fblob\u002Fmain\u002FREADME.md) for the exact paths.\n\n## HotRepo's take\n\nHotRepo only has two recorded dates, 2026-10-24 to 2026-10-25, and one observation: #10 on the NEW-REPOSITORY board for javascript at 935 total stars on 2026-10-25 — not a daily growth figure. That sample is too thin to judge whether interest persists, so treat the star count as a snapshot of a 13-day-old repo rather than evidence of momentum. Before adopting, verify that your agent and model actually satisfy the image-reading and Playwright requirements, since that is where the QA loop either works or quietly degrades.",{"slug":19,"name":20},"developer-tools","Developer Tools",[22,23,24,25,26,27,28,29,30,31,32,33],"agent-skills","astro","awwwards","claude-code","claude-skills","codex","creative-coding","frontend","gsap","nextjs","web-design","webgl","MIT",null,"2026-10-25T02:49:04Z",[],1792915197]