browser-automation

Files (5)

• EXAMPLES.md
• REFERENCE.md
• SKILL.md
• _meta.json
• setup.json

SKILL.md (first lines)

# Browser Automation

Automate browser interactions using the `agent-browser` CLI on Zo Computer.

## Available Tool

The `agent-browser` CLI is installed at `/usr/local/bin/agent-browser`. It auto-detects Chrome or falls back to Edge.

## Commands

```bash
agent-browser open <url>                    # Navigate to URL
agent-browser snapshot -i -c                # Accessibility tree with @refs (primary way to read page)
agent-browser snapshot -c                   # Full tree with text content
agent-browser screenshot                    # Viewport screenshot, image returned inline
agent-browser click <@ref>                  # Click element by ref
agent-browser fill <@ref> <text>            # Clear and type into field
agent-browser type <text>                   # Type at cursor
agent-browser press <key>                   # Enter, Tab, Escape, etc.
agent-browser select <@ref> <value>         # Select dropdown value
agent-browser scroll down/up [px]           # Scroll
agent-browser get text <@ref>               # Get text of element
agent-browser eval <js>                     # Execute JavaScript
agent-browser wait --load networkidle       # Wait for page load
agent-browser tab                           # List tabs
agent-browser tab new [url]                 # New tab
```

## Workflow

1. `agent-browser open <url>` — Navigate to page
2. `agent-browser snapshot -i -c` — Read page content with element refs
3. Interact using `@ref` values from snapshot
4. Re-snapshot after changes to verify results

## Login Flow

If the user mentions logging in or signing in:

```bash
agent-browser close
agent-browser --headed open <url>
```

This shows the browser window so the user can enter credentials. Wait for confirmation before continuing.

## Best Practices

1. **Always navigate first** before interacting
2. **Use snapshot to read page content** — don't guess element refs
3. **Use screenshot when user wants to see the page visually**
4. **Be specific** in action descriptions
5. **Close browser** when done: `agent-browser close`
6. **Stop and ask the user** after 2-3 consecutive failures

## Error Recovery

If navigation fails due to anti-bot protection, try alternative approach or ask user.
If element not found after page load, scroll down once then retry.
If buttons can't be clicked due to overlays, use keyboard: `agent-browser press Tab` then `agent-browser press Enter`.

## Zo Browser Tools (Alternative)

Zo also has built-in browser tools that work through the app:

- `open_webpage(url)` — Navigate to URL
- `view_webpage()` — Get current page content as markdown + screenshot
- `use_webpage(task)` — Interact with the current page

Use `agent-browser` for efficiency on unauthenticated sites. Use Zo browser tools for authenticated sessions and when sites block the CLI browser.

From SillyHippy/zo-skills, MIT.

More from SillyHippy/zo-skills