К основному содержанию
скилл CX Codex

codiris-agentizer — скилл для Claude Code и Codex: сайт в agentfront

Скилл codiris-agentizer разбирает сайт по страницам, выясняет, что компания продаёт, и собирает из Next.js-шаблона интерактивный agentfront под этот бизнес.

MIT ★ 1 596 обновлён 2026-10-29

Установить в Claude Code · Через Skills CLI (Vercel) Все способы и платформы ↓
npx skills add hashgraph-online/awesome-codex-plugins --skill codiris-agentizer -a claude-code
Открыть на GitHub

Что делает codiris-agentizer

Скилл открывает через WebFetch несколько страниц сайта: главную, продукты, цены, «о компании». Оттуда он вытягивает название бренда, слоган, ссылки на картинки, hex-цвета кнопок, пункты меню, тарифы и цифры. Потом разбирает рынок и тип покупателя.

На выходе получается приложение на Next.js, где секции, шрифты и главная кнопка подобраны под бизнес. Магазину достаётся каталог и кнопка покупки, продавцу API — примеры кода и бесплатный тариф. Эмитенту карт нужны сравнение планов и форма заявки. Шрифты и цвета правятся в src/app/layout.tsx и src/app/globals.css.

Шаблон лежит в ./template/ рядом с SKILL.md: скопируйте его и один раз прогоните npx next build, иначе поломку в самом шаблоне поймаете уже на своём контенте.

Какие задачи решает

  • Превратить сайт компании в интерактивный agentfront
  • Снять с лендинга палитру, шрифты и структуру навигации
  • Подобрать главную кнопку под тип бизнеса: покупка, заявка, демо
  • Собрать секции главной под конкретный рынок, а не по шаблону
  • Проверить сборку шаблона до работы над контентом

Примеры запросов агенту

  • Агентзируй <website-url> и собери agentfront под их продукт.
  • Прогони wwa transform для <website-url>: разбери продукты и цены, шрифты подбери под бренд.
  • Сделай agent-first версию сайта <website-url>, кнопки подбери по тому, как они продают.

Как установить codiris-agentizer

1Через Skills CLI (Vercel)

npx skills add hashgraph-online/awesome-codex-plugins --skill codiris-agentizer -a claude-code

2Вручную: положить файлы

~/.claude/skills/wwa-transform/SKILL.md

Папка скилла целиком; для одного проекта — .claude/skills/wwa-transform/

По шагам, с командами для каждой платформы: Как установить скилл в Claude Code: пять способов

Вопросы о codiris-agentizer

Как установить codiris-agentizer в Claude Code?

Выполните npx skills add hashgraph-online/awesome-codex-plugins --skill wwa-transform -a claude-code. Скилл появится по пути ~/.claude/skills/wwa-transform/SKILL.md.

Что делает codiris-agentizer?

Читает сайт, решает, какие секции и какой призыв к действию подходят бизнесу, и собирает agentfront на Next.js из шаблона внутри скилла.

Что делать, если файлы шаблона выглядят сломанными?

Посмотрите живое демо https://wwa.visa.codiris.app через WebFetch и примените переименования visaCards → productItems, VisaCard → ProductItem, annualFee → price. Править заглушку на месте не нужно.

Исходный текст

SKILL.md · MIT
---
name: codiris-agentizer
description: >-
  This skill should be used when the user asks to "agentize a website",
  "transform a website", "create an agentfront", "build a WWA app",
  "make an agent-first website", "turn a website into an agent experience",
  "agentizer", "wwa transform", or mentions "agentfront" in context of
  building one from a URL.
version: 2.0.0
allowed-tools:
  - Read
  - Write
  - Edit
  - Bash
  - Glob
  - Grep
  - Agent
  - WebFetch
  - WebSearch
argument-hint: <website-url>
---

# Codiris Agentizer v2 — Intelligent Website-to-Agentfront

Transform any website into a custom AI-powered agentfront. The skill deeply analyzes the company — its market, products, value proposition, customers — then builds a tailored experience. NOT a template copy-paste.

## Core Principle

**Understand the company first, build second.** Every company is different:
- Stripe sells APIs to developers → show API products, code examples, pricing tiers
- Nike sells shoes to consumers → show product catalog, categories, shopping
- Visa sells payment cards → show card tiers, comparisons, benefits
- Salesforce sells enterprise SaaS → show solutions by industry, case studies, demos

The agentfront must reflect what the company ACTUALLY is, not what Visa is.

## Template Location
`./template/` relative to this SKILL.md — the skill directory (typically `~/.claude/skills/agentizer/` after install) contains the Next.js template. Content is generated custom per company. After `cp -R`, run the template's `npx next build` once to confirm nothing regressed in the template itself.

## Reference Implementation
**https://wwa.visa.codiris.app** — live demo of a finished agentfront (payment card company). Source for the template is at https://github.com/Humiris/wwa-transform/tree/main/template.

