css-slop-detector

Score your CSS/HTML for AI-generated design cliches — the "AI slop" look: oversized border radii, glassmorphism on everything, purple gradients, neon glows, pure-black dark mode with no light toggle, and giant centered hero titles.

If AI built your site, this tells you in one number how obvious it is. v2 adds --fix: a one-click detect-and-fix loop that rewrites the slop for you (with backups and dry-run diff previews).

Zero dependencies. Python standard library only.

Install

pip install css-slop-detector

Or run the single file straight from the repo:

git clone https://github.com/hahahahahahahahah6/css-slop-detector
cd css-slop-detector
python3 css_slop.py scan style.css

Requires Python 3.9+.

Usage

Scan

$ css-slop styles.css
styles.css: score 68/100 (slop)
  L12 [big-radius] oversized border-radius (32px) on '.card'
  L12 [glass] backdrop-filter blur on '.card' (glassmorphism)
  L15 [purple-gradient] purple/violet gradient on '.hero'
  L22 [hero-center] giant centered title (font-size: 72px) on '.hero h1'
  L1 [no-light-mode] dark-only design: near-black backgrounds with no
     light-mode toggle or prefers-color-scheme: light found

Works on .css files and .html files (scans <style> blocks and inline style="..." attributes). Machine-readable output with --json; fail CI with --fail-over 50 (exit 1 when any file scores 50 or higher).

Fix (v2)

Preview what the fixer would change — writes nothing:

$ css-slop --dry-run styles.css
styles.css: 7 fixes (big-radius x1, glass x1, glow-shadow x1, hero-center x1,
  no-light-mode x1, pure-black x1, purple-gradient x2) — score 100 -> 0 [dry-run, nothing written]
--- a/styles.css
+++ b/styles.css
...

Apply the fixes. The original is backed up to styles.css.bak first:

$ css-slop --fix styles.css
styles.css: 7 fixes (...) — score 100 -> 0 [applied, backup: styles.css.bak]

$ css-slop --restore styles.css   # undo: restore the backup

What --fix rewrites (only patterns the scanner flags — it de-slops, it doesn't redesign):

Rules

Scores: 0–20 clean, 21–50 mild slop, 51–80 slop, 81–100 peak slop.

Differentiation

Generation side vs. detection+fix side. Tools like pbakaus/impeccable (and design-skill collections like ui-skills or hallmark) teach AI to generate better design — they improve the prompt. This does the opposite: it scores the finished product's AI fingerprints, then closes the loop with one command — --fix rewrites the flagged patterns for you, with a diff preview, automatic backups, and --restore to undo. Detect + fix, not just detect.

One file, no dependencies, runs in CI.

License

MIT