- Home
- Skills
- Frontend & Web Apps
- Award-Grade Web: Awwwards-Level Site Builder
More screenshots
Works with the AI Agents you already use
Award-Grade Web: Awwwards-Level Site Builder
This skill is a manually invoked workflow (`/award-grade-web`) for building sites that a design jury would take seriously, and that still work on an ordinary phone.
$10
Award-Grade Web: Awwwards-Level Site Builder
Example session with this skill installed
/award-grade-web quick
Landing page for Linea, an architecture studio in Mersin. Visitors are potential clients on phones and desktops; the one job is to get them to send an enquiry by email.
Real content (use only this):
- Deniz House, 2021 - a family house on a sloping plot, organised along one retaining wall
- Liman Library, 2023 - a public reading room with a single continuous bench
- Tuz Pavilion, 2025 - a small timber pavilion, drawn first as one roof line
- Contact: hello@linea.example
No photos yet. One self-contained HTML file.
- Read your context and instructions
- Compiled the award-grade web
- Generated the UI component
Summary
Quick mode, L1 Crafted (sample brief, content-light studio site, phone-first audience; budgets: up to 9 viewports, JS up to 100 KB). Delivered one self-contained index.html: 2.2 viewports,
no JS libraries,
0 FAIL / 0 WARN on the second checker pass.
Concept: one drawn line is the whole identity, and the page draws it as it loads.
Motif: a single 2px accent rule plus hairline dividers between works.
Signature moment: the accent rule draws left-to-right under the hero headline (CSS only, off under reduced motion).
Voice: short, declarative, no superlatives. Copy uses only the content supplied in the request; there are no invented numbers, quotes, logos or awards.
Design system (tokens set before layout)
| Token | Value | Why |
|---|---|---|
| Display type | Fraunces, clamp(3.5rem, 11vw, 12rem), line-height .88, tracking -0.035em | Extreme scale contrast, tight display setting |
| Body type | Instrument Sans 18px | Readable, different role from the display face |
| Colour | paper #f4f1ea, ink #16150f, one accent #b8431a | Neutral base plus one accent |
| Motion | one easing cubic-bezier(.2,.7,.1,1), two durations | One motion character |
Checker run 1 (render_check.py --level L1 --quick)
0 FAIL, 2 WARN
- WARN ai-tells: reveal animation on 83% of sections (fade-up everywhere)
- WARN tap-targets: 3 interactive elements smaller than 24x24 px on mobile
a 45x19 "Works" / a 46x19 "Studio" / a 57x19 "Contact"
Fixes made: reveal kept only on the three work rows (the one place it carries meaning), nav links given real tap padding.
Checker run 2
0 FAIL, 0 WARN
Families: Instrument Sans, Fraunces | Body 18px, largest text 158px
Page length: 2.2 viewports | canvases: 0 | WebGL context requests: 0
Perf (local): LCP 244 ms, CLS 0.033, load 171 ms
Transfer (compressed estimate): 113 KB in 5 requests, JS 0 KB, ~1.2 s on slow 4G
The deliverable (excerpt of the delivered index.html)
<style>
:root{
--paper:#f4f1ea; --ink:#16150f; --accent:#b8431a; --mute:#5c5a50;
--font-display:"Fraunces",Georgia,serif; --font-body:"Instrument Sans",system-ui,sans-serif;
--ease:cubic-bezier(.2,.7,.1,1); --d1:.25s; --d2:.7s;
--pad:clamp(1.25rem,4vw,4rem);
}
h1{font:400 clamp(3.5rem,11vw,12rem)/.88 var(--font-display);letter-spacing:-.035em;max-width:12ch}
.rule{height:2px;background:var(--accent);transform-origin:left;animation:draw 1.1s var(--ease) .1s both}
@keyframes draw{from{transform:scaleX(0)}to{transform:scaleX(1)}}
.js .reveal{opacity:0;transform:translateY(16px);transition:opacity var(--d2) var(--ease),transform var(--d2) var(--ease)}
@media (prefers-reduced-motion:reduce){.rule{animation:none}.js .reveal{opacity:1;transform:none;transition:none}}
</style>
<h1>One line, <em>three</em> buildings.</h1>
<p class="lede">Linea is an architecture studio in Mersin. Every project starts with a single drawn line and is judged by whether the finished building still holds it.</p>
<article class="work reveal"><span class="n">01</span><div><h3>Deniz House</h3><p>A family house on a sloping plot, organised along one retaining wall.</p></div><span class="y">2021</span></article>
The .js flag is set by a script with a 2.5 s failsafe, so every reveal ends visible even if the observer never fires.
