More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Award-Level Websites: Art Direction + Anti-Slop Design QA

    1

    Your agent designs websites that don't look AI-made, then proves it: 12 art directions, WCAG-verified tokens, original art, a motion kit and a 155-check design QA with screenshots.

    $9.99

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Award-Level Websites: Art Direction + Anti-Slop Design QA

    Award-Level Websites: Art Direction + Anti-Slop Design QA

    Example session with this skill installed

    Build a landing page for my coffee roastery, Saltbush in Cape Town. We sell subscriptions (that's the main goal) and run a small café. I don't want it to look like every other AI-made site. Facts and prices are in notes.md.

    • Read your context and instructions
    • Compiled the award-level websites
    • Generated the UI component

    Done: saltbush/site/ scores 100/100 (A) on the design QA, with 0 AI tells and 0 blockers, checked in a browser at 1440, 834 and 390 px.

    Direction: Organic Handmade, re-cut to avoid the coffee cliché. For a roastery, the predictable AI look is cream paper, a serif headline with one italic word, terracotta and bean icons. I went the other way:

    • a deep moss ground with one honey accent;
    • Caprasimo, a chunky 70s display face that nods to your restored 1970s roaster, with Karla for text;
    • saltbush-leaf silver as a second ground.

    Organic was in the top three on board.html; I chose it over Swiss Poster because moss and honey carry warmth without cream.

    The page (8 sections, chapters grammar)
    1.

    Hero: "Coffee roasted on Tuesday, posted by Thursday." Beside it, a drawn cooling tray of beans that turns slowly as you scroll, and a live "Next roast" date.
    2. The four origins and Harbour, set huge as an index with process and tasting notes.
    3. A quiet section: the founders and the 12 kg drum.
    4.

    The signature move (the peak): a green bean roasts to brown as you scroll, while a ring closes Farm → Drum → Post → Door.
    5. Thandi's quote on a honey block.
    6. Three plans with bags drawn to scale (R149 / R279 / R499).
    7. The café menu, hours, Saturday cupping and the address.
    8. Close: the next roast date and "Start a subscription".

    Checks

    Audit: 100/100 static and with the rendered checks merged in. Every text pair passes WCAG AA, CLS stays under 0.1, and there's no overflow at any width.

    • No JavaScript or reduced motion: the bean rests roasted and all four steps are listed.
    • Fingerprint gate: passed and logged in FINGERPRINTS.md, so your next site can't repeat it.

    Before you launch, please confirm (from claims-to-confirm.md):

    • Thandi's quote and permission;
    • every price, time, date and contact detail on the page (28 more items, each with its line number);
    • the domain and the phone number.

    Files: site/ (index.html, assets, 3 drawn SVGs, og.jpg, favicon), design/ (direction.md, board.html, tokens.css/json, Tailwind, DESIGN.md), qa/ (report.html, fixes.md, contact sheet, screenshots), BRIEF.md.

    Next steps:

    1. Confirm the claims.
    2. Point the plan buttons at your checkout.
    3. Deploy site/ to any static host.

    board.html

    TSX · React component

    Generated

    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

    Your agent's websites look AI-made. This skill fixes that, and proves it with a score.

    Agents build the average of every site they have seen. In 2024 that was Inter, a purple gradient and emoji cards. In 2026 it's cream paper, a serif headline with one italic word, mono "01 /" labels and pill buttons. Both read as generated, and most "award-winning design" prompts only fight the first one.

    This skill gives your agent an art director's process and the tools to back it up. The scripts make the design system and the art, then measure the page. Your agent makes the decisions, writes the copy and builds.

    What you get

    • 12 art directions, ranked for your business. Swiss Poster, Organic Handmade, Architect Monograph, Pop Sticker, Cinematic Noir, Clinical Calm and six more, each with seeded font pools and palette variants. Directions that sit too close to your industry's AI cliché are held back, with the reason given. board.html shows the top 3 as mini heroes set in your brand name before any code exists.
    • A verified design system. Every colour pair is repaired to WCAG (body 7:1, secondary 4.5:1, UI 3:1), with a fluid type and space scale. Exports to CSS variables, Tailwind v4 @theme (default palette removed), Tailwind v3 config, W3C DTCG tokens, Google DESIGN.md and WordPress theme.json. Bring your own brand with --existing or --palette.
    • Original art, no stock photos. Seeded SVG plates in your palette (topographic, flow-field, ridgeline, halftone, Bauhaus shapes, dither, layered depth for a parallax hero), plus a monogram, favicon and OG card.
    • A zero-dependency motion kit. Pinned stages with cues, horizontal rails, clip wipes, split and fit-to-width headlines, colour drift, a magnetic CTA and counters. Every scroll act publishes a --p progress variable for the one signature move each site gets. It works with JavaScript off and with reduced motion.
    • A fingerprint gate. Each site is logged on 7 design dimensions, and a new plan must differ from every earlier build. A reworded copy still fails, so your client sites never come out looking alike.
    • A 155-check design QA across craft and anti-slop, accessibility, responsive, performance, SEO & share, and conversion. It gives a 0–100 score with Gen-1 and Gen-2 AI tells named down to file:line, plus a client-ready report, fixes ranked by points gained (with code), a critique sheet, a badge, and a claims-to-confirm list that catches invented stats and placeholder testimonials. --min 90 turns it into a CI or Stop-hook gate.
    • Rendered QA (optional). Screenshots at 1440, 834 and 390 px in normal, no-JS and reduced-motion passes, checked for overflow culprits, clipped text, tap targets, pixel-sampled contrast, fonts that never loaded, CLS, dead scroll, and reveals that never fire.
    • A one-second demo. One coffee roaster built three ways: the 2024 default (39/F), the 2026 default (64/D) and the art-directed rebuild (100/A), side by side.

