# UI Template Audit

> Audit an existing UI, landing page, or screenshot for "template defaults" — the patterns that make interfaces read as generic or machine-generated — and produce a scored report with specific fixes. Use this skill when reviewing a page that "looks like every other AI-built site", before shipping a landing page, or when asked to make a design feel less generic. Activate when: generic UI, looks AI-generated, template look, design review, landing page critique, make it less generic, design audit, AI slop design.

- Skill: `latestaiagents/ui-template-audit` (Agent Skill)
- Install (CLI): `npx get-skill add latestaiagents/ui-template-audit`
- Raw SKILL.md: https://api.skillmd.com/api/skills/latestaiagents/ui-template-audit/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Security
- Author: latestaiagents (https://skillmd.com/u/latestaiagents)
- Updated: 2026-11-14
- Page: https://skillmd.com/skills/latestaiagents/ui-template-audit

---


# UI Template Audit

**Find the defaults a page inherited without deciding on them, score how much they dominate, and replace each one with a choice that comes from the product.**

A "template default" is not a forbidden pattern. A numbered list is fine when order matters; a logo row is fine when the logos are real customers who agreed to be shown. The problem is a pattern used because it was the first thing that came to hand, not because the content needed it. This audit is judgment, not law: record why each finding matters for *this* page.

## When to Use

- A page was generated quickly (by an agent, a starter kit, or a UI kit) and now needs a review
- Stakeholders say the site "feels generic" but can't say why
- Before a launch, as a final pass after functionality works
- When building a design system, to decide which defaults to keep on purpose

## When NOT to Use

- The brief explicitly asks for a conventional or familiar look (internal admin tools, government forms, a pastiche) — familiarity is a feature there
- The page is not built yet — use `visual-direction` first
- You only need an accessibility check — use `accessible-by-default`

## Relationship to Other Skills

Anthropic's official `frontend-design` skill guides *building* distinctive UI (aesthetic direction, typography, plan-review-build-critique). This skill is the *audit* counterpart: a repeatable checklist for an existing page, with a score and a fix list. Use `frontend-design` for generation and this skill for review; they do not conflict.

## Workflow

### Step 1: Gather the Page and the Brief

Collect, before judging anything:

1. The rendered page (screenshot at desktop and ~375px mobile width) or the source
2. What the product is, who uses it, and what one action the page should drive
3. Any brand assets that already exist (logo, product screenshots, real photos, real customer names)

Without the brief you can only detect sameness, not decide what should replace it.

### Step 2: Run the Checklist

Mark each item **Present**, **Partly**, or **Absent**. Note the exact selector, section, or line.

| # | Template default | What to look for |
|---|------------------|------------------|
| 1 | Default gradient palette | Purple/indigo/blue gradients on hero or buttons, especially on white or near-black, with no link to the brand |
| 2 | Eyebrow labels everywhere | Small, tracked, uppercase (often monospace) labels above every section heading |
| 3 | Decorative numbering | "01 / 02 / 03" on items whose order carries no meaning |
| 4 | Sloganeering headline | Headline that would fit any product ("Build faster. Ship smarter.", "The future of X") |
| 5 | Three equal feature cards | A row of three same-sized cards, icon + title + two lines, repeated down the page |
| 6 | Logo wall without consent or context | "Trusted by" row of logos that are placeholders, unverified, or unexplained |
| 7 | Unverifiable statistics | "10x faster", "99.9% satisfaction", "10,000+ teams" with no source |
| 8 | Stock testimonials | Quotes with first name only, generic avatar, no company or link |
| 9 | Emoji or icon bullets | Emoji or generic line icons used as list bullets or feature markers |
| 10 | Decorative arrows | "Learn more →" arrows added to every link or button |
| 11 | Glassmorphism by default | Blurred translucent panels over a gradient blob, not motivated by layered content |
| 12 | Gradient text | Headline words filled with a gradient as the main emphasis device |
| 13 | Single-word emphasis | One word in a headline highlighted, underlined, or colored as a reflex |
| 14 | Uniform rounding and shadow | Every element shares one large radius and one soft shadow |
| 15 | Centered everything | All sections centered, same max-width, same vertical rhythm |
| 16 | Generic hero illustration | Abstract 3D shapes, isometric people, or blobs instead of the actual product |
| 17 | Motion on everything | Fade-up on scroll for every block; no `prefers-reduced-motion` handling |
| 18 | Default font stack as identity | One sans-serif at default weights carrying everything, nothing chosen |
| 19 | Placeholder copy patterns | "Seamless", "powerful", "all-in-one", "supercharge", "unlock" |
| 20 | FAQ and pricing boilerplate | Three tiers with "Most popular" badge regardless of the actual offer |

