marketing-site-builder
Build company, product and landing sites that do not look generic: one system, plain copy, real imagery, restrained motion, a static finish gate
Install / Use
npx skills add borghei/Claude-Skills --skill marketing-site-builderInstalls into whichever agent you are using.
SKILL.md
Installable skill definition
Quality Score
Category
Content & MediaSupported Platforms
Our assessment of marketing-site-builder
marketing-site-builder scores 91/100 on our quality scale, 377th of 1,155 Content & Media skills we index (top 33%).
Its SKILL.md is 20 KB long, well organised into 29 sections with 3 code examples: a thorough specification that gives an agent plenty to work with.
It has 817 GitHub stars, a meaningful sign that others use it.
Maintenance, license and trust
- The repository was last updated about 2 months ago, so marketing-site-builder is actively maintained.
- No license is declared. By default that means all rights are reserved: you can read it, but reusing or redistributing it is not clearly permitted. Ask the author before building on it commercially.
- Its trust signals score 88/100, with 1 caution from licensing, adoption, age or documentation. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.
Safety scan
No issues foundOur scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review of the same text found nothing harmful.
AI review by kimi-k2.7-code on 2026-11-10. Automated pattern scan on 2026-11-10. It catches known dangerous patterns, not every risk — read a skill before letting an agent act on it.
marketing-site-builder compared with similar skills
All 4 of these similar skills score higher than marketing-site-builder; compare them before choosing.
| Skill | Score | Stars | Updated | Format |
|---|---|---|---|---|
| marketing-site-builder (this skill)by borghei | 91 | 817 | 56d ago | SKILL.md |
| taste-skillby Leonxlnx | 100 | 116.8k | 8d ago | CLAUDE.md |
| Claude-Code-Game-Studiosby Donchitos | 100 | 39.7k | 7d ago | CLAUDE.md |
| LocalAIby mudler | 100 | 50.4k | today | MCP Server |
| siyuanby siyuan-note | 100 | 47.3k | today | MCP Server |
Frequently asked questions
- How do I install marketing-site-builder?
- Run
npx skills add borghei/Claude-Skills --skill marketing-site-builder. The install tabs above show the steps for each supported agent. - Which AI agents does marketing-site-builder work with?
- It is written for Universal, as a SKILL.md file. Other agents that read the same format can often use it too.
- Is marketing-site-builder safe to use?
- Our scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review of the same text found nothing harmful. It declares no license and scores 88/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
- Is marketing-site-builder still maintained?
- The repository was last updated about 2 months ago, so marketing-site-builder is actively maintained.
Skill content
View source on GitHubname: marketing-site-builder description: > Build company, product and landing sites that do not look generic: one system, plain copy, real imagery, restrained motion, a static finish gate. Use when building, restyling or reviewing a marketing site, or a page reads as templated. license: MIT + Commons Clause metadata: version: 1.0.0 author: borghei category: engineering domain: frontend updated: 2026-11-08 tags: [frontend, web-design, landing-page, copywriting, design-system, performance, motion]
Marketing Site Builder
Build the website a good company would ship: it says plainly what is sold, shows the real thing on every screen, and looks made for this client rather than assembled from a template. The method is fixed in order. Decide the kind of business first, because it decides what the page may show and how much may move. Write the copy from facts before any layout. Plan the page as a list of screens. Build on one system of tokens. Then run the finish gate (six static checkers behind one verdict line, with written waivers and a three-round limit) and a short browser checklist, fix, and report what is still open.
Most generic pages fail in the same three places: copy that would fit any business, a stylesheet with a dozen ad-hoc spacings, radii and colours, and decoration (gradients, glass, scroll reveals) standing where the product should be. The agent catches those from source with the gate, the rest on screenshots.
When to Use
- Building a marketing, company or product site as static HTML/CSS or inside an existing React, Next.js, Vue, Svelte or Astro project
- Restyling or reviewing a page that someone says looks generic, templated or "AI-made"
- Writing or cleaning up website copy before it goes into markup
- Adding a finish gate to a site build so copy, consistency and weight are checked before handover
- Deciding how much motion and effect a page may carry for its kind of business
Clarify First
What most shapes the page is the kind of business and the real material on hand. Settle the inputs below before producing anything. When one is missing or vague and would change the result, ask about it; when it wouldn't, use a sensible default and say which one you used.
- The kind and what is sold: tech, product, service or place decides the effect budget, the photograph count and the hero; a wrong kind puts a shader on a bakery.