If any template file looks incomplete or broken, inspect the live demo via WebFetch for visual reference, then apply the `visaCards → productItems` / `VisaCard → ProductItem` / `annualFee → price` renames — don't try to fix the template stub in place.

---

## PHASE 1: Deep Company Intelligence

### Step 1.1: Crawl Multiple Pages

Use `WebFetch` on at minimum:
1. **Homepage** — hero, value prop, main CTAs, navigation
2. **Products/Features page** — what they actually sell
3. **Pricing page** (if exists) — how they charge, what tiers
4. **About page** — company story, mission, stats

For EACH page, ask WebFetch:
```
Extract everything: brand name, tagline, ALL image URLs (full absolute), 
colors (hex from buttons/headers/CSS), navigation items, products/features 
with descriptions, pricing tiers, stats/numbers, customer logos, 
testimonials, CTAs, form fields. Be exhaustive.
```

### Step 1.2: AI Company Analysis

After crawling, THINK deeply about the company. Answer these questions BEFORE writing any code:

**Market Understanding:**
- What industry is this company in? (Fintech, SaaS, E-commerce, Healthcare, etc.)
- Who are their customers? (Developers, Enterprises, Consumers, SMBs)
- What problem do they solve?
- Who are their competitors?

**Product Understanding:**
- What do they sell? (APIs, Physical products, Subscriptions, Services, Cards)
- How many products/tiers do they have?
- How do they charge? (Per-transaction, Subscription, One-time, Usage-based, Free)
- Do they have a free tier or trial?

**Content Structure:**
- What sections make sense for THIS company's homepage?
- What navigation items should exist?
- What categories should products be organized by?
- Should there be a "cards" section? (Only if they sell cards/plans/tiers)
- Should there be a pricing grid? A product catalog? Case studies?

**Value Proposition:**
- What's their main headline?
- What stats are impressive? (Users, revenue, uptime, countries)
- What social proof do they have? (Customer logos, testimonials)

**Typography (CRITICAL — match brand aesthetic):**

The template uses Geist by default. You MUST change fonts to match the brand:

| Brand Type | Fonts to Use | Implementation |
|------------|--------------|----------------|
| Luxury fashion (Dior, Hermès, Chanel) | Playfair Display + Inter | Serif headings, clean body |
| Tech/SaaS (Stripe, Notion, Linear) | Inter or Geist | Modern sans-serif |
| Premium/Editorial (Apple, Aesop) | SF Pro / Helvetica Neue / system fonts | Clean, minimal |
| Fintech (Visa, Chase) | Geist / Inter | Professional sans |
| E-commerce (Amazon, Walmart) | Inter / Roboto | Readable, utilitarian |

Update `src/app/layout.tsx`:
```tsx
import { Playfair_Display, Inter } from "next/font/google";

const serif = Playfair_Display({ variable: "--font-serif", subsets: ["latin"] });
const sans = Inter({ variable: "--font-sans", subsets: ["latin"] });

// In body: className={`${serif.variable} ${sans.variable} font-sans`}
```

Update `src/app/globals.css`:
```css
:root {
  --brand-primary: #000000;    /* match brand color */
  --brand-secondary: #C9A96E;
  --font-serif-brand: var(--font-serif), "Didot", "Bodoni MT", "Georgia", serif;
}

h1, h2, .font-serif {
  font-family: var(--font-serif-brand);
  letter-spacing: -0.01em;
}
```

**User Intent (CRITICAL — drives CTAs and modals):**
Analyze what users ACTUALLY do on this site. Map to correct CTA:

| Company Type | Primary Intent | CTA Text | Modal Type |
|--------------|---------------|----------|------------|
| E-commerce (Dior, Nike, Apple) | Shop / Buy products | "Shop Now" | Cart / Product detail |
| SaaS/API (Stripe, Notion) | Try / Book demo | "Start Free" or "Book a Demo" | Demo booking form |
| Enterprise (Salesforce) | Contact sales | "Contact Sales" | Sales form |
| Financial (Visa, banks) | Apply / Get card | "Apply Now" | Application form |
| Marketplace — Listings (Airbnb, Booking, Uber) | Search / Book | "Start Searching" | Search UI |
| Marketplace — C2C (Vinted, Depop, Etsy, eBay) | Sell AND Browse (dual intent) | "Sell Now" + "Start Browsing" | Seller signup modal (personal / pro) |
| Services (Consulting) | Contact / Meeting | "Book a Call" | Calendar booking |
| Content (Netflix, Spotify) | Subscribe / Try | "Start Free Trial" | Signup form |

**NEVER use "Book a Demo" for e-commerce.** E-commerce brands want you to SHOP, not schedule a meeting.

Update these files based on intent:
- `brand-config.ts` → `ctaPrimary` and `ctaSecondary`
- `hero-section.tsx` → button text
- `solution-detail-panel.tsx` → button text
- `book-demo-modal.tsx` → rename, change fields to match intent (e-commerce = Contact Client Services, Cart, or remove entirely)
- `actions.ts` system prompt → what the agent offers users

