wb

name	wb
description	Use the wb CLI to browse the web, search, read pages, and interact with websites. Use when the user asks to look something up online, visit a website, extract content from a page, or interact with web elements.
allowed-tools	Bash(wb *)

wb — Web Browsing CLI

wb provides web browsing through a shared daemon session. Use it to search the web, read pages, and interact with websites.

Session Setup

If no session is running (wb tab list fails), create one:

wb session create -c ~/cookies.json -d

This loads saved cookies and opens a visible browser window.

If a session is misbehaving (commands hang, errors), delete and recreate:

wb session delete
wb session create -c ~/cookies.json -d

Only do this if commands are failing — it kills all open tabs.

Important: Shared Session

You are using the default session which may be shared with other agents. Always:

1. Create a tab with a unique name before browsing
2. Close your tab when you're done
3. Never close tabs you didn't create

Use a descriptive tab name that identifies your task (e.g., search-react-hooks, read-docs-api).

Quick Start

# Create a tab and navigate to a page
wb tab new <tab-name> <url>

# Read page content
wb tab dump <tab-name>

# Close when done
wb tab close <tab-name>

Searching the Web

Use Google search to find information:

wb tab new my-search "https://www.google.com/search?q=your+search+query"
wb tab dump my-search
# Read the results
wb tab close my-search

# Possibly navigate to a result as needed
wb tab new read-result <result-url>
wb tab dump read-result
wb tab close read-result

Always URL-encode the search query (replace spaces with +).

Commands Reference

Tab Management

wb tab list                           # List all open tabs
wb tab new <name> <url>               # Create tab and navigate
wb tab close <name>                   # Close tab (always clean up!)
wb tab go <name> <url>                # Navigate tab to a new URL

Reading Content

wb tab dump <name>                    # Dump page as markdown (default limit: 8192 chars)
wb tab dump <name> -o 8192            # Dump starting at offset 8192 (pagination)
wb tab dump <name> -h                 # Dump as raw HTML
wb tab dump <name> -l 16384           # Dump with larger limit

The dump output includes a header:

---
url: https://example.com
offset: 0
limit: 8192
remaining: 4521
---

Use remaining to know if there's more content. Paginate with -o <offset> if needed.

Interacting with Pages

wb tab act <name> -a <instructions>       # Perform an action (click, type, etc.)
wb tab observe <name> <instructions>      # List available actions matching instructions

act uses AI to perform browser actions like clicking links, filling forms, pressing buttons. If act fails, try observe first to see what actions are available, then act on a more specific instruction.

Examples:

wb tab act my-tab -a "click the Sign In button"
wb tab act my-tab -a "type hello into the search box"
wb tab observe my-tab "find all navigation links"

Tips

• Dump is paginated: if remaining > 0 in the header, use -o to get the next chunk
• Act with observe fallback: act internally falls back to observe + act if direct action fails
• SVGs are stripped: dump output automatically removes SVG noise
• Google search: always use https://www.google.com/search?q=... for web searches