agentleFS
carlos2219carlos2219/web-kit/skills/new-site/SKILL.md

new-site

Skill1 starsChanged 53 days agoMITView on GitHub

Take the user from a website idea to a published site, or resume one in progress, with the web-kit pipeline. The pipeline asks for information and feedback progressively and adapts to the budget, from completely free to low-cost upgrades like a custom domain or generated animations. Use it when the user wants to start a new site ("nuevo sitio", "tengo una idea para una web", "landing para X", "arranquemos el sitio de…"), asks "what's next" or "¿qué sigue?" in a web-kit site, or runs /web-kit:new-site. This skill is the plan for building the site, so don't run a separate brainstorming or plan skill for it.

The file

We split this skill into the rules it gives an agent

10 rules. About 0 are common, 6 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Keep it simple: No jargon. When a term is unavoidable, explain it in half a sentence: "dominio (la dirección tipo tunegocio.com)".

    line 14
  2. Distinctive:

    Show, don't describe: When asking for design feedback, show screenshots (and the reference next to them).

    line 15
  3. Distinctive:

    Money: read options.md. Never sign up, buy, or spend credits without an explicit yes.

    line 16
  4. Distinctive:

    Every phase ends in a file, so check which ones exist to know where you are.

    line 20
  5. Distinctive:

    If the user hasn't described the idea, ask for it in 1–2 sentences.

    line 36
  6. Distinctive:

    Local business?: If they serve a specific area or have a place, ask for the city or area, the phone or WhatsApp, and the hours.

    line 45
  7. Very specific:

    Run web-kit:design-system.

    line 53
  8. Very specific:

    Copy: every section gets a Clave | <LANG> | … table with the final copy in every language, plus title/description per page.

    line 61
  9. Very specific:

    Run web-kit:assets for each asset: slot.

    line 80
  10. Very specific:

    Run a final web-kit:design-loop over every page, at desktop and mobile width, in every language.

    line 85

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

New site pipeline

The goal is a site that looks designed by a studio, not by AI, and that works properly on any budget. The user may not be a web developer, so you guide them: you propose, they choose or correct. You ask for more as each decision becomes relevant, never all up front.

How to talk with the user

  • Propose, don't interrogate. Draft from what you know, then ask "¿es así o lo cambio?". Correcting a draft is easier than answering a blank question.
  • Use AskUserQuestion with 2–4 options, and put your recommendation first, marked "(Recomendado)". Ask 1–3 questions per turn, and only the ones the next step needs. The user can always pick "Other" to write their own answer.
  • Keep it simple. No jargon. When a term is unavoidable, explain it in half a sentence: "dominio (la dirección tipo tunegocio.com)".
  • Show, don't describe. When asking for design feedback, show screenshots (and the reference next to them). Turn vague feedback ("no me convence") into concrete options: more sober or bolder, the colors, the typography, the spacing, the image.
  • Money: read options.md. Never sign up, buy, or spend credits without an explicit yes.

Resume or start

Every phase ends in a file, so check which ones exist to know where you are. Tell the user in one line which phase you're in and what comes next.

PhaseDone when
1 Idea + scaffoldsrc/content/site.js exists and README.md has no web-kit:placeholder
2 Design directionDESIGN.md exists and src/index.css has no placeholder
3 Contentdocs/site.md covers every page with copy in every language, and the user approved it
4 PilotHome hero, nav and footer pass web-kit:design-loop, and the user approved them
5 PagesEvery route in docs/site.md is implemented, and the user saw each page
6 VisualsEvery asset: slot in docs/site.md is filled and logged in docs/assets.md
7 Publishnode <kit>/scripts/check.mjs launch passes and the site is live

<kit> is the web-kit root, two levels above this skill's base directory (<base>/../..). Commit at the end of each phase.

