agent-browser

name	agent-browser
description	Automates browser interactions for web testing, form filling, screenshots, and data extraction. Use when the user needs to navigate websites, interact with web pages, fill forms, take screenshots, test web applications, or extract information from web pages.
metadata	{"jaguarclaw":{"requires":{"anyBins":["agent-browser","agent-browser.cmd"]}}}

Browser Automation with agent-browser

Runtime Model

Desktop mode uses a bundled runtime:

• bundled agent-browser binary
• bundled Chromium kernel
• per-agent/per-session profile persistence via AGENT_BROWSER_PROFILE

In desktop mode, users should not need manual installation or PATH setup.

If this skill is unavailable, first check runtime diagnostics:

• AGENT_BROWSER_EXECUTABLE_PATH
• AGENT_BROWSER_CHROMIUM_PATH
• AGENT_BROWSER_KERNEL_HOME

---

Platform-Specific Notes

Windows Users:

• Use agent-browser.cmd instead of agent-browser in commands
• Or add .cmd extension: agent-browser.cmd open <url>
• Bundled runtime already includes compatible launcher/binary

All Platforms:

• Verify availability with: agent-browser --version

---

Quick start

agent-browser open <url>        # Navigate to page
agent-browser snapshot -i       # Get interactive elements with refs
agent-browser click @e1         # Click element by ref
agent-browser fill @e2 "text"   # Fill input by ref
agent-browser close             # Close browser

Windows: Replace agent-browser with agent-browser.cmd in all commands above.

Core workflow

1. Navigate: .\agent-browser.cmd open <url>
2. Snapshot: .\agent-browser.cmd snapshot -i (returns elements with refs like @e1, @e2)
3. Interact using refs from the snapshot
4. Re-snapshot after navigation or significant DOM changes

Commands

Navigation

.\agent-browser.cmd open <url>      # Navigate to URL
.\agent-browser.cmd back            # Go back
.\agent-browser.cmd forward         # Go forward
.\agent-browser.cmd reload          # Reload page
.\agent-browser.cmd close           # Close browser

Snapshot (page analysis)

.\agent-browser.cmd snapshot            # Full accessibility tree
.\agent-browser.cmd snapshot -i         # Interactive elements only (recommended)
.\agent-browser.cmd snapshot -c         # Compact output
.\agent-browser.cmd snapshot -d 3       # Limit depth to 3
.\agent-browser.cmd snapshot -s "#main" # Scope to CSS selector

Interactions (use @refs from snapshot)

.\agent-browser.cmd click @e1           # Click
.\agent-browser.cmd dblclick @e1        # Double-click
.\agent-browser.cmd focus @e1           # Focus element
.\agent-browser.cmd fill @e2 "text"     # Clear and type
.\agent-browser.cmd type @e2 "text"     # Type without clearing
.\agent-browser.cmd press Enter         # Press key
.\agent-browser.cmd press Control+a     # Key combination
.\agent-browser.cmd keydown Shift       # Hold key down
.\agent-browser.cmd keyup Shift         # Release key
.\agent-browser.cmd hover @e1           # Hover
.\agent-browser.cmd check @e1           # Check checkbox
.\agent-browser.cmd uncheck @e1         # Uncheck checkbox
.\agent-browser.cmd select @e1 "value"  # Select dropdown
.\agent-browser.cmd scroll down 500     # Scroll page
.\agent-browser.cmd scrollintoview @e1  # Scroll element into view
.\agent-browser.cmd drag @e1 @e2        # Drag and drop
.\agent-browser.cmd upload @e1 file.pdf # Upload files

Get information

.\agent-browser.cmd get text @e1        # Get element text
.\agent-browser.cmd get html @e1        # Get innerHTML
.\agent-browser.cmd get value @e1       # Get input value
.\agent-browser.cmd get attr @e1 href   # Get attribute
.\agent-browser.cmd get title           # Get page title
.\agent-browser.cmd get url             # Get current URL
.\agent-browser.cmd get count ".item"   # Count matching elements
.\agent-browser.cmd get box @e1         # Get bounding box

Check state

.\agent-browser.cmd is visible @e1      # Check if visible
.\agent-browser.cmd is enabled @e1      # Check if enabled
.\agent-browser.cmd is checked @e1      # Check if checked

Screenshots & PDF

.\agent-browser.cmd screenshot          # Screenshot to stdout
.\agent-browser.cmd screenshot path.png # Save to file
.\agent-browser.cmd screenshot --full   # Full page
.\agent-browser.cmd pdf output.pdf      # Save as PDF

Video recording

.\agent-browser.cmd record start ./demo.webm    # Start recording (uses current URL + state)
.\agent-browser.cmd click @e1                   # Perform actions
.\agent-browser.cmd record stop                 # Stop and save video
.\agent-browser.cmd record restart ./take2.webm # Stop current + start new recording

Recording creates a fresh context but preserves cookies/storage from your session. If no URL is provided, it automatically returns to your current page. For smooth demos, explore first, then start recording.