- The main action: the one transaction (Book a room, Start a free trial) sets the button wording in the nav, hero and close; a page with three equal actions has none.
- The real material: client photographs, screenshots, prices, address and hours decide which sections can exist; nothing missing is invented.
- The stack: static files or an existing framework decides where tokens live and which built CSS the consistency check reads.
Ask about the two or three inputs that matter most for this request, not the whole list. If the user wants a result straight away, produce it and state your assumptions at the top.
Quick Start
# One verdict over all six checkers: the sound sample ends on GATE: PASS (exit 0)
python3 engineering/marketing-site-builder/scripts/finish_gate.py engineering/marketing-site-builder/assets/sound_page.html
# The deliberately flawed sample fails every checker (exit 1); run one alone for its full list
python3 engineering/marketing-site-builder/scripts/finish_gate.py engineering/marketing-site-builder/assets/flawed_page.html
python3 engineering/marketing-site-builder/scripts/access_links_check.py engineering/marketing-site-builder/assets/flawed_page.html --format json
Workflow
Step tags: [PROVEN] means the step has held up across many projects; [RECOMMENDED] means a strong default you can depart from with a stated reason.
Step 1 — Decide the kind [PROVEN]
Write the kind on the plan's first line and on the page as <html data-kind="...">.
| Kind | Examples | Leads with | Effects |
|---|---|---|---|
| tech | SaaS, AI and developer tools, fintech apps, games, hardware | The real interface as live HTML, or the device photographed | One per page |
| product | Furniture, bikes, food sold online, cosmetics | The product whole and large, a new view each section | None |
| service | Clinics, law firms, agencies, studios, trades | The people, the rooms, the finished work | None |
| place | Cafes, bakeries, guesthouses, shops, gyms | The room, the counter, the goods; two or more photographs per screen | None |
A maker that also opens its door is a place. When unsure between product and service, ask what the buyer takes home: a thing is a product, an outcome is a service.
Step 2 — Write the copy first, from facts [PROVEN]
- Fill the fact sheet in
assets/site-plan-template.mdfrom the brief, word for word where possible. Empty fields stay empty; their sections shrink or go. - Pick one register for the voice card (
references/copy-voice-and-tells.md, section 2). Never the default warm-wry-literary voice unless the client asks for it. - Write every line under the five rules: the h1 names the category and the buyer or place in 2 to 8 words; one hard fact on the first screen; a noun in every heading; buttons as verb plus object; at most two hero actions, each with one wording everywhere.
- Shape the page from the reader's questions; write the microcopy in the same register
(
references/copy-voice-and-tells.md, sections 2 and 5). - Lint the draft before any markup exists, words first, then devices:
python3 engineering/marketing-site-builder/scripts/copy_lint.py draft.md
python3 engineering/marketing-site-builder/scripts/copy_devices.py draft.md
Validation checkpoint: both exit 0, and the draft passes the outside reads (a rival's name in the
h1 and headings, read as the client, a fresh reader where a subagent is available;
references/copy-voice-and-tells.md, section 8).
Step 3 — Plan the page as screens [RECOMMENDED]
One line per section in the plan: its id, its one idea, its picture and that picture's kind (object, use, parts, people, place, interface, diagram, type, slot), its height (one screen or two at 1440x900) and any motion with its purpose. A section with no picture and under half a screen of content merges into the one above. Five to nine sections for a landing page; long copy goes to sub-pages.
- The product or hero object leads the hero and at most two more sections; three object lines at most.
- No card layout more than twice, never the same layout in neighbouring sections; two picture-less sections never side by side.
- Every asset the brief names (trailer, screenshots, CAD) gets a finished-looking slot with a stand-in.
Rhythm rules: references/one-system-and-rhythm.md, section 3; content rules:
references/sections-cards-forms.md, section 1.
Step 4 — Set one system [PROVEN]
Define tokens on :root before any rule: a spacing scale on a 4px grid, two to four radii, a
near-black ink, a muted text colour, two or three grounds, one hairline, one accent, two font
families, one button pair. Use var(--token) everywhere. Take the accent from the client's
material, set headlines by size and tracking rather than weight, keep everything at 13px or more and
reading text at 16px. Defaults and the reasons behind them: references/one-system-and-rhythm.md,
sections 1 and 2.
Step 5 — Build [RECOMMENDED]
- Every section is a direct child of
main, opens with its h2, and uses the one.wrapso all edges line up. - Photographs: the client's first, stand-ins listed for replacement, library stock only in detail
slots; alt, width and height on every image, the hero eager, the rest lazy
(
references/imagery-motion-effects.md, section 2). - Motion in three layers only: the hero's entrance once, one or two moments in their own sections, the hover and press layer. No scroll reveals. A reduced-motion rule on every page that moves.
