Coding agents can build anything. These skills make it look designed.
Small, sharp design skills for AI agents — encoded from 19+ years of product design leadership in financial services, by Ravi Palwe. Senior judgment, not generic guidelines: findings come with receipts (computed contrast, named principles) and every critique carries a fix.
npx skills add ravipalwe/design-skillsOn the marketplace: skills.sh/ravipalwe/design-skills
Works with any agent that reads SKILL.md — Claude Code, Cursor, Windsurf, Amp, Devin, and friends.
| Skill | What it does | Status |
|---|---|---|
| roast-my-ui | A senior design crit on demand. Screenshot, URL, Figma link, or code in → ranked roast out: what's hurting the design, why it matters, the exact fix. Contrast violations are computed (WCAG 2.1 + APCA), never guessed. | ✅ live |
| de-slop | Detect and fix the AI-generated look — the purple gradients, the glass cards, the template scroll. | 🔜 shipping |
| fix-contrast | Full computed contrast audit of your codebase's colors, with auto-fix suggestions. | 🔜 shipping |
| ship-dark-mode | Derive a designed dark palette from your light tokens — elevation by lightness, not inversion. | 🔜 shipping |
| microcopy | Rewrite your buttons, errors, and empty states in product voice. Words are the cheapest design lever. | 🔜 shipping |
| fintech-ux | Design judgment for money products — trust architecture, financial-anxiety patterns, money display craft, compliance as a design material (KYC, Reg E, TILA). Money-display violations are computed by a bundled lint (tabular figures, U+2212 negatives, fee contrast), never guessed. Encoded from 19 years in financial services. | ✅ live |
Install, then:
Roast my UI — here's a screenshot.
You get: a one-line verdict, the three things a senior can't unsee, the full crit grouped by theme with 🔥 severity ratings and computed contrast receipts, what's genuinely working, and a fix list ordered by impact-per-effort.
- Receipts, not vibes. Contrast claims run through a bundled zero-dependency script (WCAG 2.1 gate + APCA advisory), and fintech-ux ships a money-display lint that finds violations in your actual code,
file:line. A finding with computed evidence is unanswerable. - Every jab carries a fix — with exact values where possible, not "consider improving the hierarchy."
- The anti-slop stance. These skills actively steer away from the recognizable AI-median look and toward one memorable signature per product.
- A senior's calibration. Severity scales to product stage; good work gets named, not buried under invented findings.
Two flagship skills live in their own repos:
- designos — a complete design operating system: idea → direction → design system → screens → handoff, with gates and verified tokens.
- prism — UX research + UI design deliverables: personas, journey maps, interview synthesis, design tokens, component specs. Benchmarked at +64% methodology quality.
- Bundled scripts are pure stdlib Python — no network calls, no dependencies to audit.
- Like every agent skill, these run with your agent's permissions. It's all readable markdown — read it before you run it.
MIT © Ravi Palwe