Perf Expert

Name: Perf Expert
Author: hardikpandya

hardikpandya/perf-analyzer

7 installs
13 repo stars
Updated January 26, 2026
hardikpandya/perf-analyzer

Audits a website for Core Web Vitals, accessibility, and SEO issues using Lighthouse and delivers a prioritized, actionable fix plan with file paths.

About

A frontend performance skill that runs Lighthouse, identifies the stack, and categorizes issues by severity with expected impact. A developer uses it to improve LCP, INP, CLS, accessibility, and SEO scores on a site.

Establishes Lighthouse baselines and targets for LCP, INP, CLS, TTFB
Focus modes for performance, accessibility, or SEO

Perf Expert by the numbers

7 all-time installs (skills.sh)
Ranked #852 of 1,352 Code Review & Quality skills by installs in the Skillselion catalog
Data as of Aug 2, 2026 (Skillselion catalog sync)

npx skills add https://github.com/hardikpandya/perf-analyzer --skill perf-expert

Add your badge

Show developers this skill is listed on Skillselion. Paste this into your README.

[![Listed on Skillselion](https://skillselion.com/badge/skills/hardikpandya/perf-analyzer/perf-expert.svg)](https://skillselion.com/skills/hardikpandya/perf-analyzer/perf-expert)

Installs	7
repo stars	★ 13
Last updated	January 26, 2026
Repository	hardikpandya/perf-analyzer ↗

What it does

Audits a website for Core Web Vitals, accessibility, and SEO issues using Lighthouse and delivers a prioritized, actionable fix plan with file paths.

Files

SKILL.mdMarkdownGitHub ↗

Performance expert

You are a senior frontend performance engineer. When this skill is invoked, conduct a comprehensive audit and deliver an actionable improvement plan.

What you deliver

Every audit must include:

1. Current state assessment — Run Lighthouse, identify the tech stack, note existing optimizations 2. Issues found — Categorized by severity (critical, moderate, minor) 3. Actionable fixes — Specific code changes with file paths and line numbers 4. Expected impact — What each fix improves and by how much 5. Priority order — What to fix first for maximum impact

Usage

/perf-expert              # Full audit
/perf-expert performance  # Performance focus
/perf-expert a11y         # Accessibility focus
/perf-expert seo          # SEO focus

Audit methodology

Step 1: Establish baselines

Run Lighthouse and note current scores:

npx lighthouse https://yoursite.com --output json --output-path baseline.json

Extract the metrics that matter:

Metric	Target	Google ranking factor?
LCP (Largest contentful paint)	< 2.5s	Yes
INP (Interaction to next paint)	< 200ms	Yes
CLS (Cumulative layout shift)	< 0.1	Yes
TTFB (Time to first byte)	< 800ms	No, but affects LCP

Step 2: Identify the stack

Check for package.json, _config.yml, next.config.js, vite.config.js. Each framework has different bottlenecks and optimization paths.

Step 3: Audit each category

Performance — Scripts, fonts, images, bundle size. See references/performance.md.

Accessibility — Focus states, skip links, semantic HTML, keyboard navigation. See references/accessibility.md.

SEO — Title tags, meta descriptions, robots.txt, canonical URLs.

Browser compatibility — Safari bugs, minifier issues. See references/browser-gotchas.md.

Step 4: Deliver the report

Use the format in references/report-template.md. Always include:

What you found (with evidence)
How to fix it (with code)
Why it matters (impact on metrics)
Priority order (what to fix first)

Quick reference

Critical issues (fix immediately)

Render-blocking scripts without defer
Fonts preloaded as wrong format (OTF declared as WOFF2)
Missing image dimensions causing layout shift
outline: none without :focus-visible replacement
Missing skip link

High impact fixes

Fix	Typical improvement
Add `defer` to scripts	-200-500ms to TTI
Switch to WOFF2 fonts	-30% font size
Add `font-display: swap`	Eliminates invisible text
Exclude unused fonts	Often 10-70MB saved
Add image dimensions	CLS drops to near zero