    Proof, not promises

    Four complete sites were built by agents following SKILL.md: a coffee roaster, a SaaS, an architecture studio and a music festival. They use four different directions, each has one bespoke scroll moment, and all score 99–100 with zero AI tells. See the screenshots.

    Why it beats a design prompt

    A "world-class designer" promptAward-Level Websites
    "Be bold"12 ranked art directions with verified tokens
    Catches the 2024 look at bestScores 2024 and 2026 AI tells separately, with line numbers
    Same agent, same taste, same siteA fingerprint gate enforces variety in code
    Stock photos or empty boxesOriginal palette-locked SVG art
    "Looks good to me"155 checks, screenshots at 3 widths, a score and a gate
    Made-up stats and testimonialsAn honesty gate with a claims-to-confirm list

    Requirements

    An agent that can run code with Python 3.8+. Standard library only: no installs, no API keys, offline. Screenshots use Playwright if you already have it.

    • A high audit score proves the page is accessible, responsive, honest and free of AI tells; it cannot prove it is beautiful. The taste comes from the art direction, the brief and looking at the screenshots, which the workflow makes your agent do.
    • No photo or video generation: art is original generated SVG (topographic, flow-field, ridgeline, halftone, Bauhaus, layered depth landscapes, OG card). Your own photos always beat it for people, products and places.
    • The static audit reads HTML and CSS as delivered and does not run JavaScript; content rendered only in the browser is checked by the optional shoot.mjs pass (needs Node 18+ and Playwright).
    • Framework projects (Next.js, Astro, React, WordPress) are audited through their built HTML output; the tokens and motion kit drop into those stacks, but new scaffolds plain HTML.
    • Copy checks (stock phrases, formulaic cadence) are tuned for English.
    • Google Fonts are linked by default; self-host them if your privacy policy requires it (explained in references/stack-notes.md).

    How to use

    1. 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. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 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 1 month ago

    Requires

    tracewell
    Node.js
    style-dictionary

    Permissions

    Terminal / Shell
    Read Files
    Write Files

    Terminal / code execution: runs the bundled scripts: scripts/studio.py (Python 3.8+ standard library) and, optionally, scripts/shoot.mjs (Node 18+ with Playwright, only if you already have it). It starts no other programs and installs nothing. Read files: your site or build-output folder, the brief and design files it writes, and the skill's own references, assets and examples. Write files: only the folders you name (design/, site/, qa/ by default) and FINGERPRINTS.md when you log a finished site. It never overwrites an existing index.html and only deletes a previous demo folder that carries its own marker file. Network access: none required. shoot.mjs serves your site folder on a temporary 127.0.0.1 server to a local headless browser; the page itself may load its own web fonts (Google Fonts by default, self-hosting explained). Nothing is uploaded anywhere. No environment variables, API keys, secrets, telemetry or binaries.

    Listed1 month ago

    What's inside

    Frequently Asked Questions

    Those tell your agent to be bold. This one gives it an art director's process and tools: 12 art directions with verified tokens, original art, a motion kit, a fingerprint gate so sites never repeat, and a 155-check audit that scores the result and names every AI tell with the line it's on.

    Gen-1 is the 2024 look (Inter, purple gradients, emoji cards). Gen-2 is what 2026 models produce by default: cream paper, a serif headline with one italic word, mono "01 /" labels, pill buttons. Most anti-slop skills only catch Gen-1; the bundled demo shows this skill catching both.

    Yes. Tokens export as CSS variables, Tailwind v4 @theme, Tailwind v3 config, W3C DTCG JSON and WordPress theme.json, and the audit runs on your built HTML. The scaffold command makes plain HTML sites.

    No. The core is Python standard library. Screenshots at 3 widths use Playwright if you already have it; otherwise the agent tells you the one-line install and carries on with the static audit.

    Yes. `direct --existing` reads your tokens.css, Tailwind config or DESIGN.md, keeps your colours and fonts, repairs contrast and only adds the missing roles.

    No. Facts come from you or are marked [[CONFIRM]], and the audit flags invented-looking stats and placeholder names in a claims-to-confirm list before launch.

    New in Frontend & Web Apps