Web Search Skill

Search the web. Prefer the built-in WebSearch tool — it uses Electron’s built-in BrowserWindow to perform searches, so it can handle JavaScript-rendered search results and bypass basic anti-bot measures. This is NOT a simple HTTP request — it actually opens the page in a real browser window, performs the search, and extracts results from the rendered page. Fall back to curl only if WebSearch is unavailable.

DuckDuckGo Instant Answer API (No API key needed)

# Basic search — returns instant answers
curl -s "https://api.duckduckgo.com/?q=QUERY&format=json&no_html=1" | jq '.Abstract, .RelatedTopics[:5]'

DuckDuckGo HTML Search (Full results)

# Get search results page and extract links
curl -s "https://html.duckduckgo.com/html/?q=QUERY" | grep -o 'href="https\{0,1\}://[^"]*' | head -10

Google Search via SerpAPI (if API key available)

# Check if SERPAPI_KEY is set
alma config get serpapi.apiKey

# If available:
curl -s "https://serpapi.com/search.json?q=QUERY&api_key=API_KEY" | jq '.organic_results[:5] | .[] | {title, link, snippet}'

Fetching Page Content

After finding URLs from search, use the WebFetch tool to get the actual page content:

WebFetch(url="https://example.com/article")

Or via curl:

curl -sL "https://example.com/article" | head -200

Tips

• URL-encode the query: replace spaces with + or %20
• Use jq to parse JSON responses
• For complex queries, try multiple search approaches
• Always summarize findings for the user rather than dumping raw results
• If DuckDuckGo doesn’t have good results, try fetching specific known sources directly
• For freshness queries (today, latest, 今天, 最新), use the current year from runtime context instead of hardcoding a fixed year

Examples

“Latest AI news”:

curl -s "https://html.duckduckgo.com/html/?q=latest+AI+news+<current_year>" | grep -o 'href="https\{0,1\}://[^"]*' | grep -v duckduckgo | head -5

“Python 3.13 new features”:

curl -s "https://api.duckduckgo.com/?q=python+3.13+new+features&format=json&no_html=1" | jq '.Abstract'