### Step 1.3: Define the Page Architecture

Based on your analysis, decide which sections the homepage should have. Choose from:

| Section | Use When | Example |
|---------|----------|---------|
| Hero Carousel | Always | Rotating through top products/features |
| Stats Bar | Company has impressive numbers | "$1.9T processed", "99.999% uptime" |
| Product Grid | 3+ distinct products | API products, SaaS modules |
| Pricing Tiers | Clear pricing tiers exist | Free/Pro/Enterprise |
| Feature Showcase | Key differentiators | With images and descriptions |
| Customer Logos | Has notable customers | "Trusted by Amazon, Shopify..." |
| Solution Categories | Products grouped by use case | "For Startups", "For Enterprise" |
| Code Example | Developer-focused company | Interactive code snippet |
| Card Catalog | Sells cards/plans consumers browse | Credit cards, subscription plans |
| CTA Banner | Always | "Book a Demo" / "Get Started" / "Start Free" |
| World Map | Company has global presence story | Payment networks, logistics, international SaaS |

**DO NOT include sections that don't apply:**
- API company → NO credit card tiers, NO card comparison
- E-commerce → NO API docs, NO code examples
- SaaS without global story → NO world map (make `handleShowMap` a no-op in page.tsx)
- Company without browsable products → leave `productItems` array empty, the browse section won't render

---

## PHASE 2: Download Assets

### Step 2.1: Download ALL Images

**MANDATORY.** The agentfront is useless without real images.

> ❗ **The single biggest bug in past transforms** is shipping images from the wrong brand — a generic "sneaker" photo from Unsplash that happens to show a Nike Swoosh on an adidas site, or a "handbag" photo that's clearly Chanel on a Dior site. Mis-branded imagery destroys credibility instantly.
>
> **Rule**: Every image that represents a **specific named product or franchise** (e.g. "Samba OG", "Air Max 90", "Lady Dior") MUST come from a source that lets you verify it shows the correct brand. Generic-category Unsplash photos are banned for these slots. They are OK only for brand-agnostic shots (lifestyle hero, category tiles with no named product, abstract backgrounds).
>
> **And every image MUST pass visual verification (Step 2.3) before you write its path into `cards.ts` / `solutions.ts`.**

```bash
mkdir -p ~/wwa.{domain}/public/images/cards

# Use browser headers to bypass CDN blocks
curl -L -o "public/images/{name}.jpg" "{url}" \
  -H "User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36" \
  -H "Referer: https://{domain}/" \
  -H "Accept: image/*"
```

**Download at minimum:**
- Logo (SVG preferred, PNG fallback)
- Hero/banner image
- 1 image per product/solution (6-10 total)
- 2-4 lifestyle/feature images
- Customer logos if available
- Product screenshots/UI images

### Image Source Priority (USE IN THIS ORDER)

**For e-commerce catalogs (fashion retailers, beauty retailers, online-only fashion), always try method 0 FIRST:**

**0. Scrape the brand's SSR catalog JSON via Chrome MCP** — modern e-commerce sites (Zara, H&M, ASOS, Nordstrom, Sephora, etc.) ship the full product grid as a JSON blob inside a `<script>` tag on every category page. Server-side curl gets WAF-blocked, but Chrome MCP with an organic-looking tab has no trouble reading the DOM.

Two flavors to look for:

- **JSON-LD `<script type="application/ld+json">`** (cleanest — H&M, many schemas). Has `@type: Product` entries with `name`, `image[]`, `offers.price` ready to `JSON.parse`.
- **Raw hydration JSON** (Zara, ASOS — biggest `<script>` that contains the brand's CDN hostname). Not always valid JSON; regex out the image URL pattern (e.g. `/static\.zara\.net\/photos\/[^"'\s]+\.jpg/g` or `/products\/[a-z0-9-]+\/\d+-\d+[a-z0-9-]*/gi`) and dedupe by product id.

Pattern that works today (verified on Zara 259 · H&M 216 · ASOS 365 · Uniqlo 54 scrapes):

```js
// In Chrome MCP, navigate to e.g. https://www.zara.com/uk/en/woman-dresses-l1066.html, then:
(() => {
  const scripts = Array.from(document.querySelectorAll('script'));
  // JSON-LD path (H&M-style): cleanest
  const ld = document.querySelector('script[type="application/ld+json"]');
  if (ld && ld.textContent.includes('"@type":"Product"')) {
    return JSON.parse(ld.textContent);
  }
  // Raw-hydration path (Zara/ASOS): regex
  const big = scripts.find(s => (s.textContent || '').includes('<brand-cdn-hostname>'));
  const text = big.textContent;
  // Adjust regex per brand URL shape
  const urls = [...new Set((text.match(/https:\/\/static\.zara\.net\/[^"'\s]+\.jpg/g) || []))];
  return urls;
})();
```

Then get the result out of Chrome via a `Blob` download:

```js
const blob = new Blob([JSON.stringify(catalog)], { type: 'application/json' });
const a = Object.assign(document.createElement('a'), { href: URL.createObjectURL(blob), download: 'catalog.json' });
document.body.appendChild(a); a.click();
// File lands in ~/Downloads/catalog.json — move to project, feed to cards.ts rewrite
```

**Hotlink don't download.** Once you have the URLs, put them directly in `cards.ts`'s `image` field — the browser will load them via `<img src>` (no CORS issue on img tags). No local copy needed. Server-side curl typically fails with 403 / 000 (WAF), but that doesn't matter because production serves from users' browsers.

