openclaw - 💡(How to fix) Fix Feature: user interrupt during streaming / tool execution [1 participants]

TL;DR

Implement a mechanism for the agent to check the message queue at tool-return breakpoints and process any waiting messages before continuing.

Guidance

• Identify the points in the code where tool calls complete and results are streamed back to the user, as these are the potential breakpoints for checking the message queue.
• Modify the agent's logic to poll the message queue at these breakpoints, allowing for early interruption or redirection based on user input.
• Consider implementing a flag or indicator to track whether a user message is waiting in the queue, enabling the agent to prioritize processing this message over continuing its current task.
• Review the message queue's implementation to ensure it can be polled efficiently and cheaply after each tool completes, as required for this feature.

Example

# Pseudo-code example of checking the message queue at a tool-return breakpoint
def tool_completed():
    # Check if there are any waiting user messages
    if message_queue.has_waiting_messages():
        # Process the waiting message
        process_user_message(message_queue.dequeue())
    else:
        # Continue with the current task
        continue_current_task()

Notes

The exact implementation details may vary depending on the specific architecture and technologies used in the project. Ensuring the message queue is designed for efficient polling is crucial for this feature.

Recommendation

Apply a workaround by modifying the agent's logic to check the message queue at tool-return breakpoints, as this approach aligns with the desired behavior and can be implemented without requiring significant changes to the underlying system.