add Upgraded #Week 4 Sidekick project with Gradio 6.0 & LangGraph v0.2 compatibility #546
Conversation
There was a problem hiding this comment.
Pull request overview
This PR upgrades the Sidekick project to be compatible with Gradio 6.0 and LangGraph v0.2, implementing modern async/await patterns and improving resource management. The changes ensure the codebase runs smoothly with recent library updates while maintaining the core functionality of an AI agent that completes tasks with evaluation feedback.
Changes:
- Migrated to Gradio 6.0 with updated message formats and CSS handling in launch()
- Updated LangGraph nodes to accept RunnableConfig parameters and use ainvoke() for v0.2 compatibility
- Implemented native async/await patterns throughout for better event loop handling
Reviewed changes
Copilot reviewed 8 out of 9 changed files in this pull request and generated 14 comments.
Show a summary per file
| File | Description |
|---|---|
| sidekick_tools.py | Tool integrations with async Playwright initialization, web search, file management, Wikipedia, Python REPL, and push notifications |
| sidekick.py | Core LangGraph agent with async worker/evaluator nodes, RunnableConfig support, and resource cleanup |
| app.py | Gradio 6.0 UI with async event handlers and proper message formatting |
| pyproject.toml | Updated dependencies including Gradio 6.3.0+ and LangGraph 1.0.5+ |
| README.md | Documentation covering setup, features, and tech stack |
| .env.example | Environment variable template for Azure OpenAI, Serper, and Pushover APIs |
| .gitignore | Standard Python gitignore with Playwright additions |
| .python-version | Python 3.12 version specification |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| version = "0.1.0" | ||
| description = "Add your description here" | ||
| readme = "README.md" | ||
| requires-python = ">=3.12" |
There was a problem hiding this comment.
The pyproject.toml specifies 'requires-python = ">=3.12"' but the README Tech Stack section mentions 'Python 3.10+ asyncio'. This creates an inconsistency about the minimum Python version. Ensure all documentation and configuration files specify Python 3.12 as the minimum version.
| # Update or add system message | ||
| messages = state["messages"] | ||
| found_system_message = False | ||
|
|
||
| for message in messages: | ||
| if isinstance(message, SystemMessage): | ||
| message.content = system_message | ||
| found_system_message = True | ||
| break | ||
|
|
||
| if not found_system_message: | ||
| messages = [SystemMessage(content=system_message)] + messages | ||
|
|
||
| # Invoke the LLM with config for proper async context propagation | ||
| response = await self.worker_llm_with_tools.ainvoke(messages, config) |
There was a problem hiding this comment.
Mutating messages in the state directly can lead to unexpected behavior. Instead of modifying the existing SystemMessage object in place, consider creating a new messages list with the updated SystemMessage to ensure proper state management and avoid side effects.
| # Update or add system message | |
| messages = state["messages"] | |
| found_system_message = False | |
| for message in messages: | |
| if isinstance(message, SystemMessage): | |
| message.content = system_message | |
| found_system_message = True | |
| break | |
| if not found_system_message: | |
| messages = [SystemMessage(content=system_message)] + messages | |
| # Invoke the LLM with config for proper async context propagation | |
| response = await self.worker_llm_with_tools.ainvoke(messages, config) | |
| # Update or add system message without mutating existing state messages | |
| original_messages = state["messages"] | |
| found_system_message = False | |
| new_messages: List[Any] = [] | |
| for message in original_messages: | |
| if not found_system_message and isinstance(message, SystemMessage): | |
| # Replace the first SystemMessage with an updated one | |
| new_messages.append(SystemMessage(content=system_message)) | |
| found_system_message = True | |
| else: | |
| new_messages.append(message) | |
| if not found_system_message: | |
| # Prepend a new SystemMessage if none existed | |
| new_messages = [SystemMessage(content=system_message)] + new_messages | |
| # Invoke the LLM with config for proper async context propagation | |
| response = await self.worker_llm_with_tools.ainvoke(new_messages, config) |
| print("Cleaning up Sidekick resources...") | ||
| try: | ||
| if self.browser: | ||
| await self.browser.close() | ||
| if self.playwright: | ||
| await self.playwright.stop() | ||
| except Exception as e: | ||
| print(f"Exception during cleanup: {e}") No newline at end of file |
There was a problem hiding this comment.