- Effects only on tech, one per page, with a still fallback.
- Cards two levels deep at most; labelled forms that really send; every control named, every link
landing; no carousel or popup (
references/sections-cards-forms.md).
Step 6 — Run the finish gate [PROVEN]
python3 engineering/marketing-site-builder/scripts/finish_gate.py index.html --kind product \
--css dist/assets/app.css --state .site-gate.json
Fix every FAIL; fix each WARN or keep it and say why. Waive a false FAIL in site-waivers.json
with its reason (five at most). Then read screenshots at 1440, 1024, 820x1180, 768x1024 and 390
against the browser checklist (references/finish-gate.md, section 2).
Validation checkpoint: the last line is GATE: PASS, GATE: OPEN after the third round (stop and
report), or GATE: DONE when the user said to stop (--done "<their words>").
Step 7 — Report
The GATE: line first, then every waiver and open FAIL, the kind and voice card, every assumed fact
(data-assumed), stand-in photograph and client-asset slot, the form's stand-in address, and the
browser checks not yet run (references/finish-gate.md, section 4).
Decision Frameworks
How much may move
| Layer | tech | product, service | place | |---|---|---|---| | Hero entrance on load | Yes | Yes | Yes | | Moments in their own section | 1 or 2 | 1 or 2, led by photographs or type | 1 or 2, led by photographs or type | | Hover and press layer | Yes | Yes | Yes | | Card loops, counters, one timed loop | Yes | No | No | | Shader, canvas, WebGL, gradient grounds, glass | One effect | No | No | | Scroll reveals, parallax on text, scroll-jacking | Never | Never | Never |
When a check fires
| Finding | First move |
|---|---|
| hype-word, stock-phrase | Replace with the fact from the fact sheet; if there is none, cut the line |
| vague-cta | Name the transaction: verb plus what the visitor gets |
| Spacing or radius over budget | Map every off-scale value to the nearest token; delete the leftovers |
| Colour or accent hues over budget | Collapse greys into ink, muted and hairline; keep one accent |
| effects on a non-tech page | Replace the effect with the photograph it was standing in for |
| img-lazy, img-dims | Add loading="lazy" below the hero and real dimensions everywhere |
| reduced-motion | Add the media query that rests every transition and animation |
| stat-strip, testimonials, apology-note | Each number in its sentence; only the brief's quotes; real material or no element |
| a11y-name, field-label, form-action | Name the control; a visible label; a real endpoint or a listed mailto stand-in |
| anchor-target, label-target, main-action | An id that exists; one wording per target; the hero's action repeated word for word |
| violet, default-face, heavy-weight | Accent from the client's material; a pairing with a reason; size and tracking, not bold |
| contrast-pair, tiny-text, small-body | 4.5:1; nothing under 13px; reading text 16px |
| no-css, kind-missing | Pass the built stylesheet with --css; write <html data-kind> or pass --kind |
| checker-gaming | Remove the headless test; the page behaves the same for every visitor |
Anti-Patterns
The Template Section Stack
Mistake: Every section built as eyebrow label, claim heading, short lede, three icon cards, with a strip of three round figures under the hero. Why it happens: It is the shape that fits any business, so it is what templates and generators produce first, and it looks finished at a glance. Instead: Shape sections from the reader's questions (what is it, what does it cost, where, how does it work, why trust it) and the material on hand. A section with nothing true to say goes.
Synonym Swapping
Mistake: Replacing "seamless" with "smooth" and "empower" with "help" to clear the linter. Why it happens: The linter reports words, so the quickest fix looks like a word change. **I
Truncated for display — read the full file on GitHub.
Related Skills
taste-skill
116.8kAgent skills that give coding agents taste: art direction, reference images and a browser check instead of generic AI pages
Claude-Code-Game-Studios
39.7kTurn Claude Code into a full game dev studio — 49 AI agents, 72 workflow skills, and a complete coordination system mirroring real studio hierarchy.
LocalAI
50.4kLocalAI is the open-source AI engine. Run any model - LLMs, vision, voice, image, video - on any hardware. No GPU required.
siyuan
47.3kAn open-source, privacy-first, self-hosted knowledge workspace where humans and AI agents work together 开源、隐私优先、自托管的知识工作空间,让人与智能体在此协作
Languages
Trust signals
From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.