**For named products/franchises** (Samba, Air Jordan 1, Stan Smith, Lady Dior, Birkin, etc.) where you need individual-product photos not a catalog:

1. Brand's own CDN (works for most sportswear — Nike, adidas, Puma, Asics, New Balance)
2. **Wikipedia / Wikimedia Commons** — the category page `commons.wikimedia.org/wiki/Category:{Brand}_{Product}` has real, licensed product photos for every iconic sneaker, boot, bag, watch, perfume, etc.
3. Industry publication CDN for the brand's segment (WWD for fashion, Fragrantica for perfume, SneakerNews/Hypebeast for sneakers)
4. **DROP THE PRODUCT from `productItems`** if none of 1–3 yields a verified photo. A site with 6 real products is infinitely better than a site with 8 and two wrong-brand frauds.

**Unsplash is BANNED for named-product slots.** It's OK only for: hero lifestyle shots (action/atmosphere, no named product), brand-agnostic category tiles, abstract backgrounds. If you use it, still run Step 2.3 verification — generic sneaker/bag photos on Unsplash very frequently show competitor logos.

Most premium brand sites (Dior, Apple, Notion, Vercel, Hermès) AGGRESSIVELY BLOCK curl. Do NOT waste time trying to download from the brand's CDN. Skip directly to the fallback.

**Sportswear CDNs are hit-or-miss** — Puma, Nike, Adidas, Asics, Under Armour, New Balance, Reebok sometimes serve real JPEGs on their public CDNs, but **you have to know the exact style code** and the match rate is worse than you'd expect. Observed on recent transforms:

- **adidas** (`assets.adidas.com`) — returned 0 bytes on every style code tried (Apr 2026)
- **Nike** (`static.nike.com`) — works for some products, requires inspecting a product page's `<img srcset>` to get the right hash
- **New Balance** (`nb.scene7.com`) — works with a Referer header, but style codes are hard to guess; ~50% hit rate even for the most popular models
- **Puma** (`images.puma.com`) — most reliable of the sportswear CDNs

**Rule of thumb**: try the brand CDN once with a single well-known SKU. If it returns real bytes (>20KB), map your remaining products; if it 0-bytes or errors, drop straight to Wikimedia Commons. Don't spend 10 minutes guessing style codes.

| Brand | CDN pattern | How to get the `{styleId}` |
|-------|-------------|----------------------------|
| Puma | `https://images.puma.com/image/upload/f_auto,q_auto,w_800,h_800/global/{styleId}/01/sv01/fnd/PNA/fmt/png/{slug}` | The number after `/pd/slug/` on a product page |
| Nike | `https://static.nike.com/a/images/t_PDP_1280_v1/f_auto,q_auto:eco/{hash}/{slug}.jpg` | Inspect a product page's `<img srcset>` |
| Adidas | `https://assets.adidas.com/images/w_600,f_auto,q_auto/{styleId}/{slug}.jpg` | Often blocked — try once, fall back to Wikimedia |
| New Balance | `https://nb.scene7.com/is/image/NB/{styleId}_nb_02_i?wid=1200&hei=1200` | Match the style id on the product page (e.g. `m1906rcd`) |

Always include `-H "Referer: https://{brand-domain}/"` — the Cloudinary-style CDNs behind most sportswear brands accept the request with referrer, 404 it without.

**Remember**: even when the CDN works, you still MUST run Step 2.3 visual verification on the downloaded bytes — CDNs occasionally return placeholder/blank images with a 200 status.

**1. Brand's own CDN** (always try first for sportswear/athletic; ~30% of premium luxury sites; skip straight to #3 for most editorial/luxury brands)
```bash
curl -sL -o public/images/{name}.jpg "{url}" \
  -H "User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36" \
  -H "Referer: https://{domain}/"
```

**2. Wikipedia / Wikimedia Commons** — works for famous brands, logos, iconic products
```bash
curl -sL -o public/images/brand-logo.svg \
  "https://upload.wikimedia.org/wikipedia/commons/{path}/{file}.svg" \
  -H "User-Agent: Mozilla/5.0"
```
Use WebSearch: "{brand} logo SVG wikimedia commons"

