brenbuilds1/skills/slop-audit

Slop Audit

Every list of "your site looks vibe-coded" tells mixes two things: the defaults a model reaches for on any brief, and one author's taste. This audit keeps only the tells that independent sources agree on (the viral list, a Reddit study of 3.2 million posts, the published catalogs, and the design guidance shipped with coding agents), names the items those sources reject, and reports with receipts. It never rewrites; the fixes are the builder's call.

Method

1. Grep the source with the checks below, with grep -E (or rg) so

the | alternations work. Every hit is a receipt: file, line, matched text.

1. Screenshot the home page and one inner page at 1440 and 390 wide.

A fresh session judges the screenshots against the eye checks; the agent that built the page does not grade it.

1. Report every finding as one of three verdicts. A tell is on the

agreed list or is a leftover fact. A smell is named by one or two sources. A choice is on someone's list but is design when done on purpose; it is reported so the builder can confirm it was chosen, and never counted against the page.

Tells (agreed across sources)

Leftovers (facts, not taste)

Builder residue proves how the page was made regardless of how it looks: lovable-tagger, gpteng.co, /lovable-uploads/, @base44/sdk, "Edit with Lovable", "Made in Bolt", "Built with v0", a builder subdomain, and section comments like <!-- Hero --> or {/* Testimonials */} left in shipped markup (a weak signal on its own). Grep for each.

Smells (one or two sources)

Bento grids; a four-column footer; the hero, features, proof, pricing, FAQ, footer sequence in that order; ALL-CAPS eyebrow labels over every heading; one word in a headline in a different color; an arrow glued to every link; meta strings joined with middle dots; neon glow nobody asked for; radial orbs behind the hero; testimonials with stock names or placeholder text (grep lorem ipsum); no page linked at /terms or /privacy; em dashes and the negation-then-elevation line in the copy (grep — and it'?s not .{1,40}, it'?s). Also the second default: cream background, high-contrast serif, terracotta accent. Swapping purple for that is trading one predetermined look for another.

Choices (one list names them, or a source calls them fine when chosen)

Drop shadows, hover states, rounded corners, a white background, a dot grid, a terminal window, checkmark bullets, a colored left stripe, pastel palettes, a font from the tells row paired with a display face that carries the identity. The tell is never the element; it is the element applied everywhere with nothing chosen. Report these as choices, ask once, and move on.

Report (fixed format)

slop-audit: 4 tells, 2 smells, 3 choices (choices not counted)
| verdict | tell | where | receipt |
|---|---|---|---|
| tell | purple gradient | src/Hero.tsx:14 | `bg-gradient-to-r from-indigo-500 to-purple-600` |
| tell | three equal cards | src/Features.tsx:9 | `grid md:grid-cols-3 gap-6`, three identical cards |
| tell | emoji as icons | src/Features.tsx:12 | `🚀 Fast`, `✨ Simple`, `🔒 Secure` |
| tell | copy | src/Hero.tsx:8 | "Elevate your workflow" |
| smell | eyebrow labels | screenshot home@1440 | tracked-out caps over every section |
| smell | orbs | src/Hero.tsx:3 | `radial-gradient` blurred behind the hero |
| choice | drop shadow | src/Card.tsx:5 | `shadow-lg` on 3 of 3 cards. chosen? |

Reports only. Never rewrites.