Taste
Delivers a ruthless, structured critique of any screen, mockup or site — scores hierarchy, spacing rhythm, contrast, typography pairing and color discipline /10 with evidence, detects clichés (purple gradients, glassmorphism-by-default, emoji soup, cookie-cutter heroes), and ranks the top 5 highest-leverage fixes with before/after directions. Use when the user says "critique this design", "why does my site look cheap", "roast my landing page", or "does this look AI-generated?".From its SKILL.md
npx -y skills add alebgl77/claude-inc --skill tasteAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 8 stars8 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
4.0 KB, 888 tokens by cl100k_base, as published. Nobody here has run it
Taste — Taste Maker
"Design-taste critic"
When to use
- Something visual exists and needs honest judgment — "critique this screen", "review my mockup".
- The user senses mediocrity but cannot name it — "why does this look cheap / generic / off?"
- Pre-ship gate — "last look before launch"; run on every sibling skill's output.
- AI-slop suspicion — "does this look like every other AI-generated site?"
- Two directions compete — "which of these headers is stronger, and why?"
Workflow
- Inventory before judging. List what is actually on the screen: sections, type sizes in play, colors in play, competing focal points. Critique the artifact, not an imagined version.
- Score five axes, /10 each, with one sentence of pointable evidence per score.
- Hierarchy — does the eye land where the business needs it to land?
- Spacing rhythm — one consistent scale, or ad-hoc gaps everywhere?
- Contrast — figure/ground separation; AA compliance on text pairs.
- Typography pairing — pairing logic, scale discipline, line length 45–75ch.
- Color discipline — roles versus decoration; count the hues doing no work.
- Run cliché detection and flag every hit: purple/indigo gradient hero, glassmorphism-by-default, emoji soup as bullet decoration, the cookie-cutter centered hero with three feature cards, drop-shadow-on-everything, fake five-star social proof, Inter-on-white sameness, "Supercharge your workflow" copy.
- Name the missing idea. The one distinctive move this design lacks — or has and should push harder. A design with zero clichés can still be forgettable; say so.
- Rank the top 5 fixes by leverage (impact over effort). Each gets a before → after direction concrete enough to execute without a follow-up question.
- Deliver the verdict. Average the five axes, write the one-line headline, and make the call: SHIP at 7.0 or above, otherwise DO NOT SHIP.
Output format
# Taste Report — <artifact>
**Verdict: <x.x>/10 — <one-line headline>. <SHIP | DO NOT SHIP>.**
| Axis | /10 | Evidence |
|---|---|---|
| Hierarchy | 4 | CTA competes with three same-weight headings |
| Spacing rhythm | <n> | <evidence> |
| Contrast | <n> | <evidence> |
| Typography pairing | <n> | <evidence> |
| Color discipline | <n> | <evidence> |
**Strengths:** <what genuinely works — always name at least one>
**Clichés detected:** <list, or "none">
**The missing idea:** <the one distinctive move this design needs>
## Top 5 fixes, by leverage
1. <fix> — before: <current state> → after: <specific direction>
2. <next four fixes, same shape, descending leverage>
Quality bar
- Every score backed by pointable evidence, not vibes
- At least one genuine strength named — all-negative critique is noise
- Each of the five fixes is executable as written; zero "make it pop"
- Full cliché list checked, including the ones the critic personally enjoys
- Verdict includes an explicit SHIP / DO NOT SHIP call at the 7.0 threshold
Example
Invocation: "Roast my SaaS landing page." (screenshot attached)
Produces: A Taste Report scoring 5.2/10 — "Competent grid, zero point of view. DO NOT SHIP." — flagging an indigo-gradient hero and emoji feature bullets, crediting a genuinely strong pricing table, and ranking five fixes led by: collapse three competing H2s into one statement headline (before: three 24px headings shouting at once → after: a single 64px display line with an 18px muted subhead, everything else demoted).
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.