**2.5. INDUSTRY PUBLICATIONS — USE THIS FOR FASHION BRANDS**
Fashion/luxury brands block their own CDN, but fashion PUBLICATIONS host real runway photos:

```bash
# WWD (Women's Wear Daily) — 2-3MB runway photos, works without auth
# Pattern: https://wwd.com/wp-content/uploads/{YYYY}/{MM}/{brand}-{collection}-{location}-GG-{01-80}.jpg

curl -sL -o public/images/runway.jpg \
  "https://wwd.com/wp-content/uploads/2025/05/dior-cruise-2026-rome-GG-01.jpg" \
  -H "User-Agent: Mozilla/5.0"

# Fragrantica CDN — real perfume bottle photos, works without auth
# Pattern: https://fimgs.net/mdimg/perfume/375x500.{perfume_id}.jpg
# Find IDs via WebSearch: "{perfume name} fragrantica"

curl -sL -o public/images/jadore.jpg \
  "https://fimgs.net/mdimg/perfume/375x500.210.jpg" \
  -H "User-Agent: Mozilla/5.0"
```

Other working fashion/beauty publication CDNs:
- **wwd.com/wp-content/uploads/** — runway, backstage, beauty product shots
- **fimgs.net/mdimg/perfume/** — perfume bottles (Fragrantica)
- **nowfashion.com/wp-content/uploads/** — runway photos
- **showstudio.com/** — editorial fashion content
- **vogue.com/photos/** — often requires auth
- **fashionography.com/wp-content/uploads/** — runway archives

**IMPORTANT:** Sample a couple of images first (~2-3MB each). On macOS use `sips -Z 1200 input.jpg -o output.jpg` to resize before committing.

### Brand Videos for Hero Backgrounds

Many luxury brands have hero video loops that dramatically elevate the feel. Look for them:

```bash
# Brand-hosted videos often work with proper headers
curl -sL -o public/videos/hero.webm "{brand-video-url}" \
  -H "User-Agent: Mozilla/5.0" \
  -H "Referer: https://{domain}/"
```

**Finding brand video URLs:**
- Dior: `diorama.dam-broadcast.com/...` — works publicly
- Apple: `apple.com/assets-www/...` — often public
- Most fashion houses have `dam-broadcast` or `.cdn.` video CDNs
- Inspect the brand's homepage for `<video>` or `<source>` tags with `.webm` / `.mp4`
- WebSearch: "{brand} hero video webm" or check Dam broadcast networks

**Implementation in hero-section.tsx:**
```tsx
<video autoPlay loop muted playsInline className="absolute inset-0 w-full h-full object-cover" poster="/images/hero-image.jpg">
  <source src="/videos/hero.webm" type="video/webm" />
</video>
<div className="absolute inset-0 bg-gradient-to-b from-black/10 via-transparent to-black/70" />
```

Video hero is essential for luxury brands. Fall back to static image if video can't be obtained. Keep videos under 10MB (resize with ffmpeg if needed).

**American heritage brands (Ralph Lauren, Levi's, L.L. Bean, J.Crew, Tommy Hilfiger) rarely expose public video CDNs.** Plan for a static full-bleed runway or editorial hero with a subtle 1.04 scale-on-mount or parallax effect instead — don't waste time hunting for a video that isn't there.

