- Home
- Skills
- Frontend & Web Apps
- Anti-Slop Design
More screenshots
Works with the AI Agents you already use
Anti-Slop Design
Scores any frontend 0-100 for AI-slop tells — purple gradients, three-card grids, fake social proof — then gives the exact fix with file:line precision.
Free
Anti-Slop Design
Example session with this skill installed
Scan examples/sloppy.html from this skill and tell me exactly what makes it read as AI-generated. Then show me the fixed versio
- Read your context and instructions
- Compiled the anti-slop design
- Generated the design tokens stylesheet
Summary
slop_scan.py scored examples/sloppy.html at
100/100 (grade F) — 13 high-severity tells. The top five were structural, not cosmetic, so I rebuilt around one direction instead of patching classes. Same copy, same sections, re-scored at
0/100 (grade A).
$ python scripts/slop_scan.py examples/sloppy.html --top 5
anti-slop design scan v1.0.0
scanned: 1 files, 116 lines
SLOP SCORE 100 / 100 grade F
13 high, 10 medium, 3 low
[HIGH] COPY-01 AI filler vocabulary x5 (-15.7)
sloppy.html:41 "empowers modern teams to unleash their potential"
[HIGH] COLOR-01 Purple/indigo gradient x3 (-15.6)
sloppy.html:36 bg-gradient-to-r from-indigo-500 to-purple-600
[HIGH] LAYOUT-01 Centred everything x11 (-14.7)
sloppy.html:32 min-h-screen flex items-center justify-center text-center
[HIGH] A11Y-02 <img> without alt x3 (-13.9)
[HIGH] COLOR-05 Low-contrast grey text x10 (-12.2)
FIX FIRST
1. COPY-01 AI filler vocabulary (-15.7)
2. COLOR-01 Purple/indigo gradient (-15.6)
3. LAYOUT-01 Centred everything (-14.7)
What changed
| Before | After | |
|---|---|---|
| Direction | none stated | Warm Swiss Editorial |
| Non-negotiables | — | no shadows; hairline rules only; vermilion accent max 5× per screen |
| Layout | centered hero + 3 identical cards | 7/5 asymmetric split + numbered rows |
| Type | Inter 400/700, gradient-clipped h1 | Fraunces 600 display, Archivo body, -0.035em tracking |
| Depth | 6+ shadows | 1px rules at 16% ink |
| Motion | transition-all + hover:scale-105 | one 700ms reveal group + reduced-motion guard |
| Score | 100 / F | 0 / A |
The two highest-penalty fixes
/* before — a gradient doing the work a type scale should do */
h1 {
background: linear-gradient(90deg, #6366f1, #9333ea);
-webkit-background-clip: text;
color: transparent;
font-size: 3.75rem;
line-height: 1.5;
}
/* after — hierarchy from weight, size and tracking */
h1 {
font-family: var(--f-display);
font-size: clamp(2.75rem, 7.2vw, 4.25rem);
font-weight: 600;
line-height: .95;
letter-spacing: -.035em;
}
h1 em { font-style: normal; color: var(--accent); }
<!-- before: three identical cards, emoji icons -->
<div class="grid grid-cols-1 md:grid-cols-3 gap-8">
<div class="rounded-2xl shadow-lg p-8 text-center">🚀<h3>Lightning fast</h3></div>
<div class="rounded-2xl shadow-lg p-8 text-center">✨<h3>Beautiful by default</h3></div>
<div class="rounded-2xl shadow-lg p-8 text-center">🎯<h3>Built for teams</h3></div>
</div>
<!-- after: numbered rows, asymmetric columns, hierarchy in structure -->
<ol class="rows">
<li><span class="num">01</span><h3>Asks one question, once</h3>
<p>Each person gets a single prompt at a time they choose.</p></li>
<li><span class="num">02</span><h3>Pushes you to decide</h3>
<p>When answers stop changing, the room asks for a decision and an owner.</p></li>
</ol>
Next steps
- Run
python scripts/slop_scan.py src/ --top 8on your own repo; fix the top five by penalty, not the easiest five. - Add
--threshold 24to CI so a design regression fails the build next to your tests. - Keep
examples/sloppy.htmlandexamples/designed.htmlside by side — that pair is what F and A look like.
sloppy.html
CSS · design tokens
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
Generic, AI-generated UI has a fingerprint: Inter, indigo gradients, blurred blobs, three identical feature cards, and "seamless" filler copy. None of it is broken. All of it is undecided — and none of it shows up in code review.
This skill makes it measurable, then fixes it.
slop_scan.py walks your HTML, CSS, JSX, TSX, Vue, Svelte and Astro files, applies 47 weighted rules across seven categories (typography, color, layout, motion, copy, accessibility, process), and returns a score from 0 to 100 with exact file:line hits and a fix-first ranking.
Before and after — same product, same copy, both pages ship in the package:
| Score | Grade | |
|---|---|---|
| Typical agent output | 100 | F |
| After the process | 0 | A |
What it does
- Scan — 47 deterministic rules, standard library only, no network, well under a second.
--jsonfor diffs,--threshold 24to fail a CI build on a design regression. - Diagnose — every finding maps to a taxonomy entry: symptom, why agents produce it, the concrete fix.
- Direct — 19 named directions (warm swiss editorial, brutalist utility, technical instrument, quiet luxury, tabloid newsroom, Devanagari-first editorial…) with font pairings, hex palettes, layout signatures and the trap each one falls into.
- Rebuild — tokens before components: type ratio, tinted neutrals, radius ladder, one-elevation rule, motion with a
prefers-reduced-motionguard. - Repair — 14 before/after recipes, ordered so findings drop fastest.
How it works
- Extract the brief: product, audience, one job, one emotional target.
- Scan the codebase and read the top 5 penalties as structural signals, not chores.
- Commit to a direction plus three non-negotiable rules.
- Author tokens before the first component.
- Compose, then verify: re-scan to grade A or B, plus squint, 3-second, grayscale, 320px and keyboard checks.
Use cases
- Audit a frontend before a launch.
- Rescue an "AI-made" landing page that reads as a template.
- Set a design baseline for a greenfield project.
- Add a design gate to CI next to your tests.
- Review a frontend PR for hierarchy, not just for bugs.
Why this beats prompting it yourself
"Make it look modern" produces the same indigo gradient cards every time, because nothing in the prompt rejects them. This skill ships a hard Never List, a scoring function, and 19 directions that each exclude most possible designs. You cannot pass them by accident.
Known limitations
The scanner is a detector, not a judge. A page can score A and still be boring — the "one memorable thing" rule and the five visual checks cover that. It does not generate logos, images or assets. The scanner needs Python 3.9+, standard library only. A supplied brand or reference design always overrides the rules.
Compatibility: framework-agnostic, with Tailwind-aware rules for class-based tells. Works in Claude Code, Cursor, Codex CLI, Gemini CLI, GitHub Copilot, Windsurf and OpenCode.
Requires Python for the scanner. Does not generate images or logos. Cannot judge subjective 'beauty', only detects known slop patterns. Strategy may conflict with existing brand guidelines.
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
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed today
- Free with an Agensi account
Requires
Permissions
File Scopes
Scans only frontend source files. No network access, no file writes, no environment variables.