skills/web-scraper/SKILL.md

---
name: web-scraper
description: >-
  Extract structured data from web pages and load it into databases. Use when
  a user asks to scrape a website, build a data pipeline, extract data from a
  webpage, pull prices from a site, collect links, gather product listings,
  download page content, parse HTML, set up ETL, or automate data collection.
  Handles static HTML, JavaScript-rendered pages, anti-bot proxies (Bright Data),
  data transformation, deduplication, and database loading.
license: Apache-2.0
compatibility: "Requires Python 3.9+ with requests and beautifulsoup4 installed. For JS-rendered pages, requires playwright."
metadata:
  author: terminal-skills
  version: "1.0.0"
  category: automation
  tags: ["scraping", "web", "html", "data-extraction", "beautifulsoup", "etl", "bright-data", "pipeline"]
---

# Web Scraper

## Overview

Extract structured data from web pages by parsing HTML, selecting elements with CSS selectors, and outputting clean data in JSON, CSV, or other formats. Handles both static HTML and JavaScript-rendered pages.

## Instructions

When a user asks you to scrape or extract data from a web page, follow these steps:

### Step 1: Assess the target

Determine:
- **URL**: What page to scrape
- **Data needed**: What specific elements to extract (prices, titles, links, tables)
- **Rendering**: Is the page static HTML or does it require JavaScript?
- **Scale**: Single page or multiple pages with pagination?

### Step 2: Fetch the page

**For static HTML:**

```python
import requests
from bs4 import BeautifulSoup

def fetch_page(url):
    headers = {
        "User-Agent": "Mozilla/5.0 (compatible; research-bot/1.0)"
    }
    response = requests.get(url, headers=headers, timeout=30)
    response.raise_for_status()
    return BeautifulSoup(response.text, "html.parser")
```

**For JavaScript-rendered pages:**

```python
from playwright.sync_api import sync_playwright

def fetch_js_page(url):
    with sync_playwright() as p:
        browser = p.chromium.launch(headless=True)
        page = browser.new_page()
        page.goto(url, wait_until="networkidle")
        content = page.content()
        browser.close()
    return BeautifulSoup(content, "html.parser")
```

### Step 3: Extract data with CSS selectors

Identify the right selectors by inspecting the page structure:

```python
def extract_items(soup, selectors):
    items = []
    containers = soup.select(selectors["container"])

    for container in containers:
        item = {}
        for field, selector in selectors["fields"].items():
            el = container.select_one(selector)
            if el:
                if el.name == "img":
                    item[field] = el.get("src", "")
                elif el.name == "a":
                    item[field] = {"text": el.get_text(strip=True), "href": el.get("href", "")}
                else:
                    item[field] = el.get_text(strip=True)
            else:
                item[field] = None
        items.append(item)

    return items
```

**Usage example:**

```python
selectors = {
    "container": "div.product-card",
    "fields": {
        "name": "h2.product-title",
        "price": "span.price",
        "rating": "span.rating-value",
        "link": "a.product-link",
    }
}
items = extract_items(soup, selectors)
```

### Step 4: Handle pagination

```python
def scrape_all_pages(base_url, selectors, max_pages=10):
    all_items = []
    for page_num in range(1, max_pages + 1):
        url = f"{base_url}?page={page_num}"
        soup = fetch_page(url)
        items = extract_items(soup, selectors)
        if not items:
            break
        all_items.extend(items)
        print(f"Page {page_num}: {len(items)} items (total: {len(all_items)})")
    return all_items
```

### Step 5: Output structured data

```python
import json
import csv

def save_json(data, filename):
    with open(filename, "w") as f:
        json.dump(data, f, indent=2)

def save_csv(data, filename):
    if not data:
        return
    with open(filename, "w", newline="") as f:
        writer = csv.DictWriter(f, fieldnames=data[0].keys())
        writer.writeheader()
        writer.writerows(data)
```

## Examples

### Example 1: Extract product listings

**User request:** "Scrape all product names and prices from this catalog page"

**Script outline:**
```python
soup = fetch_page("https://example-store.com/catalog")

products = []
for card in soup.select("div.product-item"):
    name = card.select_one("h3.title")
    price = card.select_one("span.price")
    products.append({
        "name": name.get_text(strip=True) if name else "N/A",
        "price": price.get_text(strip=True) if price else "N/A",
    })

save_csv(products, "products.csv")
print(f"Extracted {len(products)} products")
```

**Output:**
```
Extracted 48 products
Saved to products.csv

Preview:
| name                  | price   |
|-----------------------|---------|
| Wireless Keyboard     | $49.99  |
| USB-C Hub 7-port      | $34.99  |
| Ergonomic Mouse       | $29.99  |
```

