- changed the system messages to have <sys> tag instead of <s> to avoid confusion with HTML
- Got rid of MessageMetadata, SupportedMessageTypes and implemented cleaner MessageManagerState
- Implemented a HistoryItem class to cleanly reconstruct agent history description
<!-- This is an auto-generated description by cubic. -->
## Summary by cubic
Added token caching support for Claude models, including accurate
tracking and cost calculation for cached, cache creation, and uncached
tokens.
- **New Features**
- Added `cache` flag to messages and updated Anthropic message
serialization to support Claude's token caching.
- Updated token usage and cost tracking to handle cached tokens, cache
creation tokens, and their costs.
- Improved token usage logs to show detailed breakdowns for cached and
uncached tokens.
- Added tests for Anthropic cache serialization and cost calculation.
<!-- End of auto-generated description by cubic. -->
- Flatten AgentOutput to get rid of current_state
- Add agent initialization in history.
- Update the example tool call.
- Add a current_state property to AgentOutput for compatibility.
- Enhanced logging in the Agent class to inform users about available attachments after task completion.
- Revised system prompt guidelines to clarify the use of text and files_to_display in the done action.
- Applied automated linting suggestions from pre-commit hooks, primarily updating type hints (e.g., Dict -> dict) and organizing imports.
- Refactored Playwright script execution in examples/features/playwright_script_generation.py to use asyncio.create_subprocess_exec for asynchronous handling and real-time log streaming, resolving ASYNC221 lint error and improving output visibility.
Remove comments automatically added by the LLM in agent-related files.
Removed comments from:
- browser_use/agent/playwright_script_generator.py
- browser_use/agent/service.py
- browser_use/agent/views.py
These comments were redundant and cluttered the code.
Implement the ability to generate executable Playwright scripts from an Agent's run history (`AgentHistoryList`).
This introduces the `PlaywrightScriptGenerator` class which translates recorded agent actions (clicks, inputs, navigation, etc.) into a Python script using Playwright's async API.
Key features:
- Uses XPath selectors derived from the agent's interaction history.
- Includes a robust helper function (`_try_locate_and_act`) with XPath prefix trimming for improved resilience on dynamic pages.
- Handles sensitive data placeholders by mapping them to environment variables in the generated script.
- Supports configuring the browser launch and context based on the original agent's `BrowserConfig` and `BrowserContextConfig`.
- Activated by setting the `save_playwright_script_path` option in `AgentSettings`.
Includes modifications to `Agent` service to trigger generation and `AgentHistoryList` view to add the saving method. Also adds a new example (`examples/features/playwright_script_generation.py`).
This feature enables reliable replay of agent runs, automated testing, and scripting of repetitive tasks discovered by the agent.
mertunsall