**3. Unsplash (MANDATORY FALLBACK — always works, real product photos)**
```bash
# Search URL pattern: https://images.unsplash.com/photo-{id}?w=800&q=80
# DO NOT make up IDs — use these tested working IDs by category:

# Luxury perfume / fragrance bottle
curl -sL -o public/images/perfume.jpg "https://images.unsplash.com/photo-1541643600914-78b084683601?w=800&q=80" -H "User-Agent: Mozilla/5.0"

# Men's cologne / dark bottle  
curl -sL -o public/images/cologne.jpg "https://images.unsplash.com/photo-1594035910387-fea47794261f?w=800&q=80" -H "User-Agent: Mozilla/5.0"

# Pink/feminine perfume
curl -sL -o public/images/miss.jpg "https://images.unsplash.com/photo-1592945403244-b3fbafd7f539?w=800&q=80" -H "User-Agent: Mozilla/5.0"

# Luxury handbag
curl -sL -o public/images/bag.jpg "https://images.unsplash.com/photo-1584917865442-de89df76afd3?w=800&q=80" -H "User-Agent: Mozilla/5.0"

# Canvas tote
curl -sL -o public/images/tote.jpg "https://images.unsplash.com/photo-1590874103328-eac38a683ce7?w=800&q=80" -H "User-Agent: Mozilla/5.0"

# Leather bag
curl -sL -o public/images/leather-bag.jpg "https://images.unsplash.com/photo-1548036328-c9fa89d128fa?w=800&q=80" -H "User-Agent: Mozilla/5.0"

# Red lipstick
curl -sL -o public/images/lipstick.jpg "https://images.unsplash.com/photo-1586495777744-4413f21062fa?w=800&q=80" -H "User-Agent: Mozilla/5.0"

# Foundation / makeup
curl -sL -o public/images/foundation.jpg "https://images.unsplash.com/photo-1512496015851-a90fb38ba796?w=800&q=80" -H "User-Agent: Mozilla/5.0"

# Skincare cream
curl -sL -o public/images/skincare.jpg "https://images.unsplash.com/photo-1556228720-195a672e8a03?w=800&q=80" -H "User-Agent: Mozilla/5.0"

# Luxury skincare bottles
curl -sL -o public/images/serum.jpg "https://images.unsplash.com/photo-1608248543803-ba4f8c70ae0b?w=800&q=80" -H "User-Agent: Mozilla/5.0"

# Gold jewelry / rings
curl -sL -o public/images/jewelry.jpg "https://images.unsplash.com/photo-1515562141207-7a88fb7ce338?w=800&q=80" -H "User-Agent: Mozilla/5.0"

# Fashion / clothing
curl -sL -o public/images/fashion.jpg "https://images.unsplash.com/photo-1558769132-cb1aea458c5e?w=800&q=80" -H "User-Agent: Mozilla/5.0"

# Building / atelier (hero/about)
curl -sL -o public/images/building.jpg "https://images.unsplash.com/photo-1519643225200-94e79e383724?w=800&q=80" -H "User-Agent: Mozilla/5.0"

# SaaS / tech product screenshot
curl -sL -o public/images/saas-ui.jpg "https://images.unsplash.com/photo-1551288049-bebda4e38f71?w=800&q=80" -H "User-Agent: Mozilla/5.0"

# Abstract tech / code
curl -sL -o public/images/tech.jpg "https://images.unsplash.com/photo-1518770660439-4636190af475?w=800&q=80" -H "User-Agent: Mozilla/5.0"

# ═══════ ATHLETIC / SPORTSWEAR (verified on Puma transform) ═══════

# Running / runner silhouette
curl -sL -o public/images/running.jpg "https://images.unsplash.com/photo-1508685096489-7aacd43bd3b1?w=800&q=80" -H "User-Agent: Mozilla/5.0"

# Training / gym / lifting
curl -sL -o public/images/training.jpg "https://images.unsplash.com/photo-1517836357463-d25dfeac3438?w=800&q=80" -H "User-Agent: Mozilla/5.0"

# Football / soccer
curl -sL -o public/images/football.jpg "https://images.unsplash.com/photo-1431324155629-1a6deb1dec8d?w=800&q=80" -H "User-Agent: Mozilla/5.0"

# Motorsport / racing
curl -sL -o public/images/motorsport.jpg "https://images.unsplash.com/photo-1504276048855-f3d60e69632f?w=800&q=80" -H "User-Agent: Mozilla/5.0"

# Kids athletic
curl -sL -o public/images/kids.jpg "https://images.unsplash.com/photo-1503944583220-79d8926ad5e2?w=800&q=80" -H "User-Agent: Mozilla/5.0"

# Athletic fashion / streetwear runner
curl -sL -o public/images/athleisure.jpg "https://images.unsplash.com/photo-1558769132-cb1aea458c5e?w=800&q=80" -H "User-Agent: Mozilla/5.0"
```

> **Unsplash ID format caveat:** stick to the timestamp-hex pattern (`photo-1508685096489-7aacd43bd3b1`). Modern Unsplash URLs use slug-style IDs (`hEIHXXic3yI`) that **do NOT work** with `https://images.unsplash.com/photo-{slug}`. If an ID you find is 11 random characters instead of the classic timestamp-hex form, you need the old numeric pattern or the image will 404.

**Verify downloads:** Check file size with `ls -la`. Files under 1KB are failures. Failed downloads return ~29 bytes (Unsplash not-found) or ~300 bytes (error page). Re-download with different ID if under 1KB.

**4. Placehold.co** — LAST RESORT ONLY if Unsplash also fails
```bash
curl -sL -o public/images/{name}.jpg "https://placehold.co/800x400/{hex_no_hash}/white?text={Product}"
```

### Required Downloads

Every project MUST have (minimum):
- Logo (SVG from Wikipedia or brand CDN)
- Hero image (brand or category-matched Unsplash)
- 1 image per solution (category-matched Unsplash)
- 1 image per product (category-matched Unsplash)
- Favicon from `https://www.google.com/s2/favicons?domain={domain}&sz=64`

### Catalog size — minimum item count per brand type (REQUIRED)

A thin catalog kills the demo. Every e-commerce, retail, or fashion brand transform MUST ship at least **100 real product items** spread across the brand's natural segments. The agent needs enough depth to recommend and compare realistically; an 8-item catalog makes every filter / category / chat prompt dead-end immediately.

