anti-slop-ui-redesigner

Use when a frontend project, webpage URL, or UI screenshot feels generic, template-like, visually weak, unresponsive, or visibly AI-generated and needs a product-specific redesign with preserved behav

SKILL.md

GitHub

Anti-Slop UI Redesigner

Turn one authorized page into an evidence-backed audit and a verified redesign. Deliver real code changes or a runnable standalone implementation. Never stop at a critique, mockup, or list of suggestions.

Select the input mode

Read input-modes.md, select exactly one mode, and state it.

Mode Source Required delivery
project Local code project Modify the project and preserve its contracts
url Authorized webpage URL Capture it in a real browser; build original runnable code
screenshot UI screenshot Preserve visible truth; build a runnable page and label assumptions

Process one page. Infer product purpose, primary user, route, and must-preserve behavior from available evidence. Ask one focused question only when a missing fact would materially change the implementation.

Do not copy third-party source, brand assets, copy, icons, imagery, or distinctive identity. Do not invent features, customers, testimonials, metrics, or hidden interactions.

Resolve the absolute directory containing this SKILL.md and assign it to REDESIGN_SKILL. Invoke bundled tools with node "$REDESIGN_SKILL/scripts/redesign.mjs"; never assume the user's working directory contains this Skill.

Create the brief

Run:

node "$REDESIGN_SKILL/scripts/redesign.mjs" brief \
  --mode project \
  --source /absolute/source/path \
  --page /target-route \
  --purpose "Concrete product purpose" \
  --user "Primary user" \
  --preserve "navigation, forms, real content" \
  --output ui-redesign

Change --mode and --source for url or screenshot. Review brief.json and record visible assumptions.

Establish evidence before editing

For project, inspect repository instructions and scripts, start the page, and test its primary behavior. For url, confirm authorization and inspect public visual and DOM evidence without taking source code.

For screenshot, preserve the original and create a non-fabricated baseline:

node "$REDESIGN_SKILL/scripts/redesign.mjs" ingest-screenshot \
  --source /absolute/path/interface.png \
  --output ui-redesign/before

Treat responsive and interaction behavior as unproven. The Review Board will label the repeated original image as Source, not as three captured breakpoints.

For every runnable source, capture mobile, tablet, and desktop simulators at 375, 768, and 1440 widths:

node "$REDESIGN_SKILL/scripts/redesign.mjs" capture \
  --url http://127.0.0.1:3000/target \
  --output ui-redesign/before \
  --channel chrome

Omit --channel chrome when Playwright Chromium is installed. Inspect the screenshots, console findings, accessibility results, and interaction-baseline.json. Exercise important controls; DOM presence alone does not prove behavior.

Audit and choose a direction

Run:

node "$REDESIGN_SKILL/scripts/redesign.mjs" audit \
  --manifest ui-redesign/before/capture-manifest.json \
  --output ui-redesign

Apply audit-rubric.md to extend audit.md. Every issue needs evidence, impact, a concrete change, and an acceptance criterion.

Use design-direction.md to derive one product-specific direction. Write selected-direction.md and design-tokens.json. Do not apply a fixed “premium” house style.

Implement, then verify

Follow implementation-playbook.md.

  • In project mode, preserve routes, data flow, component contracts, content truth, and recorded behaviors. Change the smallest coherent file set.
  • In url or screenshot mode, copy $REDESIGN_SKILL/assets/standalone-template/ to ui-redesign/implementation/, then replace its placeholders with an original, dependency-free page.

Implement in stages: tokens; structure; typography and color; components; responsive states; restrained motion. Test primary controls after each stage.

Capture the redesigned page to ui-redesign/after, then run:

node "$REDESIGN_SKILL/scripts/redesign.mjs" qa \
  --brief ui-redesign/brief.json \
  --before ui-redesign/before/capture-manifest.json \
  --after ui-redesign/after/capture-manifest.json \
  --interactions ui-redesign/before/interaction-baseline.json \
  --output ui-redesign

node "$REDESIGN_SKILL/scripts/redesign.mjs" review-board \
  --before ui-redesign/before/capture-manifest.json \
  --after ui-redesign/after/capture-manifest.json \
  --candidate-url http://127.0.0.1:4173 \
  --output ui-redesign/review-board

Open the Review Board and visually compare all three widths. Automated checks cannot judge hierarchy, product specificity, or whether the result still looks generated.

Completion gate

Read output-contract.md. Do not claim completion unless:

  • real code changes or a runnable standalone implementation exist;
  • before/after evidence exists at 375, 768, and 1440 widths for runnable sources; screenshot mode preserves the original source and captures the implementation at all three widths;
  • baseline controls and primary behaviors remain present and tested;
  • there is no horizontal overflow or new console error;
  • controls have accessible names, visible focus, and adequate touch targets;
  • contrast and reduced-motion checks pass;
  • audit.md, selected-direction.md, design-tokens.json, interaction-baseline.json, the Review Board, and qa-report.md exist.

Return fail for a critical behavior regression or unusable page. Return conditional pass for a runnable implementation with explicit non-critical assumptions or warnings. Make limitations visible.

Common mistakes

  • Treating gradients, cards, pills, or centered heroes as defects without product evidence.
  • Restyling before recording behavior and responsive evidence.
  • Copying a reference site instead of deriving an original direction.
  • Claiming success from a single desktop screenshot.
  • Hiding uncertain screenshot behavior instead of reporting assumptions.