Wait

.\agent-browser.cmd wait @e1                     # Wait for element
.\agent-browser.cmd wait 2000                    # Wait milliseconds
.\agent-browser.cmd wait --text "Success"        # Wait for text
.\agent-browser.cmd wait --url "**/dashboard"    # Wait for URL pattern
.\agent-browser.cmd wait --load networkidle      # Wait for network idle
.\agent-browser.cmd wait --fn "window.ready"     # Wait for JS condition

Mouse control

.\agent-browser.cmd mouse move 100 200      # Move mouse
.\agent-browser.cmd mouse down left         # Press button
.\agent-browser.cmd mouse up left           # Release button
.\agent-browser.cmd mouse wheel 100         # Scroll wheel

Semantic locators (alternative to refs)

.\agent-browser.cmd find role button click --name "Submit"
.\agent-browser.cmd find text "Sign In" click
.\agent-browser.cmd find label "Email" fill "[e-mail]"
.\agent-browser.cmd find first ".item" click
.\agent-browser.cmd find nth 2 "a" text

Browser settings

.\agent-browser.cmd set viewport 1920 1080      # Set viewport size
.\agent-browser.cmd set device "iPhone 14"      # Emulate device
.\agent-browser.cmd set geo 37.7749 -122.4194   # Set geolocation
.\agent-browser.cmd set offline on              # Toggle offline mode
.\agent-browser.cmd set headers '{"X-Key":"v"}' # Extra HTTP headers
.\agent-browser.cmd set credentials user pass   # HTTP basic auth
.\agent-browser.cmd set media dark              # Emulate color scheme

Cookies & Storage

.\agent-browser.cmd cookies                     # Get all cookies
.\agent-browser.cmd cookies set name value      # Set cookie
.\agent-browser.cmd cookies clear               # Clear cookies
.\agent-browser.cmd storage local               # Get all localStorage
.\agent-browser.cmd storage local key           # Get specific key
.\agent-browser.cmd storage local set k v       # Set value
.\agent-browser.cmd storage local clear         # Clear all

Network

.\agent-browser.cmd network route <url>              # Intercept requests
.\agent-browser.cmd network route <url> --abort      # Block requests
.\agent-browser.cmd network route <url> --body '{}'  # Mock response
.\agent-browser.cmd network unroute [url]            # Remove routes
.\agent-browser.cmd network requests                 # View tracked requests
.\agent-browser.cmd network requests --filter api    # Filter requests

Tabs & Windows

.\agent-browser.cmd tab                 # List tabs
.\agent-browser.cmd tab new [url]       # New tab
.\agent-browser.cmd tab 2               # Switch to tab
.\agent-browser.cmd tab close           # Close tab
.\agent-browser.cmd window new          # New window

Frames

.\agent-browser.cmd frame "#iframe"     # Switch to iframe
.\agent-browser.cmd frame main          # Back to main frame

Dialogs

.\agent-browser.cmd dialog accept [text]  # Accept dialog
.\agent-browser.cmd dialog dismiss        # Dismiss dialog

JavaScript

.\agent-browser.cmd eval "document.title"   # Run JavaScript

Example: Form submission

.\agent-browser.cmd open https://example.com/form
.\agent-browser.cmd snapshot -i
# Output shows: textbox "Email" [ref=e1], textbox "Password" [ref=e2], button "Submit" [ref=e3]

.\agent-browser.cmd fill @e1 "[e-mail]"
.\agent-browser.cmd fill @e2 "password123"
.\agent-browser.cmd click @e3
.\agent-browser.cmd wait --load networkidle
.\agent-browser.cmd snapshot -i  # Check result

Example: Authentication with saved state

# Login once
.\agent-browser.cmd open https://app.example.com/login
.\agent-browser.cmd snapshot -i
.\agent-browser.cmd fill @e1 "username"
.\agent-browser.cmd fill @e2 "password"
.\agent-browser.cmd click @e3
.\agent-browser.cmd wait --url "**/dashboard"
.\agent-browser.cmd state save auth.json

# Later sessions: load saved state
.\agent-browser.cmd state load auth.json
.\agent-browser.cmd open https://app.example.com/dashboard

Sessions (parallel browsers)

.\agent-browser.cmd --session test1 open site-a.com
.\agent-browser.cmd --session test2 open site-b.com
.\agent-browser.cmd session list

JSON output (for parsing)

Add --json for machine-readable output:

.\agent-browser.cmd snapshot -i --json
.\agent-browser.cmd get text @e1 --json

Debugging

.\agent-browser.cmd open example.com --headed              # Show browser window
.\agent-browser.cmd console                                # View console messages
.\agent-browser.cmd errors                                 # View page errors
.\agent-browser.cmd record start ./debug.webm   # Record from current page
.\agent-browser.cmd record stop                            # Save recording
.\agent-browser.cmd open example.com --headed  # Show browser window
.\agent-browser.cmd --cdp 9222 snapshot        # Connect via CDP
.\agent-browser.cmd console                    # View console messages
.\agent-browser.cmd console --clear            # Clear console
.\agent-browser.cmd errors                     # View page errors
.\agent-browser.cmd errors --clear             # Clear errors
.\agent-browser.cmd highlight @e1              # Highlight element
.\agent-browser.cmd trace start                # Start recording trace
.\agent-browser.cmd trace stop trace.zip       # Stop and save trace