Permissions issue: clicking "Always Allow" causes ChatGPT to reconnect and reauthorize every time

[jplew]

Whenever elevated permissions are required by ChatGPT, it prompts you to "Deny", "Allow once", or "Always allow" commands.

Clicking "Allow once" works fine, it executes the command and proceeds.

However, clicking "Always allow" instantly triggers a connect & authorize loop again, even though I already authorized once:

It pops up three modals and asks for the owner password all over again, each time I try to permanently approve a command.

This interrupts the workflow. Is there a workaround for this?

[jplew]

ok never mind, this seems to have done the trick:

chatgpt is going full clip now
just want to say I love this project ❤️

[aaamosh]

Thanks for building this. DevSpace is a genuinely interesting project; we are testing it now as a ChatGPT Developer Mode -> local coding/Codex-like bridge on Linux.

I tested the Always allow / permission-bypass path today on:

• @waishnav/devspace@1.0.1
• ChatGPT web, personal account, Developer Mode app
• OAuth app URL: https://samsnote.tail55b3e8.ts.net/mcp through Tailscale Funnel
• Browser route: Server2-routed/SplitVPN isolated Chrome

Result in my current setup:

1. I set global Settings -> Apps -> App Preferences -> Ask permission to Always ask to force a prompt.
2. In a new chat I selected the DevSpace app and asked ChatGPT to open /home/amosh/Priv/Projects.
3. ChatGPT showed the expected prompt with Deny / Allow once / Always allow.
4. Clicking Always allow opened the second elevated-risk confirmation: “Let ChatGPT use DevSpace ... without confirmation?”
5. Confirming it changed the app-specific setting to Settings -> Apps -> DevSpace app -> Preferences -> Ask permission -> Never (ELEVATED RISK).
6. I then restored the global preference to Important changes, left the DevSpace app-specific override at Never, opened a fresh chat, and verified open_workspace + bash ran without permission prompts. The DevSpace service logs showed successful open_workspace and bash tool calls.

So the workaround from your comment is confirmed here: the practical “bypass permissions” mode is the ChatGPT app-specific setting Preferences -> Ask permission -> Never, not a DevSpace environment flag.

One caveat: in my current setup I could not reproduce the reconnect/owner-password loop after the second Always allow confirmation. Before this test I had locally patched our installed DevSpace so dynamically registered OAuth clients persist across service restarts, because otherwise ChatGPT can reuse cached client_id values that DevSpace no longer remembers. If the loop appears when ChatGPT switches into the permanent-approval flow and reconnects/reuses OAuth state, OAuth client persistence may be related.

OpenAI docs for this behavior say personal accounts can override permissions per app in Settings > Apps, and eligible prompts can offer Always allow which remembers permission for the app: https://developers.openai.com/apps-sdk/deploy/connect-chatgpt#try-the-app