Codex “All Projects” Permissions Reset and Require Repeated Manual Approvals

[donoscarcaba-boop]

Can we access your project?

• I give permission for members of the FlutterFlow team to access and test my project for the sole purpose of investigating this issue.

Current Behavior

REPRODUCTION STEPS:

1. Open FlutterFlow AI Agent > Settings > Permissions > Rules.
2. Add/modify the required command permission.
3. Select ALLOW.
4. Select ALL PROJECTS.
5. Save the rule.
6. Return to the AI Agent and allow Codex to continue working.
7. Codex later interrupts the task again with another “Allow command?” permission request.
8. Return to Settings > Permissions > Rules.
9. The permission scope/interface has reverted back to “THIS WORKSPACE” instead of remaining on “ALL PROJECTS.”

This behavior is repeatable.

The important issue is not simply that Codex asks for permission again. The ALL PROJECTS permission setting appears to reset/revert after returning to the AI Agent, causing previously authorized routine commands to require manual approval again.

This prevents true unattended/autonomous Agent operation because the user must remain at the computer to repeatedly approve commands.

Expected Behavior

Once I explicitly configure Codex permissions as “ALLOW > ALL PROJECTS,” I expect those permissions to persist and be honored across projects, workspaces, Agent sessions, and FlutterFlow restarts unless I manually change or revoke them.

Routine, non-destructive commands that I have already authorized should run automatically without repeatedly stopping the AI Agent to ask for the same permissions again.

The purpose of using the AI Agent is to allow it to work autonomously. I should be able to start a repair, diagnostic, validation, or other approved task and leave the computer while Codex continues working.

I should NOT have to sit behind the computer for hours babysitting the AI Agent because it stops every few minutes waiting for another “Allow command?” approval.

If I have already explicitly granted the required permissions for ALL PROJECTS, I expect the Agent to continue working unattended and only interrupt me when there is a genuinely destructive, security-sensitive, or otherwise exceptional action that requires my approval.

Steps to Reproduce

1. Open a FlutterFlow project in the Windows Desktop app and open AI Agent.

2. Go to AI Agent Settings > Permission Rules.

3. Set the required PowerShell command permission to “Allow” and “All projects.”

4. Click Save and return to AI Agent.

5. Allow the AI Agent to continue working normally.

6. The Agent later stops again and asks for permission to run a PowerShell command that should already be authorized.

7. Return to Permission Rules and observe that the permission has reverted/reset from “All projects” to “This workspace.”

8. Change it back to “All projects,” save it again, and return to AI Agent.

9. The problem repeats. Previously granted permissions do not persist reliably, causing the Agent to repeatedly stop for manual approval.

10. This prevents unattended operation and requires me to remain at the computer babysitting the AI Agent, including during long-running or overnight work.

Reproducible from Blank

• The steps to reproduce above start from a blank project.

Bug Report Code (Required)

IT4Kj8nf049Om+EA0a6Jc8BsoCcRNjJ7RpEalu1EGUg0IbWvG7YiS8+kMxJVXemndGNEL1r88X40097PvtnxG+0rHxuCQ4NewbgBaTqVWliXVLLUCqqwY3d6QJ5VG2ag4Z+KmA9QH7dvLyEe6jqtfdiQSgjfQ++yC2sRMvqkGKXAjmyrKwDsGSdVxmNecCfn

Visual documentation

1. Open a FlutterFlow project in the Windows Desktop app and open AI Agent.

2. Open AI Agent Settings > Permission Rules.

3. Add/modify the required PowerShell command permission and select:
   Allow > All projects.

4. Click Save.

5. Return to AI Agent and allow it to continue working on the project.

6. During normal Agent work, another PowerShell command permission request appears even though the command was previously authorized.

7. Return to Permission Rules and observe that the permission is no longer behaving as an All Projects persistent permission and/or the interface has reverted to This workspace.

8. Set it back to Allow > All projects and save again.

9. Continue AI Agent work. The same permission interruption occurs again.

10. Because previously approved permissions are not persisting reliably, the AI Agent repeatedly stops and waits for manual approval, preventing unattended operation and requiring the user to remain at the computer and babysit the Agent.

Environment

- FlutterFlow version: 
- Platform:
- Browser name and version:
- Operating system and version affected:

Additional Information

This issue has a major impact on my ability to use the AI Agent as intended. I cannot leave the computer unattended while the Agent works because it repeatedly stops and waits for permission approvals that I have already granted.

I specifically want to be able to leave the AI Agent working on long-running projects overnight while I sleep, or allow it to continue working while I am away from the computer. Instead, I currently have to stay beside the laptop and constantly check whether another permission request has interrupted the work.

This effectively forces me to babysit the computer for hours. I should not have to go to bed with my laptop beside me just so I can keep approving routine commands throughout the night.

For users relying on the AI Agent for substantial projects, this is not a feasible workflow. Once I explicitly select “Allow > All Projects” for routine, non-destructive commands, that authorization should persist. The Agent should be able to continue working autonomously and only interrupt me when an action genuinely requires new or security-sensitive approval.