- Home
- Skills
- Frontend & Web Apps
- Anti-Slop Design Pro
More screenshots
Works with the AI Agents you already use
Anti-Slop Design Pro
Finds the AI tells in a React frontend and stops them returning: 12 headless components, 3 token presets, and a CI gate that fails the build on design regression.
$15
Anti-Slop Design Pro
Example session with this skill installed
Our settings page is in src/settings.tsx and it is the usual card soup. Scan it, then rebuild the form with the Pro kit: tabs, one save surface, and delete behind a dialog. Then show me what the CI gate does when this drifts six weeks from now.
- Read your context and instructions
- Compiled the anti-slop design pro
- Generated the UI component
Summary
Rebuilt the settings page on the Pro kit — three tabs, one save surface per tab, and the destructive action behind a dialog. The gate passes at
0/100 (grade A) with threshold 10. I also ran the drift that happens six weeks later, to show what the gate catches.
The page
import { Button, Dialog, Field, Tabs, useAnnouncer } from "./anti-slop/components/src";
export function Settings({ name, errors }: { name: string; errors: Record<string, string> }) {
const { announce, region } = useAnnouncer();
const [confirm, setConfirm] = useState(false);
return (
<>
{region}
<Tabs
label="Settings"
items={[
{ id: "profile", label: "Profile", content: <Profile name={name} errors={errors} announce={announce} /> },
{ id: "notify", label: "Notifications", content: <Notify /> },
{
id: "danger",
label: "Danger",
content: <Button tone="danger" onClick={() => setConfirm(true)}>Delete workspace</Button>,
},
]}
/>
<Dialog
open={confirm}
onClose={() => setConfirm(false)}
title="Delete this workspace?"
description="Windows, releases and history are removed. This cannot be undone."
actions={
<>
<Button tone="quiet" onClick={() => setConfirm(false)}>Keep it</Button>
<Button tone="danger" onClick={() => setConfirm(false)}>Delete workspace</Button>
</>
}
>
<Field label="Type the workspace name to confirm" required>
{(props) => <input {...props} autoComplete="off" />}
</Field>
</Dialog>
</>
);
}
What the kit decided for me
| Decision | Made by |
|---|---|
| Focus moves into the dialog, Escape closes, focus returns to the trigger | Dialog |
| Tab never leaves the dialog, body scroll locked | Dialog |
Label, hint and error bound to one input via useId | Field |
| One tab stop, arrow keys inside it | Tabs |
| "Settings saved" announced without stealing focus | useAnnouncer |
The gate, today
$ python scripts/ci_gate.py --path src --threshold 10
SLOP SCORE 0/100 grade A (threshold 10)
Exit code 0. Green.
The gate, six weeks later
Someone adds a "Why teams switch" card. Nobody reviews it. This is the real output:
$ python scripts/ci_gate.py --path src --threshold 10
SLOP SCORE 30/100 grade C (threshold 10)
::error file=src/Settings.tsx,line=13,title=SLOP COLOR-01 Purple/indigo gradient surface::1x in this file set, penalty -9.0. `from-indigo-500 to-purple-600` (or #6366f1 -> #a855f7) is the canonical AI hero. Commit to one accent hue and a real neutral ramp instead.
::error file=src/Settings.tsx,line=12,title=SLOP TYPE-01 Default AI font stack::1x in this file set, penalty -8.0. Inter / Roboto / Montserrat / system-ui is the agent default, not a decision. Choose a face with a point of view and defend it against the brief.
::error file=src/Settings.tsx,line=13,title=SLOP TYPE-02 Gradient-clipped heading text::1x in this file set, penalty -7.0. `bg-clip-text text-transparent` on a headline is the single most common AI tell. Use weight, size, or a colour block instead of a gradient on type.
::warning file=src/Settings.tsx,line=13,title=SLOP COLOR-02 Default indigo/violet accent hex::1x in this file set, penalty -6.0. These exact hexes ship in every generated UI. Pick an accent from the brand or from a real reference image, and derive a ramp from it.
