Agent Browser Automation

agent-browser

> Skill by ara.so — Daily 2026 Skills collection.

agent-browser is a headless browser automation CLI built in Rust, designed for AI agents. It wraps Chrome via the Chrome DevTools Protocol (CDP) and exposes a fast, ergonomic command-line interface for navigation, interaction, accessibility snapshots, screenshots, network interception, and more — with no Node.js or Playwright runtime required.

Installation

Recommended (npm global)

npm install -g agent-browser
agent-browser install  # Download Chrome for Testing (first time only)

macOS (Homebrew)

brew install agent-browser
agent-browser install

Rust / Cargo

cargo install agent-browser
agent-browser install

Local project dependency

npm install agent-browser
# Add to package.json scripts or invoke via npx

Linux (with system dependencies)

agent-browser install --with-deps

Quick Start

agent-browser open https://example.com
agent-browser snapshot                        # Accessibility tree with @refs (best for AI)
agent-browser click @e2                       # Click by ref from snapshot
agent-browser fill @e3 "hello@example.com"   # Fill by ref
agent-browser get text @e1                    # Get text content
agent-browser screenshot page.png
agent-browser close

Core Commands

Navigation

agent-browser open            # Navigate (aliases: goto, navigate)
agent-browser get url              # Get current URL
agent-browser get title            # Get page title
agent-browser close                # Close browser (aliases: quit, exit)

Accessibility Snapshot (recommended for AI agents)

agent-browser snapshot             # Returns accessibility tree with @ref IDs
agent-browser snapshot -i          # Interactive / compact mode

Snapshot output includes @eN refs you can use directly:

@e1 [button] "Submit"
@e2 [textbox] "Email" value=""
@e3 [link] "Sign in"

Then act on them:

agent-browser fill @e2 "user@example.com"
agent-browser click @e1

Interaction

agent-browser click                      # Click element
agent-browser dblclick                   # Double-click
agent-browser fill                 # Clear and fill input
agent-browser type                 # Type into element
agent-browser press                      # Press key (Enter, Tab, Control+a)
agent-browser keyboard type             # Type at current focus (real keystrokes)
agent-browser keyboard inserttext       # Insert text without key events
agent-browser hover                      # Hover element
agent-browser select              # Select dropdown option
agent-browser check                      # Check checkbox
agent-browser uncheck                    # Uncheck checkbox
agent-browser scroll down 500                 # Scroll (up/down/left/right, optional px)
agent-browser scroll down --selector "#feed"  # Scroll within element
agent-browser scrollintoview             # Scroll element into view
agent-browser drag               # Drag and drop
agent-browser upload  /path/file.pdf     # Upload file

Screenshots & PDF

agent-browser screenshot                          # Save to temp dir, print path
agent-browser screenshot page.png                 # Save to path
agent-browser screenshot --full page.png          # Full-page screenshot
agent-browser screenshot --annotate               # Numbered element labels overlay
agent-browser screenshot --screenshot-dir ./shots # Custom output directory
agent-browser screenshot --screenshot-format jpeg --screenshot-quality 80
agent-browser pdf output.pdf                      # Save page as PDF

Getting Element Info

agent-browser get text            # Text content
agent-browser get html            # innerHTML
agent-browser get value           # Input value
agent-browser get attr      # Attribute value
agent-browser get count           # Count matching elements
agent-browser get box             # Bounding box
agent-browser get styles          # Computed styles
agent-browser get cdp-url              # CDP WebSocket URL

State Checks

agent-browser is visible 
agent-browser is enabled 
agent-browser is checked 

Semantic Locators (find)

agent-browser find role button click --name "Submit"
agent-browser find text "Sign In" click
agent-browser find label "Email" fill "test@example.com"
agent-browser find placeholder "Search..." fill "rust"
agent-browser find testid "login-btn" click
agent-browser find first ".item" click
agent-browser find nth 2 "a" text
agent-browser find role textbox fill "hello" --name "Username"

