Previously, when browser launch succeeded after retry using a temporary
directory, the code immediately deleted ALL temporary directories including
the one actively being used by the running browser process. This caused
browser crashes and data corruption when Chrome tried to write to the
deleted user_data_dir during normal operation.
The fix:
- Only clean up unused temporary directories after successful launch
- Preserve the currently used temporary directory until browser cleanup
- Update _temp_dirs_to_cleanup to track only the in-use directory for
proper cleanup during BrowserKillEvent
Content blocks are dictionaries at runtime despite type annotations
indicating Pydantic models. Added type: ignore comments to allow
dictionary-style .get() access which works correctly at runtime.
Previously, cache_control was applied to every content block and tool call
in a message, which could exceed Anthropic's limit of 4 cache breakpoints
per request. This fix ensures cache_control is only applied to the last
block in each message type:
- Last text block in UserMessage/AssistantMessage content
- Last tool_use block when AssistantMessage has tool calls
- Only the last overall block when AssistantMessage has both content and tools
This prevents exceeding the cache breakpoint limit while maintaining
correct caching behavior.
- Fixes validation error when LLM returns {'scroll': {'pages': 3}} without 'down' field
- Defaults to True (scroll down) when 'down' is missing, which is the most common case
- Still supports explicit down=False for scrolling up
- Also fixes CDP client compatibility with older cdp-use versions by making max_ws_frame_size optional
Fixes validation errors with AWS Bedrock Claude and Gemini models
Cursor Agent