| Brand type | Minimum items | Split guidance |
|------------|---------------|----------------|
| Fashion mono-brand (Zara, H&M, Uniqlo, Ralph Lauren, Dior RTW) | **100+** | 40 Woman / 30 Man / 15 Kids / 10 Accessories / 5 Home — adjust to brand reality |
| Fashion multi-brand retailer (Nordstrom, Net-A-Porter, Farfetch, Zalando) | **150+** | 50 Woman / 40 Man / 25 Kids / 20 Shoes / 15 Bags — plus 20+ brand catalogue entries |
| Beauty multi-brand retailer (Sephora, Ulta, LookFantastic) | **120+** | 30 Skincare / 30 Makeup / 25 Fragrance / 15 Hair / 10 Body / 10 Men — plus 20+ brands |
| Beauty mono-brand (L'Oréal, Estée Lauder, MAC) | **80+** | 25 Skincare / 25 Makeup / 15 Fragrance / 10 Hair / 5 Men |
| Outerwear / Technical (Canada Goose, Arc'teryx, Patagonia) | **60+** | 25 Men / 20 Women / 10 Kids / 5 Accessories — fewer items OK because parkas are expensive/curated |
| Luxury mono-brand (Hermès, Chanel — no RTW catalog online) | **50+** | Scale to what's published; bags + fragrance + leather small goods heavy |
| Sportswear (Nike, Puma, Adidas, Under Armour) | **100+** | 30 Men / 30 Women / 20 Kids / 15 Footwear / 5 Accessories |
| Marketplace listings (Airbnb, Booking) | **100+ stays × 40+ destinations** | destinations.ts covers the cities, cards.ts covers the stays |
| Marketplace C2C (Vinted, Depop, Etsy) | **80+ feature tiles + category mock listings** | cards.ts holds features/flows, not individual listings |
| Fintech / SaaS | 6-12 products is fine | Few products is brand-accurate (Visa has ~8 tiers, Stripe has ~10 products) |

**How to reach 100+ on an e-commerce transform:**
1. Start by listing the brand's real product segments (Woman/Man/Kids/Home/Beauty for Zara; Skincare/Makeup/Fragrance for Sephora). Write the minimum count per segment before downloading images.
2. Scrape representative products from the brand's own site when possible, even if names are paraphrased — the image + name + price + material gives the agent enough to recommend.
3. When the brand CDN blocks (Zara, Dior, etc.), fall through to Unsplash with segment-matched searches. Each segment needs 30+ unique photos; guard against duplicates.
4. Every product entry in `cards.ts` MUST be a real-looking record with: name + price + segment (gender/category) + material/size + badge. Agent recommendations against a half-populated catalog feel hollow.

**Per-section rendering rule (NEW, critical — addresses "there's not a lot of items on Men's" feedback):**

The catalog-size rule above covers `cards.ts` data depth. The **rendering cap** is a separate trap: template components often render `homes.slice(0, 12)` or similar, so even with 100+ items in data, the user sees a thin 8-tile grid.

- **Homepage carousel rows**: 12-16 tiles OK (carousel UX, user scrolls horizontally).
- **Category / gender inner pages** (when user clicks "Men's" or "Woman" from nav or a tile): **MUST render the full filtered catalog, no slice cap.** If the brand has 30 Men's items, the Men's page shows 30, not 12. Minimum per-section display:
  - Woman / Man: 30+ tiles visible on the inner page
  - Kids / Baby: 15+ tiles
  - Accessories / Home / Beauty: 10+ tiles
- **Fix pattern**: in `inner-page.tsx` (or equivalent), the category grid must `productItems.filter(p => p.issuer === gender)` without a trailing `.slice()`. If the page feels too long, use pagination / "Load more" — not a silent truncation.
- Grep check before shipping: `grep -rnE '\.slice\(0,\s*(6|8|12)' src/components/inner-page.tsx src/components/hero-section.tsx src/app/page.tsx` — if caps are present on gender/category paths, they're bugs.

**Quality gate (add to Phase 4 verification):**
- `curl https://wwa.{slug}.codiris.app/mcp -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"search_products","arguments":{}}}'` → `count: 100+` for any e-commerce brand type.
- Curl the rendered HTML of the Men's / Woman's category pages and count `<img` tags for product tiles. Must be 30+ each (for mono-brand fashion) or 40+ (for multi-brand fashion). If below threshold, UI has a rendering cap that needs removal — the transform is incomplete.

**Post-deploy smoke check (catches stale-alias bugs):** after `vercel alias set`, always re-verify that the alias points to the LATEST deploy. A common failure mode: you fix the code locally, run `npx vercel --prod`, but forget to re-alias — so `wwa.{slug}.codiris.app` continues pointing to the previous (broken) deploy. The Zara/Uniqlo/H&M/ASOS retrofit pass uncovered 4 stale aliases where the uncap code was already correct in git but not reflected live. Always run:

```bash
npx vercel --prod --yes 2>&1 | tee /tmp/deploy.log
DEPLOY_URL=$(grep -oE 'https://[a-z0-9-]+\.vercel\.app' /tmp/deploy.log | head -1)
npx vercel alias set "$DEPLOY_URL" wwa.{slug}.codiris.app
# Then verify:
curl -s https://wwa.{slug}.codiris.app/ | grep -c 'aspect-\[3/4\]\|aspect-\[4/5\]\|aspect-square'  # rough product-tile count
```

