Use when iterating an existing design in place with A/B variants, an editable artboard, and anti-slop moodboards instead of generating from zero. Triggers on \"iterate design\", \"design variants\", \"artboard edit\", \"click to edit\", \"anti-slop\". Non-triggers: blank-page prototype generation (use cloud-design-prototyping), generic UI polish rules (use frontend-design-direction). Outcome: variant set against the existing system plus an edited winner ready to implement.
Installs into .claude/skills of the current project.
Are you the author of Design Iterate In Place?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/majinmagros-design-iterate-in-place)
---
name: design-iterate-in-place
description: "Use when iterating an existing design in place with A/B variants, an editable artboard, and anti-slop moodboards instead of generating from zero. Triggers on \"iterate design\", \"design variants\", \"artboard edit\", \"click to edit\", \"anti-slop\". Non-triggers: blank-page prototype generation (use cloud-design-prototyping), generic UI polish rules (use frontend-design-direction). Outcome: variant set against the existing system plus an edited winner ready to implement."
metadata:
origin: ECC
---
# Design Iterate In Place
Iterate mid-process on a design that already exists: read its system,
produce variants, edit on an artboard, and kill AI slop per section.
Rule: respect the existing design system; this skill is mid-process,
not zero-to-one.
## When To Activate
- The user says iterate this design, variants A/B/C/D, artboard,
click-to-edit, or fix this hero/landing.
- A design system or live page exists and must be respected.
- Output looks "AI-made" (neon glow overuse, repeated logo, 3+ text
blocks, weak grid) and needs a human finish.
- A previous run regenerated from zero and broke brand consistency.
## Workflow
1. Read first: extract tokens, type scale, spacing, and components from
the existing design or page. List what must not change.
2. Diagnose: name the competing elements (for example three dashed bands
fighting, accent color on low-priority spots) in one short list.
3. Moodboard per section: collect 2-3 consistent references per section
before generating. Famous skills all converge to "AI look" - the
moodboard is the antidote.
4. Generate variants: produce A/B/C/D against the same brief (cleaner
options first, one braver option last). Same content, different
hierarchy and rhythm.
5. Artboard edit: open the winner on an editable artboard (properties
panel, click-to-edit injects into the prompt). Edit in place, never
restart from a blank canvas mid-round.
6. Human finish: fix grid and margins, redo type by hand where needed,
add texture or glass sparingly, and run a side-by-side A/B before
locking.
7. Hand off: ship the edited variant plus a delta list (what changed vs
the original system) for implementation.
## Anti-Patterns
- Zero-to-one regeneration when a system already exists.
- Variant soup with no diagnosis of what is broken.
- One global moodboard for the whole page (slop returns per section).
- Editing screenshots by hand instead of via artboard -> prompt loop.
- Shipping without a side-by-side A/B against the original.
## Relations
- `cloud-design-prototyping`: external prototype flow; this skill is the
in-place counterpart for existing designs.
- `frontend-design-direction`: product design judgment reused in diagnosis.
- `impeccable-design`: live multi-version editing behind the artboard step.
## Sources
- No external design-tool URL confirmed at write time - verify the native
design skill and artboard protocol of your current harness before use.
- Variants A/B/C/D, Figma-style editable artboard (properties panel,
click-to-edit into prompt), mid-process-not-zero-to-one rule, and the
## Exemplo
```text
Landing com hero poluído (3 faixas tracejadas + glow neon + logo repetido)
Diagnóstico: 3 elementos brigando → moodboard por seção (não global)
Variantes A/B/C/D no mesmo brief → vencedora no artboard (click-to-edit)
Fecha: A/B lado a lado vs original + delta list p/ implementação
```
## Swarm de variantes lado-a-lado vs dividido (leva YouTube rodada 10)
Dois modos: DIVIDIR (coordenador + 2 workers, cada um numa direcao) vs LADO-A-LADO (mesma tarefa, 2-6 variantes com layout diferente ou estilo diferente). Cap de frames para economizar tokens. Passe o pre-check em swarm-readiness-gate antes de subir o enxame.