Actions: click, fill, type, hover, focus, check, uncheck, text

Waiting

agent-browser wait "#modal"                          # Wait for element visible
agent-browser wait 2000                              # Wait N milliseconds
agent-browser wait --text "Welcome back"             # Wait for text
agent-browser wait --url "**/dashboard"              # Wait for URL pattern
agent-browser wait --load networkidle                # Wait for load state
agent-browser wait --fn "window.appReady === true"   # Wait for JS condition
agent-browser wait "#spinner" --state hidden         # Wait for element to disappear

Load states: load, domcontentloaded, networkidle

JavaScript Eval

agent-browser eval "document.title"
agent-browser eval "JSON.stringify(window.__STATE__)"
agent-browser eval -b "BASE64_ENCODED_JS"
echo "return document.body.innerHTML" | agent-browser eval --stdin

Batch Execution (efficient multi-step)

echo '[
  ["open", "https://example.com"],
  ["snapshot", "-i"],
  ["fill", "@e2", "user@example.com"],
  ["click", "@e1"],
  ["screenshot", "result.png"]
]' | agent-browser batch --json

# Stop on first failure
agent-browser batch --bail  commands.json << 'EOF'
[
  ["open", "https://news.ycombinator.com"],
  ["wait", "--load", "networkidle"],
  ["get", "title"],
  ["snapshot"],
  ["screenshot", "hn.png"]
]
EOF

agent-browser batch --json < commands.json

Scrape with mocked network

agent-browser open https://api-heavy-app.example.com
agent-browser network route "**/api/slow-endpoint" --body '{"data":"mocked"}'
agent-browser snapshot
agent-browser network unroute

Full-page screenshot with annotations

agent-browser open https://example.com
agent-browser wait --load networkidle
agent-browser screenshot --full --annotate annotated.png

Connect to already-running Chrome

# Start Chrome with remote debugging
google-chrome --remote-debugging-port=9222 &

agent-browser connect 9222
agent-browser open https://example.com
agent-browser snapshot

Emulate mobile device

agent-browser set device "iPhone 14"
agent-browser open https://example.com
agent-browser screenshot mobile.png

HAR recording for network analysis

agent-browser open https://example.com
agent-browser network har start
agent-browser click "#load-data"
agent-browser wait --load networkidle
agent-browser network har stop session.har

Selector Reference

| Format | Example | Notes | |--------|---------|-------| | @ref | @e1, @e12 | From snapshot output — preferred for AI | | CSS | #id, .class, [attr=val] | Standard CSS selectors | | Text | "Sign In" | Exact text match | | XPath | //button[@type='submit'] | Full XPath |

Troubleshooting

Chrome not found

agent-browser install              # Downloads Chrome for Testing
agent-browser install --with-deps  # Linux: also installs system libs

Element not found / timing issues

agent-browser wait "#my-element"              # Wait for visibility first
agent-browser wait --load networkidle         # Wait for page to settle
agent-browser wait --fn "!!document.querySelector('#app')"

Selector issues — use snapshot refs instead

# Instead of fragile CSS:
agent-browser click ".btn.btn-primary.submit-form"

# Use snapshot refs:
agent-browser snapshot  # Find @e7 = [button] "Submit"
agent-browser click @e7

Debug what's on the page

agent-browser screenshot debug.png        # Visual check
agent-browser snapshot                    # Accessibility tree
agent-browser console                     # JS console output
agent-browser errors                      # Uncaught exceptions
agent-browser eval "document.readyState"

Auth issues between sessions

agent-browser state save ./auth.json   # After successful login
agent-browser state load ./auth.json   # At start of next session

Handling alerts/dialogs

# Set up handler BEFORE the action that triggers dialog
agent-browser dialog accept
agent-browser click "#delete-button"

Performance — use batch for multi-step workflows

# Slow: one process per command
agent-browser open https://example.com
agent-browser fill "#q" "search"
agent-browser click "#submit"

# Fast: single process, multiple commands
echo '[["open","https://example.com"],["fill","#q","search"],["click","#submit"]]' \
  | agent-browser batch --json

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: zhouguoqing
• Source: zhouguoqing/QianYuan.AIAgenticFramework
• License: Apache-2.0

Install and usage instructions live in the source repository linked above.