[Feature Request] Automatically Retry and Resume Tasks on API Stream Failure

[oldjs]

Describe

When using kilocode to execute a multi-step task, the process is often interrupted by an "API stream transmission failed" error, followed by a "terminated" status. This appears to be caused by high load on the underlying AI model's API (seemingly the Claude API).

The main issue is that after this failure, the task does not automatically attempt to recover. I have to manually intervene and re-prompt the assistant to continue its work. This is very inconvenient, especially for complex or long-running tasks.

Steps to Reproduce

1. Assign a complex, multi-step task to the AI assistant (e.g., "Integrate Redis caching into the project, including environment configuration, writing utility code, updating the service layer, and creating documentation").
2. The assistant breaks the task down into a Todo List and begins executing the steps sequentially.
3. During one of the steps (e.g., while creating the documentation file dev-docs/redis_cache_guide.md as shown in the attached screenshot), the underlying API call fails.
4. The UI displays a red error message "API流式传输失败" (API stream transmission failed) and "terminated," and the entire workflow stops.

Expected Behavior

When a temporary API request or stream failure occurs, kilocode should ideally:

• Auto-Retry: Automatically attempt to re-execute the failed step, perhaps with a 1-3 retry-on-failure logic with a short delay.
• Seamless Resume: If a retry is successful, the task should resume from the exact point of interruption without any need for user intervention.
• Status Notification: To improve the user experience, the UI could display a transient message like, "Connection lost. Attempting to reconnect..." during the retry process.

Actual Behavior

The task halts completely upon API failure, requiring the user to manually restart or continue the process. This adds friction and disrupts the workflow.

Screenshot

Implementing an automatic recovery mechanism would significantly enhance the stability and user experience of the product. Thank you for considering this suggestion.