Commands to run

# Full Lighthouse audit
npx lighthouse https://site.com --output html --view

# Scan entire site
npx unlighthouse --site https://site.com

# Accessibility audit
npx pa11y https://site.com

# Check build size
du -sh _site/ dist/ build/ .next/ 2>/dev/null

Example output format

Your audit report should look like this:

## Performance audit for example.com

### Current scores
- Performance: 67
- Accessibility: 82
- Best practices: 92
- SEO: 100

### Critical issues

#### 1. Render-blocking scripts (Performance: -15 points)

**Problem**: 3 scripts in `<head>` block rendering.

**Files**:
- `_includes/head.html:12` — analytics.js
- `_includes/head.html:15` — app.js
- `_includes/head.html:18` — utils.js

**Fix**:

<script src="/js/app.js"></script>

<script src="/js/app.js" defer></script>


**Impact**: Estimated +10-15 performance points, -300ms to FCP.

---

#### 2. Font preload MIME mismatch (Best practices: -8 points)

**Problem**: Preloading OTF files but declaring as WOFF2.

**File**: `_includes/head.html:8`

**Fix**:

<link rel="preload" href="/fonts/body.otf" as="font" type="font/woff2">

<link rel="preload" href="/fonts/body.woff2" as="font" type="font/woff2" crossorigin>


**Impact**: Eliminates console warning, proper font caching.

---

### Recommended priority

1. Add defer to scripts (quick win, big impact)
2. Fix font preloads (prevents double downloads)
3. Add image dimensions (fixes CLS)
4. Implement skip link (accessibility requirement)

### Expected results after fixes
- Performance: 67 → 85-90
- Accessibility: 82 → 95+
- CLS: 0.15 → < 0.05

License

MIT

MIT License

Copyright (c) 2026 Hardik Pandya

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

perf-analyzer

A Claude Code skill that delivers comprehensive frontend performance audits with actionable improvement plans.

This isn't a checklist. It's a methodology for diagnosing performance issues and fixing them systematically.

What you get

When you invoke /perf-expert, you receive:

1. Current state assessment with Lighthouse scores and Core Web Vitals 2. Issues found categorized by severity (critical, moderate, minor) 3. Specific fixes with file paths, line numbers, and code changes 4. Expected impact for each fix 5. Priority order for maximum impact with minimum effort

Skill structure

perf-expert/
├── SKILL.md                  # Core instructions (~180 lines)
├── references/
│   ├── performance.md        # Scripts, fonts, images, bundles
│   ├── accessibility.md      # Focus, ARIA, keyboard, contrast
│   ├── browser-gotchas.md    # Safari bugs, minifier issues
│   └── report-template.md    # Audit summary format
├── README.md
└── LICENSE

Installation

Claude Code: Add this folder as a skill.

mkdir -p .claude/skills/perf-expert
curl -o .claude/skills/perf-expert/SKILL.md \
  https://raw.githubusercontent.com/hardikpandya/perf-analyzer/main/SKILL.md

Download reference files for deep dives:

mkdir -p .claude/skills/perf-expert/references
curl -o .claude/skills/perf-expert/references/performance.md \
  https://raw.githubusercontent.com/hardikpandya/perf-analyzer/main/references/performance.md
curl -o .claude/skills/perf-expert/references/accessibility.md \
  https://raw.githubusercontent.com/hardikpandya/perf-analyzer/main/references/accessibility.md
curl -o .claude/skills/perf-expert/references/browser-gotchas.md \
  https://raw.githubusercontent.com/hardikpandya/perf-analyzer/main/references/browser-gotchas.md
curl -o .claude/skills/perf-expert/references/report-template.md \
  https://raw.githubusercontent.com/hardikpandya/perf-analyzer/main/references/report-template.md

Claude Projects: Upload SKILL.md and reference files to project knowledge.

API calls: Include SKILL.md in your system prompt. Reference files load on demand.

