visual-polish-loop
/visual-polish-loop
Improve an existing web interface through a bounded screenshot-driven loop: capture a reproducible baseline, fix the highest-impact visual issues, and verify the same states on desktop and mobile. Use when a working UI needs polish or responsive cleanup; do not use for net-new product design or broad feature development.
- Author
- qml780 ↗
- Repository
- visual-polish-loop-skill ↗
- Licence
- MIT
- Last repository change
- 06.11.2026
Assessment and security
- Files in this package
- 3
- Scripts
- 0
- HTML comments
- 0
- URLs in SKILL.md
- 0
No signals detected. Repository changed on 06.11.2026, without references to retired models or APIs.
MIT, open: use and modify while retaining author attribution.
What it does and when to use it
Improve an existing web interface through a bounded screenshot-driven loop: capture a reproducible baseline, fix the highest-impact visual issues, and verify the same states on desktop and mobile. Use when a working UI needs polish or responsive cleanup; do not use for net-new product design or broad feature development.
Original author description
Improve an existing web interface through a bounded screenshot-driven loop: capture a reproducible baseline, fix the highest-impact visual issues, and verify the same states on desktop and mobile. Use when a working UI needs polish or responsive cleanup; do not use for net-new product design or broad feature development.
How it works
- Establish the baseline
- Choose the polish pass
- Verify the delta
- Stop conditions
- Handoff
SKILL.md preview
--- name: visual-polish-loop description: "Improve an existing web interface through a bounded screenshot-driven loop: capture a reproducible baseline, fix the highest-impact visual issues, and verify the same states on desktop and mobile. Use when a working UI needs polish or responsive cleanup; do not use for net-new product design or broad feature development." --- # Visual Polish Loop Turn “make it look better” into a short, repeatable visual QA pass with before-and-after evidence. ## Establish the baseline Before editing, identify the smallest route or screen in scope and make it reproducible: - Start the app using its existing workflow. - Use stable demo data or a deterministic state; do not compare different content. - Capture one desktop viewport and one narrow/mobile viewport. Record their dimensions. - Exercise the primary interaction once and record console/page errors, horizontal overflow, and broken states. - Preserve the original screenshots outside generated build folders so the final comparison is honest. If the user supplied a reference image, treat it as direction for hierarchy, density, rhythm, and tone. Do not silently replace the product’s identity or copy unrelated branded assets. ## Choose the polish pass Review the baseline through the lenses in [references/review-lenses.md](references/review-lenses.md). Rank issues by perceptual impact on the intended task, then select at most three for one pass. Good candidates usually affect the whole experience: unclear focal hierarchy, inconsistent spacing, weak typography contrast, cramped responsive layout, missing interaction feedback, or accidental visual noise. Prefer shared tokens and existing component patterns over one-off overrides. Preserve working behavior and accessibility semantics. Do not introduce a new UI framework, font service, animation library, or design system merely to polish one screen. ## Verify the delta After editing, replay the exact baseline state and recapture the same viewport sizes. The pass succeeds only when: - the selected visual issues are materially improved in the new screenshots; - the main interaction still works; - the console has no new errors; - the narrow viewport has no page-level horizontal overflow; - focus, disabled, loading, and error states touched by the change remain understandable; - relevant tests or checks pass. If the result is merely different, revert or refine it. Do not call a style change an improvement without visible evidence. ## Stop conditions - Stop after two polish passes or when the three highest-impact issues are resolved. - Do not expand into navigation redesign, data-model work, new features, rebranding, or content strategy unless the user explicitly adds that scope. - Do not chase pixel differences caused by font rasterization, timestamps, randomized data, caret blinking, or animation frames; stabilize or mask them first. - If browser access or a runnable app is unavailable, report the blocked evidence instead of inventing a visual result. ## Handoff Lead with the visible outcome. Provide before/after screenshots or paths, viewport sizes, the three issues addressed, interaction and console results, tests run, and any visual risks deliberately left unchanged.
Files in this package (3)
- SKILL.md 3 KB
- agents/openai.yaml 1 KB
- references/review-lenses.md 2 KB
Versions
| Repository | Score | Licence |
|---|---|---|
| This page · qml780/visual-polish-loop-skill | 100 | MIT |
| FlintSH/tinderbox | 100 | MPL-2.0 |
Database snapshot: 13.11.2026 · package version 1fb58df0bd8e (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 100ux-ui-audit-loop
Audit and improve a web UI in a browser through evidence-backed capture, severity triage, minimal code fixes, and before/after verification across responsive and interaction states. Use when the user says "audit this UI", "fix the UX", "review this page visually", "make this… - 100ux-ui-audit-loop
Audit and improve a web UI in a browser through evidence-backed capture, severity triage, minimal code fixes, and before/after verification across responsive and interaction states. Use when the user says "audit this UI", "fix the UX", "review this page visually", "make this… - 90ui-slop-score
Review a rendered web or mobile interface and score how generic it looks. Use for UI critique, design review, visual polish, screenshot review, pre-merge checks, and requests to fix AI-looking UI or explain why an interface feels generic. - 95interface-polish
Audit and improve an existing web app's UI and UX while preserving its design language. Use for polish passes, inconsistent typography or icons, cramped panes, weak contrast, awkward controls, snapping menus, and requests to make a shipped interface feel more considered. Covers… - 90interface-design-refinement
Critique and refine an existing web UI within its authorized visual and functional scope. Use for awkward hierarchy, generic AI-looking pages, inconsistent visual treatment, or explicit UI polish. - 100ds-design-audit
Reviews an existing UI with scores based on screenshot evidence, then fixes the highest-impact problems and re-scores. Use when a page or app screen looks generic, cluttered, or unfinished, or before you ship UI work. Run as /ds-design-audit.