uses CDP Runtime.evaluate with includeCommandLineAPI to access getEventListeners() and mark elements with click/mousedown/pointerdown listeners before DOM snapshot capture
This adds support for connecting to an existing browser instance via
Chrome DevTools Protocol (CDP) URL in the skill_cli commands.
## Why
The current skill_cli only supports three browser modes:
- chromium: Launches a new Playwright-managed Chromium
- real: Launches users Chrome with profile
- remote: Uses Browser-Use Cloud
However, many use cases require connecting to an already-running browser:
- Remote browser automation via SSH tunnels
- Connecting to browsers in Docker containers
- Reusing authenticated browser sessions
- Integration with Claude Code on remote servers
The BrowserSession class already supports cdp_url, but it was not
exposed through the skill_cli. The TUI mode has this option, but the
simple CLI commands did not.
## Changes
- main.py: Added --cdp-url argument and pass-through to server
- server.py: Accept and store cdp_url, pass to session registry
- sessions.py: Handle cdp_url in session creation (takes precedence)
## Usage
browser-use --cdp-url http://localhost:9222 open https://example.com
browser-use --cdp-url http://localhost:9222 state
browser-use --cdp-url http://localhost:9222 click 5
## Testing
Tested with Chrome remote debugging over SSH tunnel - successfully
controlled a local macOS Chrome browser from a remote Linux server.
Vibe coded with Claude Opus 4.5
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The CLI 'run' command now supports BROWSER_USE_API_KEY as an LLM option,
using ChatBrowserUse which is optimized for browser automation tasks.
This aligns the CLI behavior with the documented supported models at
https://docs.browser-use.com/supported-models which shows ChatBrowserUse
as a valid LLM option using BROWSER_USE_API_KEY.
Changes:
- Add ChatBrowserUse as first LLM option in get_llm() when BROWSER_USE_API_KEY is set
- Update error message to mention BROWSER_USE_API_KEY
- State command outputs same format that browser-use agents see
- Added aliases: bu, browser (all work the same)
- Headed mode now correctly shows browser window
- Change main entry point to 'browser-use'
- Move TUI to 'browser-use-tui'
- Add README with human instructions
- Update SKILL.md references
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
CDP's dispatchKeyEvent doesn't trigger beforeinput events that Draft.js
relies on for text input. This causes the first character to be dropped.
Solution: Detect Draft.js editors via specific DOM attributes (DraftEditor
class, data-contents, data-block, data-offset-key) and prepend a sacrifice
space character that Draft.js will drop, preserving the actual text.
- Added 'row', 'cell', 'gridcell' to interactive_roles in ClickableElementDetector
- Added 'row', 'cell', 'gridcell' to interactive_ax_roles in ClickableElementDetector
- Verified logic with isolated test case
When use_vision is set to False, screenshots should not be sent to the
judge LLM. This is a privacy concern as websites may contain sensitive
data that users explicitly do not want shared with LLMs.
Changes:
- Add use_vision parameter to construct_judge_messages()
- Only encode and include screenshots when use_vision is not False
- Pass self.settings.use_vision from _judge_trace() to construct_judge_messages()
- Add comprehensive tests for all use_vision modes
Fixes#3726
Laith Weinberger