data-base

Mental Model

Data acquisition is converting unstructured web content into structured data. Choose tool based on page complexity: JS-heavy → chrome-devtools MCP, static → Python requests.

Tool Selection

Page Type	Tool	When to Use
Dynamic (JS-rendered, SPAs)	chrome-devtools MCP	React/Vue apps, infinite scroll, login gates
Static HTML	Python requests	Blogs, news sites, simple pages
Complex/reusable logic	Python script	Multi-step scraping, rate limiting, proxies

Anti-Patterns (NEVER)

• Don't scrape without checking robots.txt
• Don't overload servers (default: 1 req/sec)
• Don't scrape personal data without consent
• Don't use Chinese characters in output filenames (ASCII only)
• Don't forget to identify bot with User-Agent

Output Format

• JSON: Nested/hierarchical data
• CSV: Tabular data
• Filename: {source}_{timestamp}.{ext} (ASCII only, e.g., news_20250115.csv)

Workflow

1. Ask: What data? Which sites? How much?
2. Select tool based on page type
3. Extract and save structured data
4. Deliver file path to user or pass to data-analysis

Python Environment

Auto-initialize virtual environment if needed, then execute:

cd skills/data-base

if [ ! -f ".venv/bin/python" ]; then
    echo "Creating Python environment..."
    ./setup.sh
fi

.venv/bin/python your_script.py

The setup script auto-installs: requests, beautifulsoup4, pandas, web scraping tools.

References (load on demand)

For detailed APIs and templates, load: references/REFERENCE.md, references/templates.md