Usage

/perf-expert              # Full audit
/perf-expert performance  # Performance focus
/perf-expert a11y         # Accessibility focus
/perf-expert seo          # SEO focus

What makes this different

It delivers actionable fixes, not vague advice

Generic advice like "optimize images" is useless. This skill tells you:

Which files to change and what line numbers
The exact code to add or remove
How much improvement to expect

It's battle-tested

Every recommendation comes from real production issues:

The Safari bug where CSS variables in @keyframes silently fail
The Jekyll minifier that breaks calc() expressions
The invisible CLS from web fonts that only shows on slow 3G

It knows the targets

Metric	Good	Needs work	Poor
LCP	< 2.5s	2.5-4s	> 4s
INP	< 200ms	200-500ms	> 500ms
CLS	< 0.1	0.1-0.25	> 0.25
TTFB	< 800ms	800ms-1.8s	> 1.8s

Lighthouse scores are vanity metrics. Core Web Vitals are what Google ranks you on.

Files

File	Purpose
`SKILL.md`	Core instructions and audit methodology
`references/performance.md`	Scripts, fonts, images, bundle optimization
`references/accessibility.md`	Focus states, ARIA, keyboard navigation
`references/browser-gotchas.md`	Safari bugs, minifier issues, edge cases
`references/report-template.md`	Audit report format

Real results

Developed while optimizing hvpandya.com:

Font bundle: 74MB → 3MB (excluded 48 unused font families)
LCP: 4.2s → 2.3s (font preloading + font-display: swap)
Accessibility score: 88 → 98 (focus states, ARIA labels, keyboard nav)
Fixed: Safari animation bug that silently broke emoji reactions

When to skip this

Prototypes: Ship fast, optimize later
Internal tools: User tolerance is higher
MVP validation: Learn first, polish second

Performance optimization has diminishing returns. Know when good enough is good enough.

Contributing

Found a technique that's not covered? Discovered a browser quirk? Open a PR.

The best contributions include:

The problem you encountered
Why the obvious solution didn't work
What actually fixed it
How to verify the fix

License

MIT

Author

Hardik Pandya

Accessibility deep dive

Accessibility isn't charity. It's good engineering. Accessible sites are more usable for everyone.

Focus states: the keyboard user's lifeline

The problem: Designers remove focus outlines because they're "ugly." Keyboard users can't see where they are.

Find the anti-pattern:

grep -r "outline.*none" --include="*.css"

The fix:

/* Delete this */
/* *:focus { outline: none; } */

/* Add this */
:focus-visible {
  outline: 2px solid #3b82f6;
  outline-offset: 2px;
}

/* Optional: explicitly remove for mouse users */
:focus:not(:focus-visible) {
  outline: none;
}

Why `:focus-visible`? It only shows for keyboard navigation, not mouse clicks.

---

Skip links: respect your users' time

Imagine tabbing through a 50-item navigation on every page.

The fix:

<body>
  <a href="#main-content" class="skip-link">Skip to main content</a>
  <header><!-- nav items --></header>
  <main id="main-content">...</main>
</body>

.skip-link {
  position: absolute;
  top: -100%;
  left: 16px;
  padding: 8px 16px;
  background: #000;
  color: #fff;
  z-index: 9999;
}
.skip-link:focus { top: 16px; }

---

Semantic HTML: divs are not buttons

The problem: <div onclick="..."> looks like a button but isn't one. Screen readers don't announce it. Keyboard users can't reach it.

Find offenders:

grep -r "div.*onclick\|span.*onclick" --include="*.html" --include="*.jsx"
grep -r 'role="button"' --include="*.html"

The fix:

<!-- Before -->
<div class="btn" onclick="submit()">Submit</div>

<!-- After -->
<button type="button" onclick="submit()">Submit</button>

Don't	Do
`<div>` with click handler	`<button>`
`<div>` for navigation	`<nav>`
`<div>` for main content	`<main>`
`<span>` for links	`<a href="...">`

---

ARIA: the last resort

ARIA is for when HTML semantics aren't enough. Native elements are always better.

Icon-only buttons need labels:

<!-- Bad: screen reader hears "button" -->
<button><svg><!-- X icon --></svg></button>

<!-- Good: screen reader hears "Close dialog, button" -->
<button aria-label="Close dialog"><svg><!-- X icon --></svg></button>

Dynamic content needs live regions:

<div aria-live="polite" aria-atomic="true">
  <span class="count">5 reactions</span>
</div>

---

Keyboard navigation

Test manually: Tab through your entire page. Can you reach everything? Can you activate it?

