Feature Request: Streamlined Agent Permission Management with an Interactive TUI

[Iweisc]

Problem: Currently, managing AI model permissions requires direct modification of the opencode.json file. This process is cumbersome, error-prone, and inefficient for developers who need to make frequent adjustments. The current workflow interrupts the development process, requiring developers to switch contexts and manually edit a configuration file, which can lead to syntax errors and invalid configurations.

Proposed Solution: To address these challenges, I propose the development of a Terminal-based User Interface (TUI) for on-the-fly management of AI model permissions. This interactive pop up would provide developers with a streamlined and intuitive way to view, add, and revoke permissions without ever leaving the command line.

Developer Benefits:

I believe this will improve efficiency, ux, visibility, and reduced error as the users wont have to worry about syntax.

[github-actions[bot]]

This issue might be a duplicate of existing issues. Please check:

• #1961: Unifying tools and permissions - discusses the need to unify permissions and tools management in agent configuration
• #3205: Agent should request permission before reading or editing files - addresses permission management for file operations
• #2521: Feature Request: Add Prefix Requirements for Permission Prompts - proposes enhanced permission prompt interface with configuration options
• #1813: Feature Request: Yolo and non-Yolo mode (auto-accept) - requests ability to toggle between ask/allow permission modes easily

Additionally, these related issues discuss permission and configuration management:

• #2208: Support {file:...} references for agent permissions - for external permission files
• #2264: Feature request: Persist permission requests - for permission persistence between sessions

Feel free to ignore if none of these address your specific case.

[rekram1-node]

Absolutely, we have discussed this internally as well. Totally in agreement

Not 100% sure how we will surface it, because it may be like claude code where you can just toggle on and off all permissions at once, but prolly makes sense to have a more advanced configuration option too

[Iweisc]

% sure how we will surface it, because it may be like claude code where you can ju

I have something like when you press ctrl + x and then press , and that opens the settings menu where u can add new models, change the permissions and other stuff can be added later. shouldn't be that hard because the backend for all of those are already implemented im pretty sure.

[igordertigor]

My feeling is that the "accept always" choice when agents ask for permission is already going in this direction. It's just that you can't "undo" the accept always choice unless you start a new session (which would lose context). I was therefore thinking that a list of the current permissions could be in the new (?) side panel that also shows todo and modified files and so on. That would already be a benefit, because it would help keep an overview of the permissions. Maybe then there could be an interface to revoke those again.

For the config-file permissions, I feel that they are fine. I often keep my config file open and adjust the permissions while I work with open code. For the current session that adjustment can usually be done using accept always, but not necessarily for later sessions or if I want to refine things.