Ozon Shopper

Ozon Shopper — AI-Driven Product Search & Scam Detection

Overview

This skill enables searching Ozon for products, analyzing listings for scams, validating technical specs, and recommending the best options. The workflow is:

1. Search → collect product cards from Ozon
2. Analyze → deep-dive each product (specs, reviews, seller)
3. Validate → check for scam patterns and spec inconsistencies
4. Report → ranked list with trust scores and red flags

Prerequisites

pip install playwright
playwright install chromium

Scripts

All scripts live in scripts/ and accept --help for usage.

Workflow

Step 1: Search

python scripts/ozon_search.py --query "ноутбук ASUS" --max-results 10 --output results.json

This produces a JSON file with product cards:

[
  {
    "title": "Ноутбук ASUS VivoBook 15...",
    "price": "45 990",
    "url": "https://www.ozon.ru/product/...",
    "rating": "4.8",
    "reviews_count": "1240",
    "image": "https://..."
  }
]

Step 2: Analyze Each Product

For each product from search results, run three analyses in parallel:

python scripts/ozon_product_page.py --url "https://www.ozon.ru/product/..." --output product.json
python scripts/ozon_reviews.py --url "https://www.ozon.ru/product/..." --max-reviews 30 --output reviews.json
python scripts/ozon_seller.py --url "https://www.ozon.ru/product/..." --output seller.json

Step 3: Validate Against Scam Patterns

After collecting data, read references/scam_patterns.md and references/specs_rules.md, then analyze each product against these checklists.

Key validation checks:

• Price anomaly (too low for the category/specs)
• Seller trust (rating, age on platform, number of products)
• Review authenticity (bot patterns, review velocity, rating distribution)
• Spec consistency (claimed specs vs realistic pricing)
• Brand verification (official vs third-party seller)

Step 4: Generate Report

python scripts/helpers/reporter.py --input results/ --output report.md

AI Analysis Guidelines

When analyzing products, follow these principles from references/scam_patterns.md:

Red Flags (immediate disqualification)

• Price 40%+ below market average for identical specs
• Seller account less than 3 months old with high-value electronics
• All reviews posted within 48 hours
• Vague specs (e.g., "Intel i7" without model number)
• Stock photos instead of real product images

Yellow Flags (investigate further)

• Price 20-40% below market average
• Mixed review sentiment despite high rating
• Seller has few products but many sales
• Specifications that look "too good to be true" for the price

Trust Signals (positive indicators)

• Official brand store on Ozon
• Consistent review timeline spanning months
• Detailed, specific reviews with photos
• Seller responds to negative reviews professionally
• Specs match manufacturer's official page

Decision Tree

User query → Is it a search or a specific URL?
├── Search → Run ozon_search.py → Collect results
│   └── For each result → Run product + reviews + seller analysis
│       └── Validate → Generate ranked report
│
└── Specific URL → Run product + reviews + seller analysis
    └── Validate → Generate single product analysis

Report Format

Always generate reports in this structure:

## Результаты поиска: "[query]"

### 1. [Product Name] — [Price]₽ ✅/⚠️/🔴 [Verdict]
- **Продавец**: [Name] (рейтинг [X], [Y] месяцев на Ozon)
- **Отзывы**: [N] отзывов, [X]% положительных
- **Характеристики**: [Key specs]
- **Оценка доверия**: [X]/10
- **Красные флаги**: [list or "Нет"]

Extending to Other Platforms

The skill is designed for easy extension:

1. Create wb_search.py, wb_product_page.py (Wildberries)
2. Create ali_search.py, ali_product_page.py (AliExpress)
3. Each script must follow the same CLI interface:
   • Accept --url or --query
   • Accept --output for JSON file path
   • Print JSON to stdout if no --output
4. Update SKILL.md with platform-specific selectors

Quick Reference