Unslop
aahil62/unslop
Audita, entrevista e retematiza sites React existentes com verificação em cada etapa
Ver no GitHubPlugin para Claude Code que audita, entrevista e retematiza sites React + Tailwind já existentes. Executa um pipeline stateful com verificação de qualidade em cada etapa: escaneia problemas de design com evidências visuais, sintetiza uma identidade visual customizada baseada na marca, e aplica as mudanças com validação de build e screenshots após cada passo.
Instalação
npx convez-ai aahil62/unslop1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Refatorar sites vibe-coded ou com design padrão em identidades customizadas
- Melhorar conformidade WCAG e contrast ratios com base em evidências visuais
- Gerar relatórios antes/depois para clientes com validações de qualidade
Veredito

Ferramenta especializada e bem-estruturada para resgate de design em sites existentes. Destaca-se pelo pipeline determinístico com verificação hard (screenshots, build, contrast) após cada mudança, eliminando garantismo vazio — exige configuração inicial de Playwright mas entrega resultados mensuráveis com relatórios para stakeholders não-técnicos.
README
A tradução para o português deste README ainda está na fila.
Unslop
The rescue tool for sites you already built. Audit → interview → re-theme an existing React + Tailwind site into a custom identity — verified at every step, without breaking your build.
Most design skills generate new UI or lint one dimension of it. Unslop does neither: it takes a site you've already shipped and runs it through a single stateful pipeline — measure what's wrong, decide what "good" means for your brand, then apply it with a hard build-and-screenshot check after every change. The output of each stage feeds the next, and nothing claims "done" without evidence. It's a Claude Code plugin (Codex-compatible) with three stages:
/unslop:audit— scores your site against an evidence-based anti-slop checklist: screenshots at 4 breakpoints, WCAG contrast checks, and a static scan for the classic tells (default indigo buttons, purple-to-blue gradients,transition-all, font-weight sprawl…). Output:DESIGN-AUDIT.mdwith a score and file:line evidence. Read-only — it never touches your code./unslop:interview— asks about your brand, audience, and taste one question at a time, then synthesizes a custom design lane: identity, palette, type pairing, texture, motion budget, and one signature moment. Rules-driven creativity — no preset menu, no two sites alike. You approve the lane before anything changes./unslop:retheme— applies the approved lane systematically: tokens → typography → section by section → states & motion. After every pass the build must succeed and fresh screenshots are taken. Ends with a re-audit, a before/after report, and a client-readyCLIENT-REPORT.htmlyou can hand to a non-technical site owner.
See it work
A stock vibe-coded landing page (default indigo, purple→blue gradient, emoji icons, gray-on-white text failing contrast) scored 27/100. After one audit → interview → retheme pass into a synthesized "Print Run" lane — 100/100, WCAG AA in both light and dark, no horizontal scroll at any width.
| Before — 27/100 | After (light) — 100/100 | After (dark) |
|---|---|---|
![]() | ![]() | ![]() |
Install
Claude Code:
/plugin marketplace add aahil62/unslop
/plugin install unslop@unslop
Codex:
codex plugin marketplace add aahil62/unslop
Other AGENTS.md-aware tools: clone this repo and point your agent at AGENTS.md — same workflow, same rules.
The audit and retheme stages use Playwright for screenshots and contrast checks. In the project being rescued:
npm i -D playwright && npx playwright install chromium
(If the chromium download is blocked, the scripts fall back to your installed Chrome.)
Use
From your site's repo:
/unslop:audit # how bad is it, with evidence
/unslop:interview # define what *good* looks like for this brand
/unslop:retheme # make it so, verified pass by pass
What's inside
skills/ audit · interview · retheme (the three workflows)
rules/ anti-slop-checklist.md (scoring rubric) · lane-synthesis.md (creativity rules) · client-report.md (deliverable format)
examples/lanes/ 4 worked design lanes — calibration for the synthesizer, not a menu
scripts/ screenshot.mjs · contrast-check.mjs · slop-scan.mjs (deterministic checks)
AGENTS.md Codex adapter
Scope (v1, deliberately narrow)
- React + Tailwind only — what v0, Lovable, Bolt, Claude, and Cursor actually produce. Other stacks are politely declined.
- Rescues existing sites; it doesn't scaffold new ones.
- No MCP server, no framework, no methodology — three skills and a rubric.
Freelancers: a before/after retheme with an audit score delta is a sellable
deliverable — and the retheme stage ends by generating it: a client-ready
CLIENT-REPORT.html with the score delta, before/after screenshots, and what
changed in plain language. Quote the outcome, not hours.
License
MIT — see LICENSE. Pre-pivot methodology kit lives on the
legacy-archive tag.