Custom elements need keyboard handlers:

element.addEventListener('keydown', (e) => {
  if (e.key === 'Enter' || e.key === ' ') {
    e.preventDefault();
    handleClick();
  }
});

---

Focus traps for modals

When a modal opens, Tab should cycle within it.

function trapFocus(modal) {
  const focusable = modal.querySelectorAll(
    'button, [href], input, select, textarea, [tabindex]:not([tabindex="-1"])'
  );
  const first = focusable[0];
  const last = focusable[focusable.length - 1];

  modal.addEventListener('keydown', (e) => {
    if (e.key === 'Tab') {
      if (e.shiftKey && document.activeElement === first) {
        e.preventDefault();
        last.focus();
      } else if (!e.shiftKey && document.activeElement === last) {
        e.preventDefault();
        first.focus();
      }
    }
    if (e.key === 'Escape') closeModal();
  });

  first.focus();
}

let previousFocus;
function openModal() {
  previousFocus = document.activeElement;
  modal.hidden = false;
  trapFocus(modal);
}
function closeModal() {
  modal.hidden = true;
  previousFocus?.focus();
}

---

Headings: the screen reader's table of contents

Screen reader users navigate by headings. Skipping levels is like a book with missing chapters.

Rules:

One <h1> per page
Never skip levels (h1 → h3 is wrong)
Use headings for structure, not styling

<!-- Correct -->
<h1>Article title</h1>
  <h2>Section</h2>
    <h3>Subsection</h3>

<!-- Wrong -->
<h1>Article title</h1>
  <h3>Section</h3>

---

Color contrast

WCAG AA requires:

4.5:1 for normal text
3:1 for large text (18px+ or 14px+ bold)
3:1 for UI components

Common failures:

Placeholder text (#999 on white = 2.85:1)
Muted text (#868e96 on white = 3.32:1)

Use browser DevTools color picker or WebAIM Contrast Checker.

Browser gotchas

Real bugs from production that will waste your time if you don't know about them.

Safari: CSS variables in @keyframes

The bug: CSS custom properties don't work in @keyframes in Safari. It silently fails.

/* This silently fails in Safari */
@keyframes fly {
  to { transform: translate(var(--end-x), var(--end-y)); }
}

The fix: Use Web Animations API instead:

element.animate([
  { transform: 'translateY(0)' },
  { transform: `translateY(${endY}px)` }
], {
  duration: 500,
  easing: 'ease-out',
  fill: 'forwards'
});

---

CSS minifiers breaking calc()

The bug: Some minifiers remove spaces in calc() expressions with custom properties, breaking them.

/* Before minification */
padding: calc(var(--spacing) * 2);

/* After aggressive minification (broken) */
padding: calc(var(--spacing)*2);

The fix: Test your production build, not just dev. Or configure your minifier to preserve spaces in calc().

---

MIME type mismatches

The bug: You declare a font as WOFF2 but the file is actually OTF. Browser gets confused, sometimes fails silently.

<!-- Wrong: file is OTF, type says WOFF2 -->
<link rel="preload" href="/fonts/body.otf" as="font" type="font/woff2" crossorigin>

<!-- Correct: match type to actual file -->
<link rel="preload" href="/fonts/body.woff2" as="font" type="font/woff2" crossorigin>

---

Mixed content warnings

The bug: HTTPS page loads HTTP resource. Browser may block it or show warnings.

Find it:

grep -r "http://" --include="*.html" --include="*.css" --include="*.js" | grep -v localhost

The fix: Change all URLs to https:// or use protocol-relative //.

---

Preload without crossorigin

The bug: Font preload without crossorigin attribute won't match the actual font request, causing a double download.

<!-- Wrong: will download font twice -->
<link rel="preload" href="/fonts/body.woff2" as="font" type="font/woff2">

<!-- Correct: add crossorigin even for same-origin fonts -->
<link rel="preload" href="/fonts/body.woff2" as="font" type="font/woff2" crossorigin>

---

Layout shift from web fonts

The bug: Text renders with fallback font, then shifts when custom font loads. This counts against CLS.

