Problem
The agent terminal appears to be non-interactive, which limits its functionality when executing commands that require user input. According to my research, this is a confirmed behavior of agent terminals.
Issue
When running commands that need interactive input (e.g., password prompts, confirmation dialogs, or any stdin-based interaction), the agent cannot complete the request because:
- The terminal cannot receive or pass through user input
- Commands that pause waiting for input will timeout or fail
- There’s no way to provide input to a running command within the agent context
Example Scenarios That Fail
- SSH commands requiring password entry
- Package managers prompting for confirmation (e.g.,
apt installwithout-yflag) - Interactive scripts or CLI tools
- Commands requiring
reador similar input mechanisms
Question
Is there a roadmap or workaround for handling interactive terminal commands in Cursor’s agent? Should we be documenting this limitation so users don’t encounter unexpected failures?
What’s the recommended approach for automating tasks that would normally require user interaction?