---

name: web-fetch description: Fetches web content with intelligent content extraction, converting HTML to clean markdown. Use for documentation, articles, and reference pages http/https URLs.

Web Content Fetching

Fetch web content using curl | html2markdown with CSS selectors for clean, complete markdown output.

Quick Usage (Known Sites)

Use site-specific selectors for best results:

Copy# Anthropic docs
curl -s "<url>" | html2markdown --include-selector "#content-container"

# MDN Web Docs
curl -s "<url>" | html2markdown --include-selector "article"

# GitHub docs
curl -s "<url>" | html2markdown --include-selector "article" --exclude-selector "nav,.sidebar"

# Generic article pages
curl -s "<url>" | html2markdown --include-selector "article,main,[role=main]" --exclude-selector "nav,header,footer"

Site Patterns

Universal Fallback (Unknown Sites)

For sites without known patterns, use the Bun script which auto-detects content:

Copybun ~/.claude/skills/web-fetch/fetch.ts "<url>"

Setup (one-time)

Copycd ~/.claude/skills/web-fetch && bun install

Finding the Right Selector

When a site isn't in the patterns list:

Copy# Check what content containers exist
curl -s "<url>" | grep -o '<article[^>]*>\|<main[^>]*>\|id="[^"]*content[^"]*"' | head -10

# Test a selector
curl -s "<url>" | html2markdown --include-selector "<selector>" | head -30

# Check line count
curl -s "<url>" | html2markdown --include-selector "<selector>" | wc -l

Options Reference

Copy--include-selector "CSS"  # Only include matching elements
--exclude-selector "CSS"  # Remove matching elements
--domain "https://..."    # Convert relative links to absolute

Comparison

Troubleshooting

Wrong content selected: The site may have multiple articles. Inspect the HTML:

Copycurl -s "<url>" | grep -o '<article[^>]*>'

Empty output: The selector doesn't match. Try broader selectors like main or body.

Missing code blocks: Check if the site uses non-standard code formatting.

Client-rendered content: If HTML only has "Loading..." placeholders, the content is JS-rendered. Neither curl nor the Bun script can extract it; use browser-based tools.