Web Research

Headless web search and content extraction via Brave Search. Lightweight alternative to browser-based search.

When to use this skill

• User asks to research a topic
• Need to find documentation or tutorials
• Want to extract content from URLs
• Gathering information for analysis
• Need web search without browser overhead

Commands

All commands are in PATH. Invoke directly via Bash:

Search the Web

# Basic search (returns 5 results)
~/agent-tools/brave-search/search.js "your query here"

# More results
~/agent-tools/brave-search/search.js "query" -n 10

# Search AND fetch content from each result (powerful for research)
~/agent-tools/brave-search/search.js "query" -n 3 --content

Extract Page Content

# Convert any URL to readable markdown
~/agent-tools/brave-search/content.js https://example.com/article

Output Format

--- Result 1 ---
Title: Page Title
Link: https://example.com/page
Snippet: Description from search results
Content: (only with --content flag)
  Markdown content extracted from the page...

--- Result 2 ---
...

Usage Patterns

Quick Research

# Get overview of a topic
~/agent-tools/brave-search/search.js "react server components best practices" -n 5

Deep Research

# Search and extract all content in one shot
~/agent-tools/brave-search/search.js "typescript discriminated unions" -n 3 --content

Selective Deep Dive

# Search first
~/agent-tools/brave-search/search.js "rust error handling" -n 10

# Then extract specific interesting results
~/agent-tools/brave-search/content.js https://doc.rust-lang.org/book/ch09-00-error-handling.html

Documentation Lookup

# Find and extract docs
~/agent-tools/brave-search/search.js "site:docs.anthropic.com tool use" --content

Comparison with WebSearch Tool

Feature	web-research (this)	WebSearch (built-in)
Content extraction	Yes (--content)	No
Selective URL fetch	Yes (content.js)	No
Result count control	Yes (-n)	Limited
Works offline	No	No
Token efficiency	Better (text only)	Embedded in response

Use this skill when you need to extract actual page content, not just search snippets.

Tips

• Use --content sparingly (fetches each page, slower)
• For single pages, content.js is faster than search
• Site-specific search: "site:github.com topic"
• Combine with grep for filtering: search.js "query" | grep -i keyword

Error Handling

• Network errors: Retry once, then report
• Blocked pages: Some sites block non-browser requests; fall back to browser-tools if needed
• Empty results: Try broader query or different keywords