Auto-generated PR for branch: fix/multi-act
<!-- This is an auto-generated description by cubic. -->
---
## Summary by cubic
Updated the system prompt to encourage combining related actions in one
step for better efficiency when multiple actions are allowed.
- **Prompt Improvements**
- Added clear guidelines and examples for efficient multi-action
combinations.
- Explained when to use single actions versus grouped actions.
<!-- End of auto-generated description by cubic. -->
Auto-generated PR for branch: hot-fix/url-not-existing
<!-- This is an auto-generated description by cubic. -->
---
## Summary by cubic
Improved logging in session.py to handle cases where the current page
URL may not exist during crashed page recovery, making debug and warning
messages clearer.
<!-- End of auto-generated description by cubic. -->
Updated the logging messages in the require_healthy_browser function to include the current page URL during recovery attempts. This change improves the clarity of debug and warning logs by ensuring the page URL is accurately reported, even when the page is unresponsive.
<!-- This is an auto-generated description by cubic. -->
## Summary by cubic
Added missing __init__.py files to make all subdirectories proper Python
packages and enable imports across the library.
- **Refactors**
- Ensured each module and LLM integration is recognized as a package.
- Exposed main chat classes in each LLM integration’s __init__.py for
easier imports.
<!-- End of auto-generated description by cubic. -->
Auto-generated PR for: fix temperature for reasoning models
<!-- This is an auto-generated description by cubic. -->
---
## Summary by cubic
Set temperature to 1 and frequency_penalty to 0 for reasoning models to
ensure correct parameter handling.
- **Bug Fixes**
- Removed unsupported parameter filtering for reasoning models.
- Updated example to use a reasoning model and show token usage.
<!-- End of auto-generated description by cubic. -->
Auto-generated PR for branch: feat/speed-improvements-wait
<!-- This is an auto-generated description by cubic. -->
---
## Summary by cubic
Improved speed and reliability by adding timeouts to LLM calls, browser
actions, and step execution, and refined the wait action to log accurate
wait times.
- **Performance and Reliability**
- Added configurable timeouts for LLM calls (default 60s) and step
execution (default 180s).
- Reduced timeouts for browser tab info and page title retrieval for
faster responses.
- Set explicit timeouts for browser actions like scroll, click, type,
and fill to prevent hangs.
- Updated the wait action to account for initial LLM wait time and
improved logging.
<!-- End of auto-generated description by cubic. -->
improve import times by setting up lazy imports
<!-- This is an auto-generated description by cubic. -->
---
## Summary by cubic
Added lazy import logic across core modules to speed up import times and
reduce startup overhead. Updated several LLM-related dependencies to
newer versions.
- **Refactors**
- Replaced direct imports with lazy imports in main package entrypoints
and heavy modules.
- Moved some imports inside functions to avoid loading large
dependencies at startup.
- **Dependencies**
- Bumped versions for google-genai, openai, anthropic, and groq.
<!-- End of auto-generated description by cubic. -->
Disables screenshot capture when vision is disabled to improve
performance
<!-- This is an auto-generated description by cubic. -->
---
## Summary by cubic
Disabled screenshot capture when vision is turned off to improve
performance.
<!-- End of auto-generated description by cubic. -->
- Added timeout parameters to scroll_into_view_if_needed, click, type, and fill methods to improve reliability and prevent potential hangs.
- Introduced debug observation for get_dom_element_by_index and _input_text_element_node methods to aid in tracking and debugging.
- Changed the exception type from asyncio.TimeoutError to TimeoutError for better clarity in timeout handling.
- Updated the error message to suggest generating fewer tokens in case of a timeout.
- Reduced the retry timeout from 6 seconds to 3 seconds.
- Decreased the page title retrieval timeout from 3 seconds to 2 seconds.
- Enhanced timeout handling logic to prioritize user-defined timeouts while ensuring a minimum of 3000ms.
- Updated the wait action to account for the initial 3 seconds already waited, ensuring accurate logging of total wait time.
- Simplified the ActionResult return by removing unnecessary parameters.
Auto-generated PR for branch: feature/optimize-browser-extensions
<!-- This is an auto-generated description by cubic. -->
---
## Summary by cubic
Updated the default browser extensions to use only fully automatic tools
for better single-session automation. Added Decentraleyes, ClearURLs,
and FastForward; removed extensions that require manual setup or
learning.
- **New Extensions**
- Added Decentraleyes for CDN tracking protection.
- Added ClearURLs to remove tracking parameters.
- Added FastForward to bypass link shorteners.
<!-- End of auto-generated description by cubic. -->
1. Remove Reader View from description - extension not actually loaded
- Updated enable_default_extensions description to only mention extensions that are actually loaded
- Removed misleading reference to Reader View functionality
2. Resolve Chrome flag conflict for extension activity logging
- Removed --disable-extension-activity-logging from CHROME_DEFAULT_ARGS
- Keeps --enable-extension-activity-logging in _get_extension_args() when extensions enabled
- Eliminates contradictory flags that caused unpredictable logging behavior
- Track actual loaded extension names instead of hardcoding them
- Log message now accurately reflects which extensions loaded successfully
- Count and names will match even if some extensions fail to load
- Fixes bug where log mentioned 'Reader View' which was removed from extension list