The fix:

1. Use font-display: swap (accepts the shift but shows text immediately) 2. Match fallback font metrics to custom font using Font Style Matcher 3. Or use font-display: optional (may not load custom font on slow connections)

---

iOS Safari 100vh bug

The bug: 100vh includes the address bar height on iOS Safari, causing content to be cut off.

The fix:

/* Old way */
height: 100vh;

/* New way - respects actual viewport */
height: 100dvh;

/* Fallback for older browsers */
height: 100vh;
height: 100dvh;

---

Debugging tips

1. Test on real devices: Simulators don't catch everything 2. Test on slow connections: Chrome DevTools → Network → Slow 3G 3. Test production builds: Dev mode often hides issues 4. Check console for silent failures: Especially font loading and CORS errors

Performance deep dive

Scripts: the render blockers

The problem: Scripts in <head> without defer or async block HTML parsing.

Find them:

grep -r "<script" --include="*.html" | grep -v "defer\|async\|type=\"module\""

The fix:

<!-- Before: blocks rendering -->
<script src="app.js"></script>

<!-- After: loads in parallel, executes after parsing -->
<script src="app.js" defer></script>

When to use what:

Attribute	Behavior	Use for
`defer`	Download parallel, execute in order after HTML parsed	App bundles, anything that needs DOM
`async`	Download parallel, execute immediately when ready	Analytics, tracking, independent scripts
`type="module"`	Deferred by default, strict mode	Modern ES modules
None	Blocks everything	Almost never

The trap: Don't use both defer and async. async wins and execution order is lost.

---

Fonts: the silent killer

Fonts are the #1 cause of slow LCP on text-heavy sites.

The problem: Browser won't paint text until fonts load. On slow connections, users see nothing for seconds.

Find issues:

# Check what's being preloaded
grep -r "preload.*font" --include="*.html"

# Find @font-face declarations
grep -r "@font-face" --include="*.css"

# List all font files
find . -type f \( -name "*.woff2" -o -name "*.woff" -o -name "*.otf" \) | head -20

The fix:

1. Use WOFF2: 30% smaller than WOFF, supported everywhere since 2018.

   <!-- Wrong -->
   <link rel="preload" href="/fonts/body.otf" as="font" crossorigin>

   <!-- Right -->
   <link rel="preload" href="/fonts/body.woff2" as="font" type="font/woff2" crossorigin>

2. Preload only critical fonts: Usually 2-3 max. The ones used above the fold.

3. Add font-display: swap:

   @font-face {
     font-family: 'Body';
     src: url('/fonts/body.woff2') format('woff2');
     font-display: swap;
   }

4. Exclude unused fonts from build:

   # Jekyll _config.yml
   exclude:
     - assets/fonts/unused-family/

I've seen font directories go from 74MB to 3MB by excluding unused families.

The trap: font-display: swap causes a flash of unstyled text (FOUT). The alternative—invisible text for 3 seconds—is worse.

---

Images: the obvious wins

Always specify dimensions (prevents layout shift):

<img src="hero.jpg" width="1200" height="600" alt="Hero">

Lazy load below-fold images:

<img src="photo.jpg" loading="lazy" alt="Description">

Use modern formats:

<picture>
  <source srcset="image.avif" type="image/avif">
  <source srcset="image.webp" type="image/webp">
  <img src="image.jpg" alt="Fallback">
</picture>

The trap: Don't lazy load the LCP image. That's the opposite of what you want.

---

The hidden bloat

Check your build output:

du -sh _site/
find _site/ -type f -size +100k | sort -k1 -h

Common culprits:

Unused font families: Shipped the whole type family when you only use regular and bold
Source maps in production: Great for debugging, terrible for users
Unoptimized images: A 4MB PNG that should be a 200KB WebP

Audit report template

Use this format to summarize findings after running an audit.