### Example 2: Extract a data table from a page

**User request:** "Pull the statistics table from this Wikipedia article"

**Script outline:**
```python
soup = fetch_page("https://en.wikipedia.org/wiki/Example_Article")

table = soup.select_one("table.wikitable")
headers = [th.get_text(strip=True) for th in table.select("tr:first-child th")]
rows = []
for tr in table.select("tr")[1:]:
    cells = [td.get_text(strip=True) for td in tr.select("td")]
    if len(cells) == len(headers):
        rows.append(dict(zip(headers, cells)))

save_json(rows, "table_data.json")
print(f"Extracted {len(rows)} rows with columns: {headers}")
```

**Output:**
```
Extracted 25 rows with columns: ['Year', 'Population', 'Growth Rate']
Saved to table_data.json
```

### Step 6: Transform and deduplicate

Clean raw scraped data before loading — normalize prices, deduplicate by content hash, validate required fields.

```python
import hashlib

def transform_products(raw_items):
    """Clean and deduplicate scraped product data."""
    seen_hashes = set()
    clean = []

    for item in raw_items:
        # Skip items missing required fields
        if not item.get("name") or not item.get("price"):
            continue

        # Normalize price: "$1,299.99" → 129999 (cents)
        price_str = item["price"].replace(",", "").replace("$", "").strip()
        try:
            price_cents = int(float(price_str) * 100)
        except ValueError:
            continue

        # Deduplicate by content hash
        content_hash = hashlib.md5(
            f"{item['name']}|{item.get('link', '')}".encode()
        ).hexdigest()

        if content_hash in seen_hashes:
            continue
        seen_hashes.add(content_hash)

        clean.append({
            "name": item["name"][:500],
            "price_cents": price_cents,
            "url": item.get("link", {}).get("href", ""),
            "rating": item.get("rating"),
            "content_hash": content_hash,
            "scraped_at": datetime.utcnow().isoformat(),
        })

    return clean
```

### Step 7: Load into a database

Batch upsert into Postgres/Supabase for persistent storage with automatic price change tracking.

```python
from supabase import create_client

def load_to_supabase(products, supabase_url, supabase_key):
    """Batch upsert products into Supabase with conflict handling."""
    client = create_client(supabase_url, supabase_key)

    # Upsert in batches of 100
    for i in range(0, len(products), 100):
        batch = products[i:i+100]
        client.table("products").upsert(
            batch,
            on_conflict="content_hash"  # Update if exists
        ).execute()
        print(f"  Loaded batch {i//100 + 1}: {len(batch)} records")

    return len(products)
```

### Step 8: Handle anti-bot protection with Bright Data

For sites that block datacenter IPs, use Bright Data's residential proxy or Web Unlocker.

```python
import requests

def fetch_with_proxy(url, bright_data_config):
    """Fetch a page through Bright Data residential proxy."""
    proxy_url = (
        f"http://{bright_data_config['customer']}"
        f"-zone-{bright_data_config['zone']}"
        f":{bright_data_config['password']}"
        f"@brd.superproxy.io:22225"
    )
    response = requests.get(
        url,
        proxies={"http": proxy_url, "https": proxy_url},
        headers={"User-Agent": "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7)"},
        timeout=30,
    )
    response.raise_for_status()
    return BeautifulSoup(response.text, "html.parser")
```

## Guidelines

- Always set a User-Agent header. Requests without one are often blocked.
- Add a timeout (30 seconds) to all HTTP requests to avoid hanging.
- Respect `robots.txt`. Check it before scraping and honor disallow rules.
- Add delays between requests when scraping multiple pages (`time.sleep(1)`). Do not hammer servers.
- Handle missing elements gracefully. Not every item will have every field. Use `None` for missing values.
- If a selector returns no results, the page structure may have changed. Report this to the user rather than returning empty data silently.
- For pages behind login walls, inform the user that authentication is required and ask for guidance.
- Prefer CSS selectors over XPath. They are more readable and sufficient for most cases.
- When scraping tables, always validate that row cell counts match header counts before zipping.
- Output data in the format the user needs. Default to JSON for structured data and CSV for tabular data.
- For recurring scraping, use upsert (ON CONFLICT) to avoid duplicates across runs.
- Store prices as integers in cents — avoid floating-point rounding errors.
- Use Bright Data residential proxies for sites that block datacenter IPs. Budget ~$0.50/GB.
- Separate extract, transform, and load stages so each can fail and retry independently.
- Track content hashes for deduplication — same product from different scrape runs should update, not duplicate.