no-ai-design-slop — Claude Code skill: strip AI design slop
A passive quality gate for UI work: it flags reflex design choices and common interface defects while you build or revise a screen.
npx skills add MengTo/Skills --skill no-ai-design-slop -a claude-codeWhat no-ai-design-slop does
The skill runs in the background of any design task. Its test is blunt: slop is not a color, font, gradient or card, it is a choice made by reflex instead of for the product. A choice looks suspect when it could be pasted into an unrelated product unchanged, repeats a familiar generated pattern, or fights the local design system and nearby sections.
What sits inside is a set of compact gates to run before keeping an element: system, hierarchy, composition, typography, color and material, product truth, interaction states, motion and access. Then a removal test. Name the element, say what job it does, take it away in your head. If meaning, state or hierarchy survives and clarity improves, delete it. If something real is lost, make the smallest correction inside the existing system.
The skill tells the agent to read local instructions, tokens, DESIGN.md, screenshots and existing components before picking a direction, and not to widen a narrow edit into a full site audit. Full pages, redesigns and detailed reviews pull in ARTICLE.md from the repo. When you want a formal review with no edits, the author points to a separate audit-ai-design-slop workflow.
What it helps with
- catch generic AI patterns before shipping a layout
- check hover, focus, loading and error states on new components
- stop the agent from swapping an authored style for a neutral template
- find decorative layers with no job and remove them
- match a new section to the project's tokens and type scale
Example prompts for the agent
- run no-ai-design-slop over this page and cut what doesn't earn its place
- check the new card grid for slop without redesigning the whole thing
- do an anti-slop pass on the landing page, keep the current direction
How to install no-ai-design-slop
1Skills CLI (Vercel)
npx skills add MengTo/Skills --skill no-ai-design-slop -a claude-code2Manually: put the files in place
~/.claude/skills/no-ai-design-slop/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/no-ai-design-slop/
1Skills CLI (Vercel)
npx skills add MengTo/Skills --skill no-ai-design-slop -a codex2Manually: put the files in place
.agents/skills/no-ai-design-slop/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add MengTo/Skills --skill no-ai-design-slop -a cursor2Manually: put the files in place
.cursor/skills/no-ai-design-slop/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install MengTo/Skills no-ai-design-slop2Manually: put the files in place
.github/skills/no-ai-design-slop/SKILL.md1Skills CLI (Vercel)
npx skills add MengTo/Skills --skill no-ai-design-slop -a gemini-cli2Platform CLI
gemini skills install https://github.com/MengTo/Skills --path agent-skills/ui/no-ai-design-slopManually: put the files in place
.devin/skills/no-ai-design-slop/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add MengTo/Skills --skill no-ai-design-slop -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about no-ai-design-slop
How do I install no-ai-design-slop in Claude?
Run npx skills add MengTo/Skills --skill no-ai-design-slop -a claude-code. The skill then lives at ~/.claude/skills/no-ai-design-slop/SKILL.md.
What does no-ai-design-slop do?
It acts as a passive quality gate during UI work, flagging reflex choices and concrete interface defects. It can also run an explicit anti-slop cleanup while preserving the chosen art direction.
Will it flatten my design into a neutral template?
No. The skill forbids replacing a distinctive choice with a fashionable neutral template, and it does not prescribe a new font, palette or component library unless the task allows it.