Phase 1: From idea to brief

  1. If the user hasn't described the idea, ask for it in 1–2 sentences. That's all you need to begin.
  2. Round 1 (AskUserQuestion, with options inferred from the idea):
    • Main goal: get contacts or quotes, sell or book, show a portfolio, inform (event, local business).
    • The one action the site asks for: WhatsApp, form, schedule a call, buy, or visit/call.
    • Languages.
    • Budget: "Gratis total" / "Bajo: dominio propio (~10–15 USD/año)" / "Medio: dominio + imágenes y animaciones generadas" / "Decido después". Recommend "Gratis total" to start, since upgrading later is easy.
  3. Draft the brief in plain Spanish (not the README yet). Cover who it's for, their problem, why this offer and not the alternative, what the site asks them to do, and a proposed tone. Also propose a name if they don't have one. Ask "¿qué corregirías?". Iterate until they say it's right.
  4. Round 2, only what's still missing:
    • Real facts you must not invent: services, experience, contact details.
    • Local business? If they serve a specific area or have a place, ask for the city or area, the phone or WhatsApp, and the hours. These feed the site and the Google Business Profile at publish time, and must match exactly.
    • Own material? Photos or video of their work, team, or place. If they have any, ask them to put the originals in media-originals/. That material usually beats any generated image.
    • Anything that must never appear: prices, client names, claims without evidence (certifications or permits they don't have yet). Private material goes in privado/, and nothing from it ever reaches the site.
  5. Scaffold: node <kit>/scripts/new-site.mjs <dir> --name "<Brand>" --langs es,en (default <dir>: a sibling of the current directory, named after the brand slug). Work inside <dir> from here on.
  6. Fill in README.md from the approved brief: every placeholder, including §9 with the chosen options from options.md. Turn tone into 3–5 rules, each with a do/don't example. Put brand-specific banned terms in docs/copy-flags.json. If the languages aren't es,en, the copy check lists what to add or drop in site.js.

Phase 2: Design direction

Run web-kit:design-system. It shows the user 3 visual directions to choose from, then builds the system from the one they pick.

Phase 3: Content

Write docs/site.md:

  • Site map: one idea per page. Keep it small. A one-page site is a valid answer for a basic site. Five focused pages beat ten thin ones. Pages without real content yet go under "Futuro".
  • Structure: global elements, then each page split into sections, in the order pain, offer, proof, how it works, call to action.
  • Copy: every section gets a Clave | <LANG> | … table with the final copy in every language, plus title/description per page. Follow web-kit:site-copy.
  • Visual slots: mark each one asset: <what it shows>. With a free budget, plan the visuals in CSS/SVG or around the user's own photos.

Show the user the site map and each page's headline, in their language. Ask for corrections. Proceed once they approve. From here on, edits go into the spec first, then into site.js.

Phase 4: Pilot

Build only the Layout (nav, footer, and a global call-to-action block if the spec has one) and the Home hero. Run web-kit:design-loop until they pass. Show the user desktop and mobile screenshots next to the reference. Ask for feedback with concrete options.

Everything else inherits the pilot's decisions. Iterate here until they like it.

Phase 5: Pages

For each page: add the route in src/router.js and App.jsx (and in NAV in Layout.jsx if it belongs in the navigation), add its copy to site.js, build it section by section by reusing the pilot's components, and run one web-kit:design-loop pass.

After each page, show the user a screenshot and ask one question: "¿Algo que cambiar en esta página?" (options: "Está bien, sigue" (Recomendado) / "Cambiar textos" / "Cambiar diseño"). With 4 or more pages, you may build them in parallel with subagents, one page per agent. Give each agent this Phase 5, DESIGN.md, and the pilot components. Then show all the pages together.

Phase 6: Visuals

Run web-kit:assets for each asset: slot. It follows the budget in README §9.

Phase 7: Publish

  1. node <kit>/scripts/check.mjs launch, then npm run check. Fix everything they report.
  2. Run a final web-kit:design-loop over every page, at desktop and mobile width, in every language.
  3. Run web-kit:publish.

Finish with web-kit:retro, so this site makes the next one better. Later, when the user wants to improve the site (a domain, a form, animations, a new page), re-enter the phase that matches the request.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.