| name | Web Browsing |
| description | Search the web via DuckDuckGo, fetch and extract content from URLs, and look up reference material. Use for real-time information, external documentation, and validating facts against public sources. |
| version | 1.0.0 |
| tools | ["web_search","web_fetch"] |
| context | 4096 |
Web Browsing Skill
Provides access to public web content via search and URL fetching.
Operations
Search
Query DuckDuckGo for relevant results. Use when you need to find sources, validate claims, or discover documentation.
Input: Natural-language query (be specific; avoid vague terms)
Output: List of results with title, URL, and excerpt
Usage pattern:
SEARCH: <specific query with key terms>
→ Review results, select the most authoritative source
→ FETCH the URL if you need full content
Fetch
Retrieve the full text content of a URL. Use after a Search to read a specific page.
Input: Valid URL
Output: Page content as plain text (HTML stripped)
Usage pattern:
FETCH: https://example.com/page
→ Extract the relevant section
→ Cite the source URL in your response
Reference Lookup
For well-known documentation (MDN, Bun docs, Ollama API), fetch directly without searching first.
When to Use This Skill
- You need current information not in your training data
- You need to verify a specific fact, version number, or specification
- You need to read external documentation before making a technical recommendation
Anti-Patterns
- Do not search for information you already know with high confidence
- Do not fetch URLs that require authentication — they will return login pages
- Do not use web search as a substitute for reasoning — it supplements, not replaces
- Do not make more than 3 search/fetch calls per task without reassessing necessity
- Do not store raw fetched content in memory — extract and summarize instead
Output Citation Format
When referencing web-sourced content:
Source: Page Title — retrieved {{DATE}}