The cleanup method uses a generic print statement instead of the configured logger. For consistency with the rest of the codebase which uses the logging module, consider using logger.info() and logger.error() instead of print statements.
| print("Setting up Sidekick...") | ||
| sidekick = Sidekick() | ||
| await sidekick.setup() | ||
| return sidekick | ||
|
|
||
|
|
||
| async def process_message( | ||
| message: str, | ||
| success_criteria: str, | ||
| history: Optional[list], | ||
| sidekick: Sidekick | ||
| ) -> tuple: | ||
| """ | ||
| Process user message and return updated chat history. | ||
| Returns: (updated_history, cleared_message_box) | ||
| """ | ||
| if not message or not success_criteria: | ||
| return history or [], "" | ||
|
|
||
| updated_history = await sidekick.run_superstep( | ||
| message, success_criteria, history or [] | ||
| ) | ||
| return updated_history, "" # Clear the message box after submission | ||
|
|
||
|
|
||
| async def reset_conversation(sidekick: Optional[Sidekick]): | ||
| """Reset the conversation and create a new Sidekick instance""" | ||
| if sidekick: | ||
| await sidekick.cleanup() | ||
| new_sidekick = await setup_sidekick() | ||
| return [], "", "", new_sidekick # chatbot, message, success_criteria, sidekick | ||
|
|
||
|
|
||
| async def free_resources(sidekick: Optional[Sidekick]) -> None: | ||
| """Cleanup callback when app closes""" | ||
| if sidekick: | ||
| try: | ||
| await sidekick.cleanup() | ||
| print("Sidekick resources freed") | ||
| except Exception as e: | ||
| print(f"Error during cleanup: {e}") |
There was a problem hiding this comment.
The print statements in the app.py file are inconsistent with the logging approach used in sidekick.py and sidekick_tools.py. Consider importing and using the logging module for consistent logging throughout the application.
| async def evaluator(self, state: State, config: RunnableConfig) -> Dict[str, Any]: | ||
| """ | ||
| Evaluator node: determines if success criteria met and if more input needed. | ||
| """ | ||
| last_response = state["messages"][-1].content | ||
|
|
||
| system_message = """You are an evaluator that assesses task completion by an Assistant. | ||
| Determine if the task meets the success criteria based on the assistant's final response. | ||
| Also assess whether the user needs to provide more input, clarification, or if the assistant is stuck.""" | ||
|
|
||
| user_message = f"""Evaluate this conversation: | ||
|
|
||
| {self.format_conversation(state["messages"])} | ||
|
|
||
| Task Success Criteria: | ||
| {state["success_criteria"]} | ||
|
|
||
| Assistant's Final Response: | ||
| {last_response} | ||
|
|
||
| Decide: | ||
| 1. Is the success criteria met? | ||
| 2. Does the user need to provide more input (clarification, stuck, etc.)? | ||
| 3. Provide constructive feedback. | ||
|
|
||
| Note: If the assistant says they wrote a file, assume they did. Give the benefit of the doubt but reject if more work is needed.""" | ||
|
|
||
| if state["feedback_on_work"]: | ||
| user_message += f""" | ||
|
|
||
| Previous Feedback Given: | ||
| {state["feedback_on_work"]} | ||
|
|
||
| If the assistant is repeating the same mistakes, mark that user input is required.""" | ||
|
|
||
| evaluator_messages = [ | ||
| SystemMessage(content=system_message), | ||
| HumanMessage(content=user_message), | ||
| ] | ||
|
|
||
| eval_result = await self.evaluator_llm_with_output.ainvoke(evaluator_messages, config) | ||
|
|
||
| return { | ||
| "feedback_on_work": eval_result.feedback, | ||
| "success_criteria_met": eval_result.success_criteria_met, | ||
| "user_input_needed": eval_result.user_input_needed, | ||
| } |
There was a problem hiding this comment.
The evaluator node is marked as async but the RunnableConfig parameter is not utilized in the function body. While it's correctly passed to ainvoke(), consider documenting why this parameter is required for LangGraph v0.2 compatibility to help future maintainers understand its purpose.
| logger.error(f"Error initializing file tools: {str(e)}") | ||
| return [] |
There was a problem hiding this comment.
