Agent Skills: browser-automation

What I do

• Provide a safe, composable workflow for CDP-backed browsing tasks
• Use explicit browser_url values; do not assume a hidden browser singleton
• Use browser_snapshot UIDs for click and fill actions
• Confirm state changes after each action with browser_snapshot or browser_eval

Best-practice workflow

1. Inspect targets with browser_list({ browser_url })
2. Pick a target_id, or omit it to use the first page target
3. Navigate with browser_navigate if needed
4. Discover candidates using browser_snapshot
5. Click or fill using a UID from the latest snapshot
6. Confirm using browser_snapshot or browser_eval

CLI-first debugging

• List all available tools: npx @different-ai/opencode-browser tools
• Run one tool directly: npx @different-ai/opencode-browser tool browser_list --args '{"browser_url":"http://127.0.0.1:9222"}'
• Set a default endpoint: OPENCODE_BROWSER_URL=http://127.0.0.1:9222 npx @different-ai/opencode-browser status

This path is useful for reproducing CDP endpoint or snapshot issues quickly before running a full OpenCode session.

Available tools

• browser_list
• browser_navigate
• browser_snapshot
• browser_click
• browser_fill
• browser_eval
• browser_screenshot

Troubleshooting

• If browser_list fails, verify the browser was started with --remote-debugging-port
• If browser_click or browser_fill cannot find a UID, refresh the snapshot first
• Use browser_eval for page-specific checks that are not visible in the accessibility tree
• Confirm results after each action