Web Search Skill

This skill enables Claude to search the web using DuckDuckGo to access current information, recent events, and real-time data beyond the training cutoff.

When to Use

Use this skill when you need:

• Current events and breaking news — Latest happenings, political updates, global events
• Recent information after training cutoff — New releases, updates, or changes since training data
• Real-time data — Current weather, sports scores, stock prices, cryptocurrency values
• Fact-checking verifiable claims — Confirm accuracy of statements, dates, statistics
• Fast-changing topics — Tech releases, API documentation updates, library versions
• Finding recent examples or tutorials — Latest best practices, code examples, how-to guides
• Availability checks — Is a service up? Is an event still happening?
• Contact information — Business hours, addresses, phone numbers for services

When NOT to Use

Avoid using this skill for:

• Basic knowledge questions — Information well within training data (historical facts, established concepts)
• Math or logic problems — Calculations, reasoning tasks that don't require external data
• Coding help — Writing code, debugging, or explaining programming concepts (unless searching for recent API changes)
• Creative writing or brainstorming — Tasks requiring imagination rather than facts
• Deep analysis — Tasks requiring understanding beyond surface-level facts from search results
• Private or local information — User's personal files, codebase details, local configurations
• Opinion or subjective matters — When training data provides sufficient context for reasoned opinions

Usage

The skill is invoked by running the search script with a query:

python ~/.openclaw/skills/web-search/scripts/search.py "your search query" [options]

Options

• --max-results N

  — Number of results to return (default: 5, max: 10)

• --time-range d|w|m|y

  — Filter results by time: d=day, w=week, m=month, y=year

• --news

  — Search news sources instead of general web

• --region CODE

  — Region code for localized results (default: "us-en")

Common Region Codes

• us-en

  — United States (English)

• uk-en

  — United Kingdom (English)

• ca-en

  — Canada (English)

• au-en

  — Australia (English)

• de-de

  — Germany (German)

• fr-fr

  — France (French)

• jp-jp

  — Japan (Japanese)

Examples

1. Check Current Weather

python ~/.openclaw/skills/web-search/scripts/search.py "current weather Hartford CT"

2. Find Recent News

python ~/.openclaw/skills/web-search/scripts/search.py "latest AI developments" --news --max-results 3

3. Search for Recent Releases

python ~/.openclaw/skills/web-search/scripts/search.py "Python 3.13 release notes" --time-range m

4. Get Stock Information

python ~/.openclaw/skills/web-search/scripts/search.py "AAPL stock price"

5. Find API Documentation

python ~/.openclaw/skills/web-search/scripts/search.py "OpenAI API rate limits 2026"

Workflow

When using this skill, follow this workflow:

1. Run the search — Execute the script with your query
2. Parse JSON output — The script outputs structured JSON with results
3. Synthesize results — Read through results and extract relevant information
4. Attribute sources — When presenting findings, cite sources with URLs
5. Refine if needed — If results are insufficient, try:
   • Different keywords (more specific or more general)
   • Time range filtering (--time-range)
   • News search (--news flag)
   • Multiple searches for complex topics

JSON Output Structure

The script returns JSON in this format:

{
  "results": [
    {
      "title": "Page Title",
      "url": "https://example.com/page",
      "snippet": "Preview text from the page..."
    }
  ],
  "query": "original search query",
  "result_count": 5
}

Attribution Guidelines

When presenting search results to users:

• Always include sources — Provide URLs in markdown format:

  [Title](URL)

• Cite multiple sources — For important claims, reference 2-3 sources
• Format sources section — End responses with a "Sources:" section listing all URLs
• Be transparent — Make it clear when information comes from search results vs. training data

Example:

Based on recent reports, Python 3.13 was released in October 2024 with
several performance improvements and new features.

Sources:
- [Python 3.13 Release Notes](https://docs.python.org/3/whatsnew/3.13.html)
- [Python Insider: Python 3.13.0 is now available](https://blog.python.org/...)

