Browser Skill

Control the user's real Chrome browser via Chrome Relay CLI commands. This gives you access to their existing sessions, cookies, and logins — no need to log in again.

All browser operations use alma browser CLI commands via Bash.

Available Commands

Workflow

1. Check connection: alma browser status — confirm Chrome Relay is connected
2. List tabs: alma browser tabs — see what's already open (note the [tabId])
3. Navigate: alma browser goto <tabId> <url> or alma browser open <url> for a new tab
4. Discover elements: alma browser read-dom <tabId> — find interactive elements with CSS selectors
5. Read content: alma browser read <tabId> — extract text as markdown
6. Interact: alma browser click, alma browser type, alma browser eval
7. Verify: alma browser screenshot <tabId> then Read the .png file to see the result

When to Use This vs WebFetch/WebSearch

• Use this skill when you need to: fill forms, click buttons, navigate multi-step flows, log in, interact with web apps, or read a page the user already has open
• Use WebFetch when you just need to read a public page's content (faster, no browser needed)
• Use WebSearch when you just need to search the web

Finding Elements

To click or type, you need CSS selectors:

• Run alma browser read-dom <tabId> — returns all interactive elements with CSS selector hints
• Take a screenshot to visually understand the layout
• Common selectors: input[name="email"], button[type="submit"], a[href*="login"], #search-input, .submit-btn

Example: Search on a Site

# 1. Find existing tabs
alma browser tabs

# 2. Open a new tab (note the tab ID in output)
alma browser open https://example.com

# 3. Discover form elements
alma browser read-dom 123

# 4. Type into search and press Enter
alma browser type 123 "input[name='q']" "search query" --enter

# 5. Read the results
alma browser read 123

Example: Multi-Step Form

# 1. Open the page
alma browser open https://example.com/register

# 2. Discover form fields
alma browser read-dom 123

# 3. Fill in fields
alma browser type 123 "#name" "John Doe"
alma browser type 123 "#email" "john@example.com"

# 4. Submit
alma browser click 123 "button[type='submit']"

# 5. Verify — screenshot then read the image
alma browser screenshot 123
# Use Read tool on the printed file path to see the screenshot

Example: Reading Browser Content

# List tabs to find the one the user is looking at
alma browser tabs

# Read page content as markdown
alma browser read 456

Tips

• Always run alma browser tabs first to find existing tabs — don't create unnecessary new tabs
• For reading text content, prefer alma browser read over alma browser screenshot
• After alma browser screenshot, use the Read tool on the returned file path to view the image
• For advanced interactions (dropdowns, scrolling to specific elements), use alma browser eval
• Tab IDs are shown in brackets like [123] in the tabs listing output