- Home
- Skills
- Research & Analysis
- Website Design & UX Research
Works with the AI Agents you already use
Website Design & UX Research
Before building a landing page or planning a redesign, turn your reference website into an evidence-backed research brief for your designer or coding agent.
$9.99
Website Design & UX Research
Example session with this skill installed
Research resend.com (Daily mode). In English.
- Read your context and instructions
- Compiled the website design
Resend
2026-10-27 | https://resend.com/ | Mode: Daily | Source: sample site for a demo run
Research reliability: Capture A | Interaction — (not measured in Daily) | Responsive A | DOM/CSS — (not measured in Daily) |
Overall A (how reliable the evidence capture was, not how good the site's design is)
Capture (Final Capture Quality, quality_check.py final): desktop Initial A → Fallback not needed →
Final A; mobile Initial A → Fallback not needed →
Final A; gaps: none (two ~300px blank bands on desktop, y=9360–9660 and y=11392–11696, are normal spacing between blocks on the slices)
What this site does: Resend is an email-sending service for developers (transactional and marketing email). The page asks visitors to click "Get started" and sign up. The audience is engineers who want to wire email into their own code.
3 things learned today
Make "send your first email" the star of the page. The screen right below the hero is "Integrate tonight": an 18-line Node.js example (new Resend(...), emails.send) with 13 language tabs above it (Node.js, Serverless, Ruby, Python, PHP, CLI, Go, Rust and more) and framework tabs inside the code box (Next.js, Remix, Express, Nuxt…) (screenshots: d00, d01).
Why it works: a developer's biggest worry is "how long will integration take?", and runnable code answers it faster than any adjective. How I can use it: for a product sold to engineers, put the shortest runnable example on the second screen and switch languages with tabs, instead of writing "integrate in minutes".
2. A black page, a serif headline and one textured object create a quiet, upscale feel. The background is black, the hero headline "Email for developers" is set in a serif face while body text is sans-serif, and a black 3D puzzle cube sits on the right (screenshot: d00). The branding tool estimates h1 ≈ 96px, h2 ≈ 56px, body 18px, with the heading font stack starting with domaine and body in Inter.
Why it works: most email services lead with envelopes and blue; this page deliberately doesn't, turning a utility into an object with taste. How I can use it: a technical product that wants to stand out from its category can change only the headline typeface and the hero object instead of redesigning the whole interface.
3.
Use the product itself as the evidence for each feature. The Test mode HTTP 200 response list and the Webhooks event cards (Clicked, Opened) (screenshot: d01), the React example source next to the email it renders (screenshot: d03), and a real-looking metrics dashboard (screenshot: d04) are interface or source code, not illustrations.
Why it works: visitors see "what I will actually see" rather than a description of the feature. How I can use it: give each feature block one readable real fragment (a log, a setting, an output) on the same screen as its explanation.
Dimension notes (one or two lines each)
Business / conversion: sign-up → primary CTA "Get started" (about 4 times on the screenshots: nav, hero, React section, footer; secondary "Documentation", "Check the docs", "Explore Enterprise"). Path: hero → customer logo wall → code example → feature cards → CEO quote → dashboard → customer quotes (horizontal cards cropped at the edges, may auto-scroll (inference)) → footer CTA "Email reimagined. Available today." (screenshots: d00, d03, d05). The clearest mechanism is placing the "first email" code very early; no obvious conversion friction was observed this time.
Brand / style: probably wants developers to feel "stable, polished, no fluff" (inference), mainly through the black background, serif headline and a single 3D object; the copy is short engineer-to-engineer sentences ("We are a team of engineers who love building tools for other engineers").
Visual: background #000000, body grey #A1A4A5 (branding estimate). The branding tool reports the primary button as blue #00A3FF, but on the screenshot the nav and hero "Get started" buttons are dark-grey pills (screenshot: d00); the blue appears only in the example email's button code (screenshot: d03), so the screenshot wins.
UX: the nav has 7 items (six with dropdowns, plus Pricing), one Log in and one Get started; the page is a single-column long page where each block carries one claim and usually one "Learn more"; no sidebar is visible (screenshots: d00, d01).
Motion: the Test mode and Webhooks cards show truncated lists (screenshot: d01) that may auto-scroll (inference); the hero 3D cube may be video or WebGL (inference). Motion was not measured this time.
Responsive: on mobile (360 wide) the nav collapses to a menu button; the hero cube shrinks and moves above the headline; the logo wall goes from 12 logos on desktop (two rows of 6) to 6 visible logos (three rows of 2); "Get started" becomes a full-width button (screenshot: m00).
Patterns: ① "big heading + one or two sentences + Learn more" repeats in Test mode, Modular webhooks, Contact management and Broadcast analytics; ② every product block uses a real interface or code as evidence — the code example, Test mode, the React example and the dashboard (screenshots: d01, d02, d03, d04).
