Refactor v2 - j

[voxmenthe]

Refactor: Enhance modularity, maintainability, and async handling

This major refactoring separates the CLI LLM agent into distinct components:

1. LLMAgentCore (src/llm_agent_core.py):

   • Encapsulates core LLM interaction logic: Gemini client management, chat session handling, conversation history, token counting.
   • Manages tool registration and execution, supporting both synchronous and asynchronous tools (using AsyncTaskManager for async tools).
   • Handles GenerationConfig.

2. CodeAgent (src/main.py):

   • Acts as the CLI interaction layer using prompt-toolkit.
   • Delegates LLM operations to LLMAgentCore.
   • Delegates background task management to AsyncTaskManager.
   • Manages CLI-specific state.

3. AsyncTaskManager (src/async_task_manager.py):

   • Centralizes asyncio event loop management (runs in a separate thread).
   • Handles submission, tracking, and cancellation of background tasks (e.g., PDF processing, script execution).
   • Provides callbacks for task results.

4. Slash Commands (src/slashcommands.py):

   • Handlers updated to interact with LLMAgentCore for relevant backend operations and state modifications.

5. Tools (src/tools.py):

   • Browser-based tools (google_search, open_url) converted to async def.
   • Standardized google.generativeai imports.

Key Improvements & Fixes:

• Resolved critical ImportError for google.genai by standardizing import style.
• Fixed AttributeError by migrating from ThinkingConfig to GenerationConfig.
• Addressed ImportError for BrowserContext in agent_browser_utils.py.
• Improved separation of concerns for state management.
• Ensured clean configuration loading and component initialization.
• Validated structural integrity through iterative application startup checks.

Outstanding Considerations (for future work):

• Runtime testing of browser-based tools due to a potential underlying dependency conflict with browser-use.
• Full functional validation of all commands and LLM interactions in a live environment.
• Further investigation into mapping the /thinking_budget command's intent to GenerationConfig capabilities.