Visual Cook
ideel-1/visual-cook
Transforma discussões de design vagas em decisões documentadas e testáveis
Ver no GitHubSkill para Claude Code que sistematiza decisões de design através de entrevistas estruturadas, convertendo linguagem vaga em documentação concreta. Cria dicionários visuais, testes em harness e Design Decision Records para trabalhos de UI/UX.
Instalação
npx convez-ai ideel-1/visual-cook1 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
- Definir vocabulário visual e sistema de design
- Resolver discordâncias sobre estética através de renders concretos
- Documentar trade-offs de design com Design Decision Records
- Policiar consistência visual em componentes
- Integrar decisões de design no fluxo de desenvolvimento
Veredito

Ferramenta bem pensada que resolve um problema real: converter conversas vagas sobre design em documentação durável e referências visuais. Complementa bem workflows de design-to-code ao forçar clareza antes da implementação.
README
A tradução para o português deste README ainda está na fila.
visual-cook
A Claude Code skill for grilling visual work — pinning fuzzy design language ("cleaner", "premium", "pop", "like Linear") to concrete decisions, grounding disagreements in real renders, and recording the result in a durable visual vocabulary.
Use it when:
- a UI "feels off" and you can't say why
- you're setting up or extending a visual system
- you're polishing a single detail and want a real conversation about trade-offs
What it does
Instead of generating a design and asking "is this good?", visual-cook interviews you. It walks the decision tree one question at a time, with its own recommended answer for each. Every vague word either gets pinned in a dictionary (DESIGN.md) or shown on screen via a scratch harness that imports the real component — so when the answer lives in the pixels, you point rather than argue in prose.
The skill documents and decides; it does not edit your component code. When a variant wins, the decision is recorded; a later session (or you) lands it.
It leaves behind:
DESIGN.mdat the repo root — a fuzzy→concrete dictionary (Premium = one accent max, zero shadows. _Not_: more whitespace.)docs/ddr/— Design Decision Records, but only for choices that are hard to reverse, surprising without context, and the result of a real trade-off..visual-cook/<topic>/— a scratch harness showing current vs. 2–3 labeled variants in one viewport. Gets deleted once the implementer lands the values.
Install
Drop the three files into a Claude Code skills directory:
# user-level (available in every project)
mkdir -p ~/.claude/skills/visual-cook
cp SKILL.md DDR-FORMAT.md DESIGN-FORMAT.md ~/.claude/skills/visual-cook/
# or project-level (committed alongside the repo)
mkdir -p .claude/skills/visual-cook
cp SKILL.md DDR-FORMAT.md DESIGN-FORMAT.md .claude/skills/visual-cook/
Claude Code auto-discovers skills in those locations. Restart your session and the skill is available.
Use
In any Claude Code session, point Claude at the work:
"Let's visual-cook the empty state on the dashboard."
Claude will start the interview, spin up a harness at .visual-cook/<topic>/, announce the preview URL, and grill from there. No need to invoke a command — the skill description triggers on UI-feel and visual-system phrasing.
Files
| File | What it is |
|---|---|
SKILL.md | The skill itself — loaded by Claude Code when a session matches. |
DESIGN-FORMAT.md | Format reference for the DESIGN.md dictionary. |
DDR-FORMAT.md | Format reference for Design Decision Records. |
Philosophy
Most visual work goes wrong in one of two ways: arguing about fuzzy words without referents, or making decisions in the harness that vanish at handoff because nobody wrote them down. visual-cook is structured around both failure modes — pin or render, never argue in prose, and everything compositional gets a dictionary entry before the harness goes away.