Foreword

Thank you for taking the time to fill this feature request fully. Without it we may not be able to address it, and the issue may be closed without resolution.

Problem Statement

When running tasks with claude-code, the agent frequently pauses execution to ask: "Do you want to proceed? (y/N)".

This is very frustrating because it breaks the "parallel" and automated nature of the app. I often start a task and step away (e.g., to get a coffee), but when I return, I find the agent has been stuck for 10 minutes waiting for me to manually type "y". This issue is even more frequent when using a proxy or when the agent needs to perform multiple file operations.

Possible Solution

I would like the ability to run agents in a non-interactive/unattended mode. Specifically, for Claude, I would like to be able to pass the --dangerously-skip-permissions flag.

Possible implementations:

1. Global Toggle: A setting to "Always skip permissions (dangerous)" in the app settings.
2. Custom CLI Flags: A field in the "New Task" dialog that allows me to append custom arguments to the agent command.

Describe alternatives you've considered

• Manual Monitoring: Staying at the desk just to press "Enter," which defeats the purpose of an automated coding assistant.
• Piping "yes": I tried to think of ways to pipe "yes" to the terminal, but that is unreliable and often breaks the terminal UI (xterm.js) within the app.

Additional context

The claude CLI introduced the --dangerously-skip-permissions flag specifically to solve this "stuck on prompt" issue for automation. Since Parallel Code is all about managing multiple agents efficiently, having a way to bypass these interactive gates would significantly improve the user experience for long-running background tasks.

Thank you so much for this amazing project! It’s incredibly useful and has quickly become a core part of my workflow—I absolutely love it!