**MANDATORY: push API keys from a sibling project before declaring "deployed" (catches silent-AI-fallback bug):** every brand-new Vercel project starts with **zero env vars**. Without `GEMINI_API_KEY` / `GOOGLE_GENERATIVE_AI_API_KEY` / `OPENAI_API_KEY`, the AI panel silently falls back to canned `assistant-shared.tsx` strings ("I can help you find..."), and the user thinks the AI is dumb. Caught on the Joivy deploy after the user complained the AI was "stupid". The 200-response on `/api/gemini-key` is the actual smoke check — until that endpoint returns a real key, the demo is broken even if every page renders correctly.

```bash
# Pull keys from any sibling project that already has them (wwa-vinted is reliable)
mkdir -p /tmp/sibling-env && cd /tmp/sibling-env && mkdir -p .vercel
cat > .vercel/project.json << 'EOF'
{"projectId":"prj_eoF8zwRk4i3blOS7Cy95N9Zn5I1s","orgId":"team_CVjPPr2WoqlNcN6zmRDjk04Q","projectName":"wwa-vinted"}
EOF
npx vercel env pull .env.production --environment=production --yes

# Push to the new project via Vercel API
TOKEN=$(python3 -c "import json; print(json.load(open('/Users/joel/Library/Application Support/com.vercel.cli/auth....

© Iris Lab · MIT · hashgraph-online/awesome-codex-plugins

Похожие скиллы

Все скиллы по теме «Настройка агента: память, контекст, субагенты» →
stitch::code-to-design официальный google-labs-code Скилл stitch::code-to-design для Claude Code переносит существующий фронтенд в Stitch: извлекает HTML и дизайн-систему и загружает их в проект. CX Codex +1 11 700установок awwwards juliobsz Собирает сайт-портфолио или лендинг из одной фразы, держа планку сайтов, которые побеждают на Awwwards. CX Codex +1 2 700установок design-lab 0xdesign Скилл design-lab проводит дизайн-интервью, собирает пять разных вариантов интерфейса во временной лаборатории и доводит выбранный до плана внедрения. CX Codex +1 2 100установок ui-design mblode Скилл ui-design для Claude Code проектирует и собирает интерфейсы на React, Next и Tailwind, а ещё находит визуальные дефекты и ломает компоненты худшими данными. CX Codex +1 1 600установок webflow-cli:devlink webflow Скилл webflow-cli:devlink для Claude Code выгружает компоненты Webflow Designer в React и Next.js, синхронизирует правки дизайна и показывает изменения после каждого sync. CX Codex +1 1 000установок 21st-dev-builder-v2 trin-zenityx Скилл 21st-dev-builder-v2 ведёт агента по реестру 21st.dev: находит компоненты, отбирает подходящие и собирает из них страницы и приложения на React и Tailwind. CX Codex +1 474установок

Из того же набора391

Весь набор →
ship Скилл ship разбивает проект на фазы и проводит каждую от вопросов и плана до сборки, проверки и отчёта. Продолжает работу по сохранённому состоянию после остановки или /clear. CX Codex +1 ★ 1 596звёзд репо pr Справочник по bkt pr для агента: создать, посмотреть, одобрить и смержить pull request в Bitbucket Data Center или Cloud. CX Codex +1 ★ 1 596звёзд репо maintaining-modding-environments Скилл maintaining-modding-environments помогает обновлять базы знаний, очищать кэш пакетов и обслуживать CLI переводчика xtl для моддинга игр Bethesda. CX Codex +1 ★ 1 596звёзд репо review Скилл review запускает проверку изменений несколькими агентами: в быстром режиме работают 5 агентов, а флаг --deep включает полный конвейер проверки. CX Codex +1 ★ 1 596звёзд репо zotero-research-tools Плагин zotero-research-tools для Claude Code и Codex подключает Zotero Desktop через Zotero MCP: агент ищет и читает записи библиотеки, готовит цитаты и разбирает коллекции с тегами. CX Codex ★ 1 596звёзд репо youtube-skills Плагин youtube-skills для Claude Code извлекает транскрипты YouTube с таймстампами, ищет видео и каналы, просматривает плейлисты через TranscriptAPI без yt-dlp и браузеров. CX Codex ★ 1 596звёзд репо yandex-direct-for-all Плагин yandex-direct-for-all для Claude Code и Codex связывает их с Яндекс Директом, Метрикой и Wordstat v2 через MCP-серверы. CX Codex ★ 1 596звёзд репо x-twitter-scraper Плагин x-twitter-scraper для Claude Code и Codex подключает их к данным X через Xquik. Ищет посты и профили; запросы требуют API-ключа. CX Codex ★ 1 596звёзд репо

Спонсоры проекта

Компании, которые помогают держать аналитику открытой