- Home
- Skills
- Frontend & Web Apps
- Site from References: Brief to Website in 8 Steps
Works with the AI Agents you already use
Site from References: Brief to Website in 8 Steps
The reason is simple — the AI improvises from images and a vague request.
$9.99
Site from References: Brief to Website in 8 Steps
Example session with this skill installed
I want a landing page for my pilates studio in Austin,
Balance Studio. The goal is booking a free trial class through WhatsApp. I like the look of https://reference-one.example and https://reference-two.example, and I saved 2 screenshots in etapas/00_briefing/prints/.
- Read your context and instructions
- Compiled the site from references
Summary
All 8 steps are done for
Balance Studio. Steps 1, 5, 6 and 7 were approved by you and logged in etapas/aprovacoes.md. The site lives in site/, validated against the build plan, with a Dockerfile ready.
Progress
| # | Step | File | Status |
|---|---|---|---|
| 1 | Brief | etapas/00_briefing/briefing.md | ✅ approved |
| 2 | Code extraction (2 sites) | etapas/01_extracao/codigo/*/resumo.md | done |
| 3 | Screenshot extraction (2) | etapas/01_extracao/prints/resumo.md | done |
| 4 | Preview | etapas/02_previa/previa_site.md | done |
| 5 | Site image | etapas/03_mockup/mockup_home-desktop.svg | ✅ approved (v2) |
| 6 | Wireframes (1 page × desktop/mobile) | etapas/04_wireframes/ | ✅ approved |
| 7 | Design system | etapas/05_design_system/ | ✅ approved |
| 8 | Build & validation | site/, etapas/06_construcao/ | done |
Design system (excerpt of generated tokens)
| Token | Value | Source |
|---|---|---|
--color-primary | #2F5D50 | your decision in the brief |
--color-bg | #FAF7F2 | measured on reference 1 |
--color-secondary | #D98E5F | estimated from screenshot 2 |
--font-heading | Fraunces 600 | measured on reference 2 |
--font-body | Inter 400 | measured on reference 1 |
--space-24 | 96px between sections (--space-16 on mobile) | measured on reference 1 |
WCAG contrast: body text on background 9.8:1 (AAA); primary button 5.1:1 (AA).
:root {
--color-primary: #2F5D50;
--color-bg: #FAF7F2;
--color-secondary: #D98E5F;
--font-heading: "Fraunces", serif;
--font-body: "Inter", sans-serif;
--radius-full: 999px;
--space-24: 96px;
}
Connects securely to your tools. The creator never sees your data.
About this skill
Stop getting generic websites from AI
Ask an AI agent to "make a site like this one" and you usually get the same thing: a template-looking layout, guessed colors, random typography, and hours of rework. The reason is simple — the AI improvises from images and a vague request.
Site from References replaces improvisation with a process. It is a bundle of 10 skills that walks your agent through 8 steps, from understanding the business to a validated, deploy-ready site. Each step writes a structured text file that the next one reads, and the steps that shape the result only move forward with your approval.
The 8 steps
| # | Step | What you get | Approval |
|---|---|---|---|
| 1 | Interview & brief | Questions in rounds, each with a recommended answer → briefing.md | ✅ |
| 2 | Design from code | Colors, fonts, spacing, radii, shadows and breakpoints measured in a real browser (computed styles) | |
| 3 | Design from screenshots | Palette with area and contrast (script) + typography and shape analysis, marked as estimated | |
| 4 | Site preview | Pages, sections, content and visual direction in Markdown | |
| 5 | Site image | Prompt + full-color SVG mockup | ✅ |
| 6 | Wireframes | SVG for every page, desktop and mobile, with a review page | ✅ |
| 7 | Design system | design_system.md + visual HTML version + tokens.css + Tailwind tokens, with WCAG contrast checks | ✅ |
| 8 | Build & validate | Build plan, section-by-section code (tokens only), Docker/deploy and a validation report |
Why it works
- Measured, not guessed. Step 2 opens your references in a real browser and reads computed styles — including what JavaScript renders.
- Explicit trust order. Your decision > value measured in code > value estimated from a screenshot > mockup proposal. No silent conflicts.
- One visual source of truth.
design_system.mdrules; code uses tokens only. Changing a color? Change it in the design system and regenerate tokens with a script. - Approvals at the right moments. Fixing a brief takes minutes; fixing a finished site takes hours.
- Resumable. Progress lives in
etapas/aprovacoes.md. Stop today, continue tomorrow — the orchestrator figures out which step you're on. - Ethical by default. References inform structure and style only. Their text, images, logos and code never reach your site — your brief's content or clearly marked placeholders do.
- Secure. Ships with
auditar-skill, which scans any third-party skill for prompt injection, data exfiltration, credential access, obfuscated code and destructive commands before installing it.
What's included
- 10 skills: orchestrator + one skill per step + security audit
- 9 Python/JS scripts (computed-style extraction, screenshot palette, WCAG contrast, CSS/Tailwind tokens, HTML design system, wireframe validation)
- Templates for brief, preview, image prompt, wireframe, design system, build plan and validation report
- Ready project structure (
etapas/,site/,CLAUDE.md) and setup scripts that audit and install the companion skills - Stack chosen in the interview (HTML/CSS, Tailwind, Vite or frameworks); default HTML + Tailwind, mobile-first
Who it's for
Freelancers and agencies shipping landing pages, developers tired of "looks like a template", and founders who know what they want but can't translate it into design.
Use cases
- Competitor Redesign: Replicating the design language of a reference site while using custom business logic.
- Landing Page Sprint: Moving from a blank slate to a deployed, containerized landing page with a full design system.
- Safe Plugin Integration: Auditing external AI skills for security vulnerabilities before adding them to your workspace.
Known limitations
Design extraction from screenshots provides estimated values rather than exact CSS. Playwright extraction requires a local browser environment to capture computed styles accurately.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Frontend & Web Apps.
Screenshot extraction uses estimated values. Playwright requires a local browser for computed style capture. Large JS-heavy sites may require manual cookie consent bypass during extraction.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 15 days ago
- Software
Python
Node.jscreate-video
Git
Permissions
Allowed Hosts
Reads and writes stay inside the project: etapas/ holds the brief, extractions, wireframes and design system; site/ holds the generated code. Network is used only to fetch the HTML/CSS of the reference URLs you provide and, during setup, to download companion skills from GitHub — each audited before install. No data is sent to external servers and no API key is needed.