The function returns an empty list if an exception occurs during FileManagementToolkit initialization, but this failure is silently ignored by the caller. This could lead to the agent running without file management capabilities without the user being aware. Consider propagating the exception or logging a warning that's more visible to users.
| logger.error(f"Error initializing file tools: {str(e)}") | |
| return [] | |
| # Log the error with full traceback and propagate so callers are aware | |
| logger.error(f"Error initializing file tools: {str(e)}", exc_info=True) | |
| raise |
| worker_response = ( | ||
| result["messages"][-2].content if len(result["messages"]) > 1 else "" | ||
| ) | ||
| evaluator_feedback = result["feedback_on_work"] or "No feedback" | ||
|
|
There was a problem hiding this comment.
The worker_response extraction logic assumes the second-to-last message contains the worker's response, but this may not always be accurate depending on the conversation flow. If there's only one message or if the message structure is different than expected, this could extract the wrong content or fail. Consider adding validation to ensure the message exists and is of the expected type.
| worker_response = ( | |
| result["messages"][-2].content if len(result["messages"]) > 1 else "" | |
| ) | |
| evaluator_feedback = result["feedback_on_work"] or "No feedback" | |
| messages = result.get("messages") or [] | |
| worker_response = "" | |
| if isinstance(messages, list) and messages: | |
| # Prefer the most recent AIMessage as the worker response | |
| for msg in reversed(messages): | |
| if isinstance(msg, AIMessage): | |
| worker_response = getattr(msg, "content", "") or "" | |
| break | |
| # Fallback: use the last message's content if no AIMessage was found | |
| if not worker_response: | |
| last_msg = messages[-1] | |
| if hasattr(last_msg, "content"): | |
| worker_response = last_msg.content or "" | |
| elif isinstance(last_msg, dict) and "content" in last_msg: | |
| worker_response = last_msg.get("content") or "" | |
| evaluator_feedback = result.get("feedback_on_work") or "No feedback" |
| def get_file_tools() -> List[Any]: | ||
| """Get file management tools (read, write, list files)""" | ||
| try: | ||
| toolkit = FileManagementToolkit(root_dir="sandbox") |
There was a problem hiding this comment.
The 'sandbox' directory used by FileManagementToolkit is hardcoded but not created if it doesn't exist. This could cause the file tools to fail on first run. Consider either creating the directory during initialization or documenting that users need to create it manually.
| toolkit = FileManagementToolkit(root_dir="sandbox") | |
| sandbox_dir = "sandbox" | |
| os.makedirs(sandbox_dir, exist_ok=True) | |
| toolkit = FileManagementToolkit(root_dir=sandbox_dir) |
| @@ -0,0 +1,129 @@ | |||
| import gradio as gr | |||
| from sidekick import Sidekick | |||
| import asyncio | |||
There was a problem hiding this comment.
Import of 'asyncio' is not used.
| import asyncio |
| from pydantic import BaseModel, Field | ||
| from sidekick_tools import playwright_tools, other_tools | ||
| import uuid | ||
| import asyncio |
There was a problem hiding this comment.
Import of 'asyncio' is not used.
| import asyncio |
|
Thanks so much. Would you be able to remove the uv.lock, and remove anything not needed, to reduce the amount of review work needed? As it stands, this is 3,500 lines and I won't have capacity to review.. thanks so much |
- Remove uv.lock (users generate their own) - Remove .python-version (local-only tool) - Simplify .gitignore to essential entries - Simplify .env.example for clarity - Update README with pip-first installation - Update pyproject.toml with proper project name
|
Hi! 👋 |
2 participants
Hi! I've really been enjoying the Sidekick project. While working through it, I noticed that recent updates to Gradio (v6.0) and LangGraph (v0.2) introduced some deprecation warnings and syntax changes.
I've updated the codebase to align with these newer standards to ensure it runs smoothly for future students.
Summary of Changes:
show_copy_button) and moved CSS tolaunch().RunnableConfig(required for v0.2+) and switched to.ainvoke.Tested and running successfully! Hope this is helpful.
User Interface