web-scraper
由 guifav 提供
Web & Frontend Development简介
Web scraping and content comprehension agent — multi-strategy extraction with cascade fallback, news detection.
描述
该技能的功能
A 5-stage web scraping pipeline that detects news articles, extracts content using a cascade of strategies (static HTTP, Playwright, Scrapy), removes boilerplate, and pulls structured metadata. Handles JS-rendered pages, soft paywalls, bulk crawls, and encoding normalization. Optionally extracts named entities via LLM using clean text only, never raw HTML.
The cascade fallback (static HTTP first, Playwright only when needed) avoids headless browser overhead for simple pages while still handling JS-rendered SPAs that single-strategy scrapers fail on.
何时使用
- Extracting full article text and metadata from a news site
- Bulk crawling all articles on a domain into structured JSON
- Scraping content from a React or Next.js SPA that requires JS rendering
- Extracting people, organizations, and events mentioned in news articles
- Checking which parts of a paywalled article are publicly accessible
示例工作流
User asks: Scrape this news article URL and extract all named entities
- 1
Analyzes URL pattern and page HTML to confirm it is a news article (Stage 1)
- 2
Fetches page with static HTTP; detects JS rendering need and escalates to Playwright (Stage 2)
- 3
Removes boilerplate with trafilatura and normalizes encoding (Stage 3)
- 4
Extracts structured metadata: title, author, publication date, paywall status (Stage 4)
- 5
Sends clean text to LLM via OpenRouter and parses entity JSON response (Stage 5)
JSON file with article text, metadata fields, and structured entities grouped by people, organizations, locations, and events
前置要求
- OPENROUTER_API_KEY environment variable (for LLM entity extraction in Stage 5)
- Playwright Chromium browser installed via 'playwright install chromium'
安装
clawhub install guifav/web-scraper运行 CLI 命令,或把这段提示词粘贴到 Claude Code、Cursor、Codex 或你在用的任意编程智能体里。