Skip to content

browser-agent-v0.4.1

browser-agent-v0.4.1: Release v0.4.1 - Activity Poller Bug Fixes & Password Manager Support
@guyernest guyernest tagged this 18 Nov 04:43 
- Fixed script executor selection ordering bug causing wrong wrapper to be used
  * Environment variables now set BEFORE finding script executor
  * Correctly routes to computer_agent_wrapper.py when browser_engine=computer_agent
  * Previously always used nova_act_wrapper.py due to timing issue

- Fixed headless mode hardcoded to true in Activity Poller
  * Now respects config.headless setting for server tasks
  * User preference properly honored

- Added real-time Python output streaming to Activity Poller
  * Changed from buffered .output().await to .spawn() with async streaming
  * Catches early Python import errors before logging setup
  * Provides real-time execution feedback

- Password Manager Autofill Support (Three-Layer Defense)
  * Layer 1: Disabled automation detection banner (--disable-blink-features=AutomationControlled)
  * Layer 2: New 'press' step type for keyboard actions (supports ArrowDown + Enter)
  * Layer 3: Programmatic password manager trigger when autofill fails
  * Resolves Windows password autofill issues

- Fixed AppSync deployment error in listTemplatesByAgent function
  * Corrected duplicate placeholder name in DynamoDB filter expression
- Added config save warning about restart requirement
- Removed exposed API keys from repository (legacy/to_delete.json)

- lambda/tools/local-browser-agent/src-tauri/src/script_executor.rs
  * Reordered environment setup to fix timing bug (lines 51-62)
  * Added real-time stdout/stderr streaming (lines 179-230)
  * Enhanced logging for debugging

- lambda/tools/local-browser-agent/src-tauri/src/activity_poller.rs
  * Fixed hardcoded headless=true to use config.headless (line 233)

- lambda/tools/local-browser-agent/python/openai_playwright_executor.py
  * Added browser args to disable automation detection (lines 303-308)
  * Implemented press step handler (lines 757-783)

- ui_amplify/amplify/data/listTemplatesByAgent.js
  * Fixed filter expression placeholder conflict

- ✅ macOS (dev mode) - All features working
- ✅ Activity Poller working with computer_agent engine
- ✅ Browser opens correctly (non-headless mode)
- ✅ Real-time Python output visible in logs
- ⏳ Windows (production) - Ready for testing password manager autofill
- ⏳ Linux (production) - Ready for remote testing

If upgrading from v0.3.1:
1. Stop Activity Polling if running
2. Install new version
3. Verify config.yaml settings (especially browser_engine and headless)
4. Restart application
5. Start Activity Polling
6. Password autofill improvements work automatically (no config changes needed)

🤖 Generated with Claude Code
Assets 2
Loading