AI coding agents are incredible at generating interfaces. They're also incredible at generating the same interface — gradient text on a dark background, glass cards with glow shadows, Inter font everywhere, identical three-column feature grids, and CTAs that technically exist but nobody can actually read because the contrast is 2.3:1.
That's AI slop. It's not broken. It's not ugly. It's just... the same. Every time.
Pixelslop opens your actual pages in a real browser, screenshots them at three viewports, extracts computed styles, measures contrast ratios, checks 25 known AI slop patterns, scores design quality on 5 pillars, evaluates from 8 user personas, and fixes what it finds. It doesn't read your CSS and guess — it renders the page and measures what's actually there.
npx pixelslop installInstall is interactive by default. It detects Claude Code and Codex CLI, lets you pick runtimes and scope, then copies agent specs and verifies a direct Playwright browser runtime.
Need: Claude Code or Codex CLI installed first.
npx pixelslop@latest update # upgrade to latest — backs up old files, shows diff
npx pixelslop doctor # verify installation health
npx pixelslop status # show scope, install root, and all installed runtimes
npx pixelslop uninstall # remove everything cleanly/pixelslop http://localhost:3000
Or skip the URL — pixelslop finds running servers, detects static HTML sites, and asks before touching anything:
/pixelslop
It scores 5 pillars (Hierarchy, Typography, Color, Responsiveness, Accessibility), detects slop patterns, groups findings by priority, asks what you want to fix, then runs the fix loop with automatic checkpoints and rollback.
Full walkthrough: docs/getting-started.md
I built stitch-kit to teach AI agents how to generate beautiful UIs using Google Stitch. Pixelslop is the other side of that coin. Stitch-kit helps agents create good design. Pixelslop catches when they didn't. Together: generate with taste, verify with evidence, fix what slipped through.
| Doc | What's in it |
|---|---|
| Getting Started | Install, first scan, flags, static site support |
| Troubleshooting | --debug flag, session logs, common issues, filing bug reports |
| Architecture | Agent roles, two-phase execution, subagent limitations, file layout |
| pixelslop-tools | Full CLI reference — plan, checkpoint, gate, discover, serve, log |
| Personas | Built-in personas, custom persona creation, JSON schema |
| Contributing | Adding patterns, fix guides, checkpoint protocol, persona system |
npm test # 540 tests, zero dependencies
npm run test:tools # pixelslop-tools CLI
npm run test:installer # installer unit tests
npm run test:orchestrator # agent spec validation
npm run test:persona # persona schema validation
npm run validate # resource file structure checksAutomated via release-please. Conventional commits, auto-changelog, npm publish with OIDC provenance. feat: bumps minor, fix: bumps patch.
- stitch-kit — 35 skills that teach AI agents to generate beautiful UIs with Google Stitch. Pixelslop catches what slips through.
- slopbuster — strips AI patterns from text and code. Same philosophy, different domain — pixelslop catches visual slop, slopbuster catches written slop.
Apache 2.0 — see LICENSE and NOTICE.
Built by Gabi @ Booplex — because AI agents are getting scary good at generating UIs, and someone needs to make sure "generated" doesn't mean "generic." Apache 2.0.