Roo forgets context when VS Code unexpectedly crashes

[cornelhattingh]

App Version

3.26.6

API Provider

xAI (Grok)

Model Used

xai/grok-code-fast-1

Roo Code Task Links (Optional)

Whenever the AI model starts timing out, or run into other errrors I have to close VS Code in order for it to behave again. This leads to the current chat context being lost. Previously completed chats remain available.

🔁 Steps to Reproduce

Ask roo to perform a step by step task.
Wait for the AI to misbehave, either being unable to edit files, API errors or even MCP Tool errors.
Close VS Code and re-open it again.
The last active chat does not appear in the chat history and there is no way to recover the context.

💥 Outcome Summary

Expected to be able to continue where Roo left off, found chat history and context of the last chat missing

📄 Relevant Logs or Errors (Optional)

I don't have any output since context, terminal history and chat history has been reset

[roomote]

I'm exploring a potential fix for the chat context persistence issue. I'll investigate how to ensure chat history is properly saved and restored when VS Code restarts unexpectedly.

[roomote]

I've submitted a pull request to fix this issue: #7660

The solution persists the current active task ID to VS Code's global state and automatically restores it when the extension restarts. This ensures that users don't lose their chat context when VS Code crashes or needs to be restarted.

The implementation includes:

• Tracking the active task ID in global state
• Persisting changes whenever tasks are added/removed from the stack
• Restoring the last active task during extension initialization
• Comprehensive tests to ensure the functionality works correctly
• Graceful handling of edge cases (missing/corrupted tasks)

All existing tests pass and I've added 9 new tests specifically for the task persistence functionality.

[daniel-lxs]

Thanks for reporting this. Closing VS Code unexpectedly can indeed result in the current session’s context being lost. That’s expected behavior and not something the extension can reliably prevent.

Also, restarting VS Code is not required to recover from API errors or other model issues - simply starting a new task is usually enough to reset the state.

I’ll close this as there isn’t an actionable bug here. If you run into a scenario where the model consistently fails and does not recover even after starting a new task, please share the exact steps to reproduce and we can revisit.