fix: handle Ask mode responses without tool usage for Google Gemini grounding

[roomote]

Summary

This PR fixes an infinite loop issue that occurs when using Google Gemini with grounding/search enabled in Ask mode.

Problem

When using Google Gemini with URL context and grounding/search enabled in Ask mode, the system expects a tool use but the model provides direct answers. This causes Roo Code to repeatedly send the noToolsUsed() error, creating an infinite loop that eventually errors out with "This may indicate a failure in the model's thought process or inability to use a tool properly".

Solution

The fix adds logic to detect when the task is in Ask mode and skips the tool validation for responses without tool usage. This is appropriate because Ask mode is specifically designed to provide direct answers without requiring tool usage.

Changes

• Modified Task.ts to check if the current mode is "ask" before triggering the noToolsUsed() error
• Added the same check in both locations where this validation occurs:
  • In recursivelyMakeClineRequests method (line ~1789)
  • In initiateTaskLoop method (line ~1349)

Testing

• All existing tests pass
• The fix allows Ask mode to work properly with Google Gemini's grounding feature without entering an infinite loop

Fixes #6671

---

Important

Fixes infinite loop in Ask mode with Google Gemini by skipping tool validation for direct answers in Task.ts.

• Behavior:
  • Fixes infinite loop in Ask mode with Google Gemini by skipping tool validation for responses without tool usage.
  • Applies to recursivelyMakeClineRequests and initiateTaskLoop methods in Task.ts.
• Testing:
  • All existing tests pass.
  • Ask mode now works correctly with Google Gemini's grounding feature without infinite loops.

^{This description was created by for 2e41089. You can customize this summary. It will automatically update as commits are pushed.}

[ellipsis-dev]

Good fix: the code now checks task mode before pushing the no-tools-used error. In Ask mode, the loop breaks gracefully. Consider abstracting the check (e.g. a helper function isAskMode()) to reduce duplication.

[roomote]

I wrote this code 5 minutes ago and already forgot why. Classic me.

[roomote]

Missing test coverage for this new behavior. This is a critical change that prevents infinite loops in Ask mode - we should have tests to ensure it works correctly and prevent regressions.

Consider adding tests that verify:

• Ask mode allows responses without tool usage
• Other modes still trigger the noToolsUsed() error
• The consecutiveMistakeCount is not incremented in Ask mode

[roomote]

This logic is duplicated from lines ~1349-1360. Consider extracting into a helper method to follow DRY principles:

Suggested change

	const isAskMode = currentMode === "ask"
	private async shouldAllowResponseWithoutToolUsage(): Promise<boolean> {
	const currentMode = await this.getTaskMode()
	return currentMode === "ask"
	}

Then both locations could use:

if (!await this.shouldAllowResponseWithoutToolUsage()) {
    // Handle the error case
}

[roomote]

Have we considered if other modes might benefit from allowing responses without tool usage? The Orchestrator mode, for example, has an empty groups array and might face similar issues with certain LLM providers.

[woodworker]

This is a good idea, maybe not check explicit for Ask mode but maybe have a setting for Models with Grounding Mode?

[daniel-lxs]

Closing, see #6671 (comment)