build-landing-page

Use when building or restructuring a landing, marketing, or other public conversion page — checks it against a section grammar, one conversion goal, a semantic skeleton, and a hero LCP/CLS budget. Wor

SKILL.md

GitHub

Build Landing Page

A landing page has one job. This skill names it, audits which sections serve it, adds only what's missing, and holds the hero to a performance budget — because on a public page LCP and CLS are ranking and revenue, not polish.

1. Gate — is this a public page?

Run the gate from ../_shared/page-types.md: must this page be findable and readable without JS? If it's an authenticated app surface, stop — this skill doesn't apply; the skills/frontend/ catalogue owns app screens.

2. Audit current state

Work from the built/served HTML (a URL or a file from dist//build//out/):

curl -s "$URL" -o /tmp/page.html        # or cp the built file
grep -ci "<h1" /tmp/page.html                                  # exactly 1?
grep -oE "<(header|main|section|footer|nav)\b" /tmp/page.html | sort | uniq -c   # landmarks
grep -oiE "<(img|picture|video)[^>]*" /tmp/page.html | head -5                   # hero media + attrs

Read the page and list: which grammar sections exist, what the h1 claims, how many competing CTAs there are. Prerequisite check: if the main content isn't in the HTML (view-source test, ../_shared/page-types.md), fix crawlability first — ../set-up-seo/SKILL.md.

3. Name the one job

Every landing page converts toward exactly one action: sign up, buy, book, subscribe, download. Name it in one sentence ("this page exists to get a demo booked"). If nobody can, that is the first finding — no section work matters until the job is named. Every primary CTA on the page is that action; anything else is subordinate (footer, text links).

4. Decide what to do

  • No page yet → compose from the grammar (step 5), required sections first.
  • Page exists, sections missing for its goal → add only those.
  • All sections present → tighten against the grammar's must-haves (step 5) and the budget (step 7). Re-running on a conforming page changes nothing.

5. Section grammar

Walk top to bottom; details and anti-patterns in ./section-grammar.md.

Section Its job Must contain Skip when
Hero state value + action h1 = outcome (not product name); one primary CTA above the fold; LCP visual on budget never
Social proof borrow trust real logos / named quotes / specific numbers nothing real yet — omit, don't fake
Problem → solution mirror the pain the problem in the visitor's words, then the turn product self-evident (waitlist)
Features as benefits capability → outcome each item leads with what the visitor gets single-feature page
Pricing remove the price question real numbers or honest "from …"; one recommended tier no self-serve price
FAQ answer created objections 4–8 real questions, direct answers (visible text first; schema is optional, see ../set-up-seo/structured-data.md) nothing left open
Final CTA catch the convinced scroller one line of value + the same primary action; no new links single-screen page

6. Semantic skeleton

Landmarks and heading order are accessibility and machine-readability at once:

<header><!-- logo, minimal nav --></header>
<main>
  <section aria-labelledby="hero-h">
    <h1 id="hero-h">Outcome the visitor gets</h1>
    <p>One supporting line — the mechanism.</p>
    <a class="cta" href="#signup">The one action</a>
  </section>
  <section aria-labelledby="proof-h"><h2 id="proof-h">…</h2>…</section>
  <!-- further sections: h2 each, h3 inside, no skipped levels -->
</main>
<footer><!-- legal links: imprint/privacy — required, not optional chrome --></footer>

Exactly one h1. Every section gets a real heading — screen readers and answer engines navigate by them.

7. Hero performance budget

The hero owns LCP and most CLS risk. Core Web Vitals "good" thresholds, judged at the 75th percentile of real page loads (mobile and desktop separately): LCP ≤ 2.5 s, CLS ≤ 0.1, INP ≤ 200 ms (web.dev/articles/vitals, verified 2026-10-09). A public page with near-zero JS rarely fails INP; LCP and CLS are where it is won or lost.

<!-- the LCP image: dimensions reserve space (CLS), priority wins the network race -->
<img src="/hero.avif" alt="What the image shows" width="1200" height="800"
     fetchpriority="high" decoding="async">
<!-- a hero font: preload + swap so text is never invisible -->
<link rel="preload" href="/fonts/heading.woff2" as="font" type="font/woff2" crossorigin>
@font-face { font-family: Heading; src: url(/fonts/heading.woff2) format('woff2');
             font-display: swap; }
  • Modern format (AVIF/WebP), sized for its largest rendered width — not the original. Bare AVIF accepts a small no-support tail; wrap in <picture> with WebP/JPEG <source> fallbacks when the audience skews to older browsers.
  • width/height (or CSS aspect-ratio) on every image so nothing shifts.
  • No carousel, no autoplay video, no JS-gated rendering in the hero — each delays the LCP element or shifts layout. Scripts use defer (or type="module") and sit below the fold; a public page's JS budget starts at zero (../_shared/page-types.md).

8. Verify

curl -s "$URL" | grep -ci "<h1"          # 1
curl -s "$URL" | grep -ci "fetchpriority"   # ≥1 (the hero image)

View the page with JS disabled: full content, readable order. Run Lighthouse (mobile profile) on the deployed URL: LCP ≤ 2.5 s, CLS ≤ 0.1. Lighthouse is a lab run on one device; the pass/fail Google uses is field data (CrUX, or RUM via configure-analytics), which a new page only has after about four weeks of traffic.

References

  • ./section-grammar.md — per-section rules, anti-examples, when to deviate.
  • ./stack-pointers.md — where hero/meta wiring lives in Next, Astro, Nuxt, Vite SPA.
  • ../set-up-seo/SKILL.md — crawlability, metadata, structured data for this page.
  • ../set-up-lead-capture/SKILL.md — when the one action is a form.
  • ../../frontend/optimize-performance/SKILL.md — app-side CWV work (the other side of the priority inversion).