fix: prevent subtask query from being returned instead of result #7252
Conversation
![roomote[bot]](https://avatars.githubusercontent.com/in/1546624?s=60&v=4){kind=small}
#7251) - Modified newTaskTool to not include the original subtask message in pushToolResult - The actual subtask result will be provided when the subtask completes via finishSubTask - Updated test to match the new expected message format This fixes the issue where GPT-5 and other models were receiving the subtask query instead of the actual subtask completion result when orchestrating tasks.
dosubot
bot
added
size:S
![ellipsis-dev[bot]](https://avatars.githubusercontent.com/in/64358?s=60&v=4){kind=small}
| // Don't include the original message in the tool result to avoid confusion | ||
| // The actual result will be provided when the subtask completes | ||
| pushToolResult( | ||
| `Successfully created new task in ${targetMode.name} mode. Waiting for subtask to complete...`, |
There was a problem hiding this comment.
The pushToolResult message now returns a generic acknowledgment (excluding the original query), which fixes the issue. Consider using the t() function for localization to keep user-facing text consistent.
| `Successfully created new task in ${targetMode.name} mode. Waiting for subtask to complete...`, | |
| t('tools:newTask.success', { mode: targetMode.name }), |
This comment was generated because it violated a code review rule: irule_C0ez7Rji6ANcGkkX.
| // Don't include the original message in the tool result to avoid confusion | ||
| // The actual result will be provided when the subtask completes | ||
| pushToolResult( | ||
| `Successfully created new task in ${targetMode.name} mode. Waiting for subtask to complete...`, |
There was a problem hiding this comment.
Minor: The success message has redundant wording - it says "Code Mode mode" (the word "mode" appears twice). Could we simplify this to just Successfully created new task in ${targetMode.name}. Waiting for subtask to complete...?
| pushToolResult(`Successfully created new task in ${targetMode.name} mode with message: ${unescapedMessage}`) | ||
| // Don't include the original message in the tool result to avoid confusion | ||
| // The actual result will be provided when the subtask completes | ||
| pushToolResult( |
There was a problem hiding this comment.
The fix correctly addresses the issue by removing the original message from the tool result. This ensures the actual subtask result (provided via the finishSubTask mechanism) is what gets passed back to the parent task. Good approach! 👍
| @@ -99,7 +99,9 @@ describe("newTaskTool", () => { | |||
| expect(mockCline.emit).toHaveBeenCalledWith("taskSpawned", expect.any(String)) // Assuming initCline returns a mock task ID | |||
| expect(mockCline.isPaused).toBe(true) | |||
| expect(mockCline.emit).toHaveBeenCalledWith("taskPaused") | |||
| expect(mockPushToolResult).toHaveBeenCalledWith(expect.stringContaining("Successfully created new task")) | |||
| expect(mockPushToolResult).toHaveBeenCalledWith( | |||
| "Successfully created new task in Code Mode mode. Waiting for subtask to complete...", | |||
There was a problem hiding this comment.
Good job updating the test to match the new expected behavior. Consider adding an integration test in the future to verify the end-to-end flow of subtask results being properly passed back to parent tasks.
hannesrudolph
added
the
Issue/PR - Triage
|
This does not solve the issue. The core issue is actually returning the subtask result. |
|
Closing, this doesn't seem right |
Summary
This PR fixes issue #7251 where the orchestrator mode (GPT-5) was receiving the original subtask query instead of the actual subtask completion result.
Problem
When using the orchestrator mode to delegate subtasks:
Solution
Modified the
newTaskToolto not include the original subtask message in thepushToolResultcall. Instead, it now returns a simple acknowledgment that the subtask was created and is waiting for completion. The actual result is properly provided when the subtask completes via thefinishSubTaskmechanism.Changes
Testing
Related Issues
Fixes #7251
Important
Fixes issue #7251 by modifying
newTaskToolto return acknowledgment instead of the original subtask query, ensuring the actual result is provided upon subtask completion.newTaskToolto return acknowledgment instead of the original subtask query.finishSubTask.newTaskTool.ts: ChangespushToolResultto exclude the original subtask message.newTaskTool.spec.ts: Updates test to verify new message format.This description was created by
for 5fb4441. You can customize this summary. It will automatically update as commits are pushed.