What not to copy (at most 3)
- Small grey text (
#A1A4A5) on pure black may be weak for long explanations (inference; contrast was not measured). - The 3D cube relates only indirectly to "email"; it relies on brand feel rather than explanation. Without such a brand asset, the hero may not show what is being sold.
One-line takeaway: it wants engineers to "try it right now", builds trust with a quiet black look, and finishes the persuasion with code that runs.
Still uncertain: motion and hover/click interactions were not measured this time; the CTA count was counted from screenshots (Capture A, 100% coverage); the Contact management and Broadcast analytics cards show only cropped lists (screenshot: d02).
Connects securely to your tools. The creator never sees your data.
About this skill
Found a website you like, but unsure what to borrow for your own project?
Before building a landing page or planning a redesign, turn your reference website into an evidence-backed research brief for your designer or coding agent.
You get:
- A Markdown research report saved to your project, covering layout, typography, visual hierarchy, UX and responsive behaviour.
- Reusable design principles—and elements you should avoid copying.
- Supporting evidence appropriate to the selected mode: screenshots, browser measurements and clearly labelled inferences.
- A handoff summary when you ask how to apply the findings to your own project.
Use it to study one reference website or compare the design approaches of several sites.
Three depths
- Daily (default, ~5 min): a one-page summary — 3 things you can learn today, one-line notes per dimension, and what not to copy.
- Standard (15–20 min): adds real-browser measurements at 1440 / 390 / 768 px (screenshots, DOM, computed CSS, CTA inventory, hover / focus / click, menu), every claim sourced.
- Deep (15–40 min): adds evidence IDs and grades, pixel sampling, reduced-motion and motion measurement, a 3-width comparison table, design-system inference and a self-review.
What makes it different
- A reliability grade on every report when the scripts are installed (A–D for capture, interaction, responsive, DOM/CSS). It rates how complete the evidence was — not how good the site is. Blank WebGL screenshots, bot-check pages and "browser not supported" walls are detected and reported as research limits, never as design findings.
- "Not observed" is never "does not exist". Capture gaps downgrade the claim.
- Interaction checks verify the target, the action and the outcome. A click that lands on a cookie banner, a carousel arrow or a dead
href="#"is reported as unverified, not as a working CTA flow. - No invented conversion claims. Without analytics it describes conversion strategy and labels hypotheses as inference.
- Research only. It does not build the implementation. Ask for "build it like this" and you get a hand-off summary (principles, token starting points, what not to reuse).
- Compare several researched sites, or pick which of ten Daily sites deserve a Standard pass.
What you need
- Claude Code and a Firecrawl MCP server with your own Firecrawl account (third-party service, separate pricing; you configure it in your own agent settings).
- Optional but recommended: Python 3.9+,
pillow,playwrightand Chromium. With them you get objective capture grading, reliability ratings and the Standard/Deep browser measurements. Without them you still get Daily-level research, and the report says "reliability: unverified" instead of inventing a grade. - Windows consoles that are not UTF-8 (e.g. Traditional Chinese cp950) need
PYTHONUTF8=1.
Language
The skill is written in Traditional Chinese, so reports are Traditional Chinese by default. Add "in English" (or another language) to your request and the whole report follows: the text, the reliability line and the screenshot citation tags. The research rules and evidence grading stay the same.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Research & Analysis.
- Public pages only. Login walls, paywalls, bot protection (e.g. Cloudflare challenges) and "browser not supported" pages are recorded as research-environment limits, not design findings.
- Sites that render blank in screenshots (heavy WebGL, scroll hijacking, lazy reveal) are retried with segmented capture; remaining gaps are reported as "not observed", and the reliability grade drops.
- Motion cannot be seen in screenshots: it is inferred from CSS values and before/after measurements, and labelled as inference.
- Brand positioning is interpretation and is labelled as such.
- Without the optional scripts, only Daily-level research is delivered and reliability is reported as "unverified".
- Reports default to Traditional Chinese. Add "in English" (or another language) to your request and the whole report follows.
- It researches existing sites only; it does not write the implementation.
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
- Accounts
Firecrawl MCP
- Optional
Python 3.9+PillowPlaywrightChromium
Install Chromium with playwright install chromium.
Permissions
Allowed Hosts
File Scopes
Shell: runs the bundled Python scripts and curl to download screenshots. Browser: headless Chromium via Playwright to measure the page being researched. Network: fetches the target website and calls your configured Firecrawl MCP. Write Files: saves reports and screenshots under research/. Environment Variables: reads PW_CHROMIUM and HTTPS_PROXY only (both optional). Nothing is sent to the skill author.