Browser Debugging with agent-browser
AI-optimized browser automation for visual debugging.
Commands
bash
# Navigate to a URL agent-browser navigate http://localhost:3000 # Take screenshot agent-browser screenshot # Get accessibility tree (AI-optimized DOM) agent-browser snapshot # Click element by accessibility ref agent-browser click @e5 # Type text agent-browser type @e3 "search query" # Get page info agent-browser info
Workflow
- •Navigate to the target URL
- •Screenshot to see current state
- •Snapshot to get accessibility tree with element refs
- •Interact using element refs (@e1, @e2, etc.)
- •Screenshot again to verify changes
Element References
The accessibility tree gives each element a unique ref:
code
@e1: button "Submit" @e2: textbox "Email" @e3: link "Home"
Use these refs for reliable element targeting.
Common Debugging Tasks
Visual Bug
bash
agent-browser navigate http://localhost:3000/broken-page agent-browser screenshot # Analyze screenshot for visual issues
Interactive Testing
bash
agent-browser navigate http://localhost:3000/form agent-browser snapshot agent-browser type @e2 "test@example.com" agent-browser click @e5 agent-browser screenshot
Layout Inspection
bash
agent-browser navigate http://localhost:3000 agent-browser snapshot # Analyze accessibility tree for structure
Prerequisites
Requires agent-browser CLI:
bash
npm i -g agent-browser@0.8.3
Output
Return findings:
- •Visual state: What the page looks like
- •Issues found: Layout problems, missing elements
- •Accessibility tree: Key element structure
- •Recommendations: How to fix issues