::error::Design regression: score 30 exceeds threshold 10. Run the scanner locally with --top 8 to see the fix order.
Exit code 2. The pull request is red before anyone reads a design review.
Next steps
- Swap the token sheet for a preset:
@import "./tokens.rail.css"after the shape. - Replace the three remaining raw
<button>elements withButton. - Wire the workflow from
examples/ci/github-actions.ymlinto this repo.
rail.css
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
One good page is a prompt. The twelfth one is a system.
This is the paid edition of Anti-Slop Design. The scanner is identical to the free edition — nothing was removed to sell this. What you get on top is the machinery that stops the design drifting back after the first good page.
What it does
- Scores a frontend with 47 deterministic rules — default gradients, generic layouts, filler vocabulary, missing focus states — and points at the exact file and line.
- Ships 12 headless React components that get keyboard, focus and ARIA right without
a styling opinion:
Button,Field,Textarea,Select,Checkbox,RadioGroup,Tabs,Dialog,Disclosure,Table,Figure,LiveRegion. - Gives 29 aesthetic directions written as constraints rather than mood boards, plus 3 finished token presets: Swiss Rail Timetable, Night Shift, Desert Modern.
- Fails the build on regression with a composite GitHub Action. Score above your threshold, exit code 2, one annotation per finding.
- Keeps Figma and code in sync with a W3C / Tokens Studio token export generated from a single CSS source.
What is new versus the free edition
| Free | Pro | |
|---|---|---|
| Scanner | 47 rules | identical 47 rules |
| Directions | 19 | 29 |
| Token presets | 1 shape | 3 finished presets + the shape |
| React components | — | 12 |
| Figma export | — | W3C and Tokens Studio |
| CI | --threshold flag | composite action + GitHub, GitLab, Husky recipes |
| Docs | slop taxonomy, fix recipes | 6 page archetypes, CI policy, component contract |
Quickstart
unzip anti-slop-design-pro-1.0.0.zip -d anti-slop
# score what exists
python anti-slop/scripts/slop_scan.py src --top 8
# gate it: exit code 2 when the score crosses the threshold
python anti-slop/scripts/ci_gate.py --path src --threshold 24
import { Button, Field, Textarea } from "./anti-slop/components/src";
<Field label="What broke?" hint="One sentence is enough." required>
{(props) => <Textarea {...props} rows={4} />}
</Field>
<Button tone="primary" type="submit">Send report</Button>
No build step. React 18+ is the only peer dependency — no runtime packages, no CSS framework, no config file.
How it works
- Write the brief — product, audience, job, one emotional target. Name the one memorable thing before writing code.
- Scan what exists —
slop_scan.pyprints a score, a grade, and the fix order sorted by penalty. - Commit to one direction — from 29, with three non-negotiables that would visibly hurt to break.
- Tokens before components — pick a preset, then replace every value you cannot defend from the brief.
- Compose with the kit, then gate it — set a threshold in CI so the next twelve pull requests cannot quietly undo the decision.
Known limitations
- Components are React 18+ only. The scanner, directions, presets and CI gate are framework-agnostic and run on Vue, Svelte, Astro and plain HTML.
- The scanner reads code. It cannot judge composition or taste — the five visual checks
in
SKILL.mdcover that. - No date picker, combobox or data grid. Those need product decisions, and a half-correct one is worse than none.
- A supplied brand or reference design always overrides these rules.
License
Commercial, one license per person, team license covers 5. You may ship the components
in client work and closed-source products. You may not redistribute or publish the files.
Full text in LICENSE.
Requires
Python 3.9+ for the scanner — standard library only, nothing to pip install.
React 18+ for the components. No network access, no API key, no telemetry.
Components are React 18+ only. The scanner, 29 directions, token presets and CI gate work with any frontend, including Vue, Svelte, Astro and plain HTML. Python 3.9+ is required for the scanner. The scanner reads code-level tells; it cannot judge composition or taste.
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 today
Requires
Permissions
File Scopes
Reads and edits frontend source only. No network access, no environment variables, no telemetry. The scanner writes nothing — output goes to stdout.