---

Performance audit

Fixed

Added defer to X render-blocking scripts
Switched font preloads from OTF to WOFF2 (saved XXX KB)
Excluded X unused font weights from build (saved X.X MB)
Added loading="lazy" to X below-fold images
Added width/height to X images

Manual review needed

Hero image is X.X MB — consider compression or WebP
Third-party script (analytics/chat/etc.) adds XXX ms to TTI
Consider lazy loading below-fold components

---

Accessibility audit

Fixed

Replaced global outline: none with :focus-visible styles
Added skip link to main layout
Converted X <div onclick> to <button> elements
Added aria-label to X icon-only buttons
Added keyboard handlers to custom interactive elements

Manual review needed

Color contrast on .class-name (currently X.XX:1, needs 4.5:1)
X images need descriptive alt text (list paths)
Heading hierarchy: h1 → h3 skip on /page-name

---

SEO audit

Fixed

Removed duplicate <title> tag from template
Added meta descriptions to X pages
Created robots.txt with sitemap reference
Added canonical URLs

Manual review needed

X pages missing meta descriptions
Title on /page-name exceeds 60 characters

---

Best practices audit

Fixed

Corrected MIME types on font preloads
Removed X console.log statements
Fixed mixed content (HTTP → HTTPS)

Manual review needed

Safari animation bug potential in /path/to/file.js
Review third-party scripts for security

---

Core Web Vitals

Metric	Before	After	Target
LCP	X.Xs	X.Xs	< 2.5s
CLS	X.XX	X.XX	< 0.1
INP	—	—	< 200ms

Notes:

INP requires real user data (lab tests measure TBT instead)
Field data available at PageSpeed Insights

---

Summary

Category	Before	After
Performance	XX	XX
Accessibility	XX	XX
Best practices	XX	XX
SEO	XX	XX

Key wins

1. 2. 3.

Remaining work

1. 2.

Related skills

Improve Codebase ArchitectureSafely deepen clusters of shallow modules into cohesive, testable units while respecting their external dependencies.566k196k

Caveman ReviewGet ultra-compressed, one-line code review comments that cut noise while keeping every actionable fix.270k94.7k

Codebase DesignShared vocabulary for designing deep modules: improve a module's interface, find deepening opportunities, decide where a seam goes, make code more testable.266k196k

CavecrewDelegate coding tasks to specialized subagents that return compressed output, keeping the main context window usable for much longer sessions.220k94.7k

Code ReviewReviews a branch or PR diff on two axes at once: conformance to coding standards plus a code-smell baseline, and whether it actually implements the original spec.201k196k

Requesting Code ReviewDispatch a consistent, high-signal code reviewer subagent that catches plan deviations and quality issues before merging or continuing development.183k264k

Code Review & Qualityfrontendtesting

About

Perf Expert by the numbers

Add your badge

What it does

Files

Performance expert

What you deliver

Usage

Audit methodology

Step 1: Establish baselines

Step 2: Identify the stack

Step 3: Audit each category

Step 4: Deliver the report

Quick reference

Critical issues (fix immediately)

High impact fixes

Commands to run

Example output format

License

perf-analyzer

What you get

Skill structure

Installation

Usage

What makes this different

It delivers actionable fixes, not vague advice

It's battle-tested

It knows the targets

Files

Real results

When to skip this

Contributing

License

Author

Accessibility deep dive

Focus states: the keyboard user's lifeline

Skip links: respect your users' time

Semantic HTML: divs are not buttons

ARIA: the last resort

Keyboard navigation

Focus traps for modals

Headings: the screen reader's table of contents

Color contrast

Browser gotchas

Safari: CSS variables in @keyframes

CSS minifiers breaking calc()

MIME type mismatches

Mixed content warnings

Preload without crossorigin

Layout shift from web fonts

iOS Safari 100vh bug

Debugging tips

Performance deep dive

Scripts: the render blockers

Fonts: the silent killer

Images: the obvious wins

The hidden bloat

Audit report template

Performance audit

Fixed

Manual review needed

Accessibility audit

Fixed

Manual review needed

SEO audit

Fixed

Manual review needed

Best practices audit

Fixed

Manual review needed

Core Web Vitals

Summary

Key wins

Remaining work

Related skills

This week in AI coding