Best Practices

Query Formulation

• Keep queries short — 1-6 words works best (e.g., "React hooks documentation")
• Use specific keywords — Instead of "how to make cake", use "chocolate cake recipe"
• Include year for recent info — "TypeScript features 2026" vs. "TypeScript features"
• Use quotation marks for phrases — Not in the script args, but in the query string itself

Time Filtering

• Use

  --time-range d

  for breaking news or very recent events
• Use

  --time-range w

  for current events and weekly updates
• Use

  --time-range m

  for recent developments and monthly changes
• Use

  --time-range y

  for annual updates and year-over-year changes

Multiple Searches

For complex questions, break them into multiple searches:

• Bad: "What are the differences between React and Vue and which is better for large projects"
• Good:
  1. "React vs Vue comparison 2026"
  2. "React large projects scalability"
  3. "Vue large applications performance"

Result Verification

• Cross-reference — Check multiple sources for important facts
• Check dates — Ensure information is current enough for the query
• Assess authority — Official docs, established news sites, and recognized experts are more reliable
• Be skeptical — Search results may contain outdated, biased, or incorrect information

Error Handling

The script handles common errors gracefully:

Network Failures

If the search fails due to network issues:

• Error message will be printed to stderr
• Exit code will be non-zero
• Retry after a moment or check internet connection

Rate Limiting

DuckDuckGo may rate-limit requests:

• Wait 30-60 seconds before retrying
• Reduce search frequency if hitting limits repeatedly
• Consider alternative providers (see

  references/providers.md

  )

No Results Found

If a query returns no results:

• JSON will contain empty results array:

  {"results": [], ...}

• Try rephrasing the query with different keywords
• Remove time filters if using them
• Check for typos in the query

Import Errors

If

duckduckgo-search

library is missing:

• Script automatically attempts to install it
• Requires

  pip

  to be available
• May require

  --break-system-packages

  flag in some environments
• If auto-install fails, manually install:

  pip install duckduckgo-search

Technical Details

Dependencies

• Python 3.x — Required (python3 command must be available)
• duckduckgo-search — Auto-installed if missing
• pip — Required for dependency installation

Auto-Installation

The script automatically installs missing dependencies:

subprocess.check_call([
    sys.executable, "-m", "pip", "install",
    "duckduckgo-search", "--break-system-packages", "--quiet"
])

The

--break-system-packages

flag is needed for externally-managed Python environments (common on modern Linux distributions and macOS with Homebrew Python).

Provider Design

This skill uses a provider-agnostic architecture:

• Search provider can be swapped by replacing

  scripts/search.py

• CLI interface remains consistent (same arguments)
• JSON output format stays the same
• See

  references/providers.md

  for alternative providers

Exit Codes

• 0

  — Success

• 1

  — General error (network failure, invalid query, etc.)

• 2

  — Import error (dependency installation failed)

Limitations

DuckDuckGo Constraints

• Rate limiting — Frequent searches may be throttled
• No API key — Cannot authenticate for higher limits
• Basic search — Less sophisticated than paid APIs (no filters for academic papers, patents, etc.)
• No search history — Each search is independent
• Result quality — May be less relevant than Google/Bing for niche topics

General Limitations

• Surface-level information — Search results provide snippets, not deep analysis
• No PDF/file downloads — Only returns web page URLs, not file contents
• No image search — Text search only (web and news)
• English-centric — Best results for English queries, though region codes help
• Recency bias — Very recent events (< 1 hour) may not be indexed yet

Troubleshooting

Script Not Executable

If you get "Permission denied" error:

chmod +x ~/.openclaw/skills/web-search/scripts/search.py

Python Not Found

If

python3

command doesn't exist, create an alias or use full path to Python interpreter.

Installation Fails in Virtual Environment

If auto-install fails due to venv restrictions, manually install globally:

python3 -m pip install duckduckgo-search --user

No Results for Valid Query

• Try removing time filters
• Use more general keywords
• Check if the topic is too recent (< 1 hour)
• Verify spelling and phrasing

See Also

• references/providers.md

  — Alternative search providers and migration guide
• DuckDuckGo Search Python Library — Upstream library documentation

Notes

• The skill is designed to be proactive — Claude should use it whenever current information would improve the response
• Always cite sources when using search results
• For controversial or critical information, cross-reference multiple sources
• Privacy-respecting — DuckDuckGo doesn't track searches or build user profiles