agentleFS
madebymustafamadebymustafa/design-taste/skills/taste-design-review/SKILL.md

taste-design-review

Skill1 starsChanged 3 months agoMITView on GitHub

Structured visual and usability audit of existing interfaces — landing pages, SaaS apps, portfolios. Screenshot-first review against taste rules, Nielsen heuristics, and accessibility floor; severity-ranked findings with specific fixes; before/after verification. Use when asked to critique, review, or improve an existing UI.

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 4 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Never fix while reviewing.

    line 20
  2. Distinctive:

    Recognition over recall: no invented jargon; navigation names are self-explanatory.

    line 45
  3. Distinctive:

    Errors communicated well: every error states what happened, why, and what to do — never just "Error".

    line 48
  4. Distinctive:

    Every input labeled; every button/link has text.

    line 54

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

taste-design-review

Audit like a design director: screenshot first, then list findings by severity, then fix only what earns it.

1. THE PROCESS (Audit → Rank → Fix → Verify)

Never fix while reviewing. The pass order:

  1. Capture: full-page screenshots of every viewport (desktop 1440, tablet 768, mobile 390). Review what is, not what you think is.
  2. First read (8 seconds): what does this page say it is? Can you name the product, audience, and primary action without scrolling?
  3. Structured pass: apply the checks below, in order.
  4. Rank findings: P0 (broken/legal/blocking) / P1 (seriously undermines quality) / P2 (polish) / P3 (preference).
  5. Fix in ranked order with atomic changes; re-screenshot and verify each fix didn't break the design argument.
  6. Report: findings table with screenshot references, severity, and the one-sentence fix for each.

2. THE TASTE PASS (Visual Quality)

  • The kill question: could this page have been generated by a default agent (slop patterns: purple gradients, centered hero + 3 cards, emoji icons, Inter-only, rounded-2xl, browser-chrome screenshot)? Every "yes" is a P1.
  • Hierarchy: one dominant element per viewport; does the eye land where the business wants it?
  • Type: any text under 10px, more than 3 weights, no display contrast, body full-width, 6-line paragraphs, widows in headings.
  • Color: more than ~3 meaningful colors, gradients doing decoration, accent used >4 times per viewport, white text on pale fills, contrast failures (exact: 4.5:1 body / 3:1 large / 3:1 UI).
  • Alignment: elements sharing an edge line up (check visually); hairline grid vs. floating cards scattered; spacing scale integrity (any 7px or 13px paddings?).
  • Depth: shadows as the only depth source; glass everywhere; cards inside cards inside cards.

3. THE HEURISTIC PASS (Usability — Nielsen's 10, Applied)

  1. Visibility of system status (2–3s): every action produces feedback; no dead buttons; loading states exist.
  2. Match with the real world: language is the user's, not the company's ("deploy instance" on a consumer site = fail).
  3. User control & freedom: exit routes everywhere; undo exists for destructive acts.
  4. Consistency: same action looks the same everywhere; one style of button; one way to reach the same destination.
  5. Error prevention: destructive actions confirmed; submit buttons guarded; autocomplete on forms.
  6. Recognition over recall: no invented jargon; navigation names are self-explanatory.
  7. Flexibility: shortcuts for repeat users (search, keyboard) without cluttering the novice path.
  8. Aesthetic & minimalist: anything that hides the primary action is noise. If an element doesn't serve a job, it's deleted in the report.
  9. Errors communicated well: every error states what happened, why, and what to do — never just "Error".
  10. Help & documentation: help exists where users get stuck, not in a help center nobody visits.

4. THE ACCESSIBILITY PASS (Short Version — See taste-accessibility)

  • Tab through: focus visible at every stop; no traps; logical order.
  • Every input labeled; every button/link has text; alt text is meaningful or empty.
  • Contrast spot-check on every distinct surface (heading/body/button/placeholder).
  • prefers-reduced-motion handled; anything autoplaying or auto-advancing paused.

5. THE CONVERSION PASS (If Asked to Improve, Not Just Audit)

  • One primary CTA above the fold; secondary actions visually quieter or deferred.
  • The value proposition lands in ≤ 8 seconds (headline does it, not the sub-sub-line).
  • Friction inventory: fields, steps, and clicks between "interested" and "started" — each one justified.
  • Trust signals exist where decisions happen (near CTA, in pricing): social proof, guarantees, no-fine-print.

6. REPORT FORMAT (Fixed)

## Audit — <page> (<viewport set>)

### P0 — Blocking
- [screenshot ref] finding + one-sentence fix

### P1 — Quality
...

### P2 — Polish
...

### P3 — Preference
...

### Verdict (one paragraph)
Is the design argument clear? What single change would improve the page most?

Severity discipline: P3 items are listed but never fixed in the same session as P1s. The verdict names ONE change — if you can't name one, the audit didn't find the real issue.

7. CHECKLIST (Fail the Audit If…)

  • No screenshots were taken/reviewed (never audit from code alone).
  • Findings lack severities or screenshot references.
  • Slop patterns dismissed as "style preference" (they're P1).
  • Anything on the page was fixed before the full audit pass completed.
  • The verdict doesn't name a single most-important change.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.