fix: allow conversational responses in Ask mode without forcing tool use #6582
Conversation
- Modified Task.ts to check if we're in Ask mode before forcing tool use - When no tool is used in Ask mode, the conversation ends gracefully instead of retrying - Added tests to verify Ask mode allows conversational responses - Added tests to ensure other modes still enforce tool use Fixes #6581
dosubot
bot
added
size:L
hannesrudolph
added
the
Issue/PR - Triage
![ellipsis-dev[bot]](https://avatars.githubusercontent.com/in/64358?s=60&v=4){kind=small}
| askTask.userMessageContentReady = true | ||
|
|
||
| // Spy on recursivelyMakeClineRequests to check if it returns true (ends loop) | ||
| const recursiveSpy = vi.spyOn(askTask, "recursivelyMakeClineRequests") |
There was a problem hiding this comment.
The spy on 'recursivelyMakeClineRequests' is created but not asserted anywhere. Consider either removing it or adding an assertion (e.g. verifying call count or arguments) to increase test clarity.
| const recursiveSpy = vi.spyOn(askTask, "recursivelyMakeClineRequests") |
| expect(askTask.consecutiveMistakeCount).toBe(0) | ||
| }) | ||
|
|
||
| it("should still enforce tool use in non-Ask modes", async () => { |
There was a problem hiding this comment.
In the test for non-Ask modes, the logic is manually duplicated (checking current mode and updating nextUserContent) rather than invoking the actual task loop. Consider refactoring the test to exercise the real behavior to reduce test fragility.
![roomote[bot]](https://avatars.githubusercontent.com/in/1546624?s=60&v=4){kind=small}
| nextUserContent = [{ type: "text", text: formatResponse.noToolsUsed() }] | ||
| this.consecutiveMistakeCount++ | ||
| // Check if we're in Ask mode before forcing tool use | ||
| const currentMode = await this.getTaskMode() |
There was a problem hiding this comment.
Consider adding a comment here explaining why Ask mode is special-cased. Something like:
| const currentMode = await this.getTaskMode() | |
| // Check if we're in Ask mode before forcing tool use | |
| // Ask mode is designed for conversational Q&A and doesn't require tool use for every response | |
| const currentMode = await this.getTaskMode() |
This would help future maintainers understand the design decision.
| if (currentMode === "ask") { | ||
| // In Ask mode, we don't force tool use for conversational responses | ||
| // This prevents the repetitive response issue | ||
| return true // End the loop successfully |
There was a problem hiding this comment.
The return value could benefit from a clarifying comment:
| return true // End the loop successfully | |
| // In Ask mode, we don't force tool use for conversational responses | |
| // This prevents the repetitive response issue | |
| return true // End the loop successfully - conversational response is valid in Ask mode |
Summary
This PR fixes issue #6581 where Ask mode was generating repetitive responses and making multiple API calls for simple conversational queries.
Problem
In Ask mode, when the agent provides a conversational response without using any tools, the system was flagging it as an error and forcing retries. This resulted in:
Solution
Modified the task loop logic in Task.ts to:
Changes
src/core/task/Task.ts: Added mode check before enforcing tool use in two locations:
src/core/task/tests/Task.spec.ts: Added comprehensive tests to verify:
Testing
Related Issues
Fixes #6581
Important
Fixes issue #6581 by allowing conversational responses in Ask mode without forcing tool use, with changes in
Task.tsand new tests inTask.spec.ts.initiateTaskLoopandrecursivelyMakeClineRequestsinTask.tsto check for Ask mode and prevent unnecessary retries.Task.spec.tsto verify Ask mode allows conversational responses without tool use.This description was created by
for f0813a5. You can customize this summary. It will automatically update as commits are pushed.