Description: When asking the AI to perform a multi-step task (e.g., translating multiple comments in a file via PowerShell), the IDE shows a separate confirmation dialog for every single command, even though they are all part of the same user-requested action.
Steps to reproduce:
- Load a Python file with multiple Spanish comments
- Ask the AI: “Translate all Spanish comments to English”
- Observe that the IDE shows a “Confirm the command is safe to run” dialog for each individual replace command
Expected behavior: One single approval should authorize all commands that are part of the same task execution. The user should not need to click “Yes, allow this time” repeatedly.
Actual behavior: A confirmation dialog appears for every single sub-command. The “Yes, allow this time” button sometimes scrolls out of view or disappears when the command output is large, making it impossible to click without scrolling.
Impact: High — This makes multi-step tasks extremely tedious and breaks the workflow. It also creates a false sense of security since users end up clicking “allow” blindly due to fatigue.
Suggested fix: Add a “Yes, allow for this entire task” option that grants approval for the duration of the current agent task, similar to how browser permissions work (“Allow always for this site”).