Source text
SKILL.md · MIT
--- name: no-ai-design-slop description: Prevent and remove generic AI-generated design defaults, incoherent visual choices, and established UI defects while creating, revising, or reviewing websites, apps, screenshots, mockups, and design code. Use as a passive quality gate during UI work or for an explicit anti-slop cleanup; preserve the chosen art direction instead of forcing a neutral redesign. --- # No Design Slops Act as a passive design quality gate. Keep the product's direction, personality, and useful choices. Catch generic defaults and design failures before they compound. ## Core Judgment Slop is not a color, font, gradient, card, or animation. It is a choice made by reflex rather than for the product. Treat a choice as suspect when several of these are true: - it could be pasted into an unrelated product unchanged - it repeats a familiar generated-design pattern - it conflicts with the local design system or nearby sections - it communicates no useful information, state, action, hierarchy, or brand meaning - it competes with content, weakens trust, or makes the interface harder to use Do not guess whether AI made the artifact. Judge the visible result. ## Apply Passively For every applicable design task: 1. Read the local instructions, tokens, `DESIGN.md`, screenshots, references, and existing components before choosing a direction. 2. Identify the product, audience, primary task, primary action, content hierarchy, and one visual thesis for the surface. 3. Preserve established decisions unless the user asks for a redesign. 4. During implementation, run the compact gates below whenever adding a section, component, effect, or state. 5. Render the result at relevant viewports and inspect the actual interface, not only the source. 6. Remove or correct the highest-impact problem created or exposed by the work. For a narrow edit, do not widen the task into a site audit. Fix the requested surface and avoid introducing new slop. For a full page, redesign, anti-slop pass, or detailed review, read [ARTICLE.md](ARTICLE.md) and use the relevant catalog sections. If the user asks only for a formal review, use the separate `audit-ai-design-slop` workflow and do not edit the artifact. ## Design Principles ### Start from context - Prompt and design from evidence: product content, real constraints, references, and the local system. - Use references to extract hierarchy, pacing, contrast, material, and interaction principles. Do not copy identity, layout, assets, or copy. - When no system exists, define a compact one before polishing: type roles, spacing rhythm, palette, radii, borders, shadows, imagery, icons, and motion. ### Choose one coherent idea - Give each viewport one focal point and each flow one clear primary action. - Let typography, media, color, material, and motion support the same visual thesis. - Prefer one strong authored moment over many unrelated effects. - Use contrast and asymmetry deliberately; do not neutralize character in the name of cleanliness. ### Make relationships visible - Use proximity before containers. Related items sit closer than unrelated items. - Use hierarchy before labels. Size, weight, placement, and contrast should do more work than pills, eyebrows, numbers, and captions. - Use depth only when the interface has a real layering model. - Let repeated components express a real repeated content type, not a convenient template. ### Make the product specific - Choose sections from the product's story, buying journey, and usage flow instead of a default landing-page sequence. - Show real product behavior, useful screenshots, or honest placeholders. Do not manufacture proof. - Art-direct imagery and iconography to the subject. Generic stock, abstract SVG filler, and default icon tiles are not substitutes for product meaning. - Adapt imported patterns to the current typography, palette, shape language, and density. ### Make motion explain something - Motion may explain state, causality, hierarchy, continuity, or spatial change. - Do not animate merely to prove that the page is interactive. - Keep reading and controls stable. Support reduced motion and complete static states. ## Compact Quality Gates Before keeping a choice, check: - **System:** Does it use the established type, color, spacing, radius, icon, and motion language? - **Hierarchy:** Is the most important content or action obvious without decorative labels? - **Composition:** Are alignment, balance, proximity, overlap, and negative space intentional at every relevant viewport? - **Typography:** Are roles distinct, readable, and optically spaced without oversized tracking or forced display treatments? - **Color and material:** Does each gradient, glow, glass layer, border, shadow, and accent have a clear role? - **Product truth:** Are imagery, copy, metrics, screenshots, logos, and states specific and honest? - **Interaction:** Are active, hover, focus, loading, empty, error, disabled, selected, and success states present when needed? - **Motion and access:** Does motion help, remain performant, preserve input, and respect reduced-motion preferences? ## Removal Test For every suspect element: 1. Name it precisely: selected state, eyebrow, radial light, nested card, icon tile, marquee, fake proof, clipped popover, or another concrete pattern. 2. State what job it performs. 3. Remove it mentally. If meaning, state, action, hierarchy, or brand character survives and clarity improves, delete it. 4. If deletion creates a real loss, make the smallest correction using the existing system. 5. Add a replacement only when the interface needs one. Do not compensate with a new effect. ## Boundaries - Do not treat a technique as slop in isolation. - Do not replace a distinctive choice with a fashionable neutral template. - Do not prescribe a new font, palette, layout, component library, or art direction unless the task authorizes it. - Do not turn every content block into a card or every improvement into decoration. - Do not add new sections, effects, colors, fonts, assets, or dependencies during a cleanup unless they solve a demonstrated loss. - Do not remove useful density, edge, humor, asymmetry, or expressive motion merely because it is unusual. - Do not invent customers, metrics, testimonials, awards, ratings, product screens, or activity. ## Verification Before finishing: - compare the rendered result with the local system and supplied references - inspect the primary flow, relevant states, and responsive breakpoints - confirm that text does not clip, overlap, overflow, or lose contrast - confirm that controls remain labeled, reachable, stable, and responsive - confirm that every remaining decorative layer has a defensible role - confirm that the result is more specific to this product, not merely more fashionable Report the material removals or corrections and why they improved the design. Do not return an AI-authorship guess or a generic numeric taste score.
© MengTo · MIT · MengTo/Skills