Next steps
- Add real photographs: the skill defines one image treatment to apply to all of them once they exist.
- Replace
hello@linea.examplewith the studio's real address. - For a Full-mode run (official or client work) the skill adds the mid-range phone pass, no-JS and no-WebGL passes and a three-juror blind review. Quick mode is a self-check only and reports no jury score.
index.html
TSX · React component
Example file from a real run - the skill writes it into your workspace.
Connects securely to your tools. The creator never sees your data.
About this skill
Most AI-built sites look the same: a competent template with no idea. This skill is a manually invoked workflow (/award-grade-web) for building sites that a design jury would take seriously, and that still work on an ordinary phone. It is calibrated on a dataset of 116 jury-scored sites and ships as a folder of plain Markdown and Python, so it works in any agent that reads the SKILL.md format.
What it does
- Concept engine: inventories the subject's real nouns, writes the obvious idea first, generates three more through different lenses, scores them, and pressure-tests the winner. The signature moment must survive the phone's first screen.
- Baseline duel (Full mode): two fresh agents build the same brief without the skill, and blind jurors compare your build against the stronger one in randomised order. The skill has to win a majority of jurors, with one revision round and a rematch if it loses.
- Intensity levels L1 to L3: Crafted, Expressive and Immersive, each with a length budget and a JS budget, so spectacle never silently breaks real devices.
- Design system before layout: type scale, colour tokens, grid, one easing family, one image treatment.
- Imagery pipeline: a source ladder (your photos, then made-for-the-site renders, then graded stock, then art-directed SVG) and
scripts/image_pipeline.py, which applies one crop, surface, grade and grain to every image and writes WebP srcsets. - Render checker:
scripts/render_check.pyrenders desktop, phone, mid-range phone, reduced motion, no-JS and no-WebGL, and flags text cut-offs, placeholders, dead bands, fixed-bar overlap, heavy fonts, generated-looking patterns and copy smells (invented numbers, quotes, logo walls). - Honest copy rules: never invents numbers, testimonials, client logos, awards or addresses.
How it works
- Mode: Quick (self-check) or Full (baseline duel and blind jurors). Full is mandatory for official, client or competition work.
- Brief and level: subject, audience, devices, the one primary job.
- Baselines, research, concept: measured against what the same model builds without the skill.
- Design system, then page architecture, then build in layers: semantic HTML, CSS, a motion layer gated by reduced-motion, WebGL only as an enhancement.
- Verify and review: the checker must end with 0 FAIL; Full mode adds the jury verdict.
Use cases
- Portfolios, launch pages and campaign sites that need a distinct idea
- Official institutional sites where accessibility and performance matter as much as looks
- Competition or investor-facing sites
- Auditing an existing page for generated-looking patterns
Known limitations
- Needs the whole folder installed, not SKILL.md alone.
- The automated check needs Python 3 and Playwright with Chromium; without them the skill falls back to a manual checklist and says so.
- Full mode takes roughly 1.5-2x the time and tokens of Quick mode because it builds and judges extra sites.
- Jury scores are estimates from simulated jurors, not predictions of real Awwwards results, and it is not a WCAG audit.
- Sites that depend on photography (restaurants, hotels, jewelry) need real photos or made-for-the-site images; the skill will not invent them.
Requires Python/Playwright for automated audits. Performance scores are estimated. Effectiveness is capped when real assets (photos/copy) are replaced by placeholders.
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 any payment.
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 14 days ago
Requires
Permissions
Allowed Hosts
Shell/Read/Write: runs scripts/render_check.py and scripts/image_pipeline.py and writes the page, processed images and check output to the work folder. Browser: headless Chromium (Playwright) renders the page at desktop, phone, reduced-motion, no-JS and no-WebGL. Network: the checker answers blocked CDN and Google Fonts requests from npm tarballs (registry.npmjs.org). Environment variables: only optional proxy variables and AGW_CACHE (cache folder). The scripts never send your files anywhere.