### Step 3: Score

Score each item: Absent = 0, Partly = 1, Present = 2. Weight the items that most affect trust and comprehension:

| Weight | Items |
|--------|-------|
| x3 (trust) | 6, 7, 8 |
| x2 (identity) | 1, 4, 16, 18, 19 |
| x1 (style) | everything else |

Interpret the total as a priority signal, not a grade:

| Total | Reading |
|-------|---------|
| 0-8 | Mostly intentional; fix individual items |
| 9-20 | Noticeably templated; fix the trust and identity items first |
| 21+ | The page has no direction yet; run `visual-direction` before patching |

The thresholds are a working heuristic for prioritising, not a validated metric. Say so if you report the score.

### Step 4: Write the Fix for Each Finding

Every finding needs a replacement that comes from the product. "Remove the gradient" is incomplete; say what goes there instead.

| Finding | Weak fix | Specific fix |
|---------|----------|--------------|
| Sloganeering headline | Rewrite to be punchier | State what it does for whom: "Turn court bookings into a schedule your members can see" |
| Unverifiable stat | Soften the number | Remove it, or replace with a fact you can source ("Used at 14 clubs in Pune" if true and approved) |
| Three equal cards | Make them four | Lead with the single most important capability at full width; list the rest compactly |
| Eyebrow labels | Make them smaller | Delete them; let the heading carry the section. Keep a label only where it adds information (a date, a status) |
| Generic hero illustration | Change the illustration | Show the product: a real screenshot, a short loop of the core flow, or a real photo |
| Decorative numbering | Use letters instead | Remove numbers unless the steps must happen in order |
| Logo wall | Fewer logos | Keep only logos you have permission to show; add one sentence on how each uses the product, or drop the row |

### Step 5: Report

Use this structure:

```markdown
## Template audit: <page>

Brief: <product, audience, one goal>
Score: <n> (heuristic; trust items weighted x3)

### Trust (fix before launch)
- [7] Hero stat "10x faster" (hero.tsx:42) has no source. Remove, or cite the measurement.

### Identity
- [4] Headline "The future of scheduling" fits any product. Proposed: "<specific line>".

### Style
- [2] Uppercase eyebrow above all 7 section headings. Delete; keep the date label on Changelog.

### Keep (intentional)
- Numbered steps in "How setup works": order matters here.
```

Always include a **Keep** section. An audit that strips everything produces a different template, not a better page.

## Example: Before and After

```html
<!-- Before -->
<section class="hero bg-gradient-to-r from-indigo-500 to-purple-600 text-center">
  <span class="eyebrow">AI-POWERED PLATFORM</span>
  <h1>Supercharge your <span class="gradient-text">workflow</span></h1>
  <p>The all-in-one platform trusted by 10,000+ teams.</p>
  <a href="/signup">Get started →</a>
</section>

<!-- After (for a hypothetical invoicing tool for freelance translators) -->
<section class="hero">
  <h1>Invoice by word count, in the client's currency</h1>
  <p>Paste a project's word count and rate; get a PDF invoice with the
     exchange rate you used printed on it.</p>
  <a href="/signup">Create your first invoice</a>
  <img src="/screens/invoice-from-wordcount.png"
       alt="Invoice form with 4,200 words at 0.08 EUR per word, total 336 EUR">
</section>
```

The "after" is not more decorated; it is more specific. Visual styling then follows from the direction, not from a kit.

## Best Practices

1. Audit against the brief, not against a list of banned styles
2. Fix trust items (fake stats, unconsented logos, stock testimonials) first; they carry legal and credibility risk
3. Pair every removal with a specific replacement
4. Record intentional choices in a Keep section so the next pass doesn't undo them
5. Check mobile: many template patterns (equal cards, centered stacks) look worse at narrow widths
6. Re-run the checklist after fixes; replacing one default with another is common
7. Run `accessible-by-default` as a separate pass; distinctive is never a reason to drop accessibility

## References

- Anthropic `frontend-design` skill: https://github.com/anthropics/skills/blob/main/skills/frontend-design/SKILL.md
- `frontend-design` Claude Code plugin (official marketplace): https://github.com/anthropics/claude-plugins-official/tree/main/plugins/frontend-design
- MDN, `prefers-reduced-motion`: https://developer.mozilla.org/en-US/docs/Web/CSS/@media/prefers-reduced-motion

