Windows app hangs on splash when saved auth references a deactivated workspace instead of prompting re-login

Confirm. Had the same issue (that's how I found this).
Workaround also solved the issue.

Confirmed. The workaround doesn't work for me apparently.

It worked actually. but yeah, this gotta be fixed

Can confirm that the workaround helped for me!

Can confirm that the workaround helped for me!

This worked for me as well!

same problem and same resolution

yes it has solved my problem

Deleting the auth.json did not solve my problem. But deleting the %USERPROFILE%\.codex\ and reinstalling worked.

I hit the same Windows splash deadlock. I cannot create a deactivated workspace for a clean repro, but the public app-
server code seems consistent with the reported state loop.

The startup/auth check appears to trust cached auth first:

• AppServerSession::read_account() sends account/read with refresh_token: false
• get_login_status() treats Some(Account::Chatgpt { .. }) as authenticated
• should_show_login_screen() only shows login for LoginStatus::NotAuthenticated
• backend get_account_response() builds account state from cached auth via provider.account_state()

Relevant lines:

• codex-rs/tui/src/app_server_session.rs:302
• codex-rs/tui/src/lib.rs:1686
• codex-rs/tui/src/lib.rs:1749
• codex-rs/app-server/src/codex_message_processor.rs:1884
• codex-rs/model-provider/src/provider.rs:147

Then app/list performs the real workspace/app backend calls. A non-success ChatGPT backend response is converted
into a generic error string, and the app/list path returns that as an internal JSON-RPC error rather than converting
invalid workspace/auth into logout or NotAuthenticated.

Relevant lines:

• codex-rs/app-server/src/codex_message_processor.rs:5843
• codex-rs/app-server/src/codex_message_processor.rs:5938
• codex-rs/app-server/src/codex_message_processor.rs:6012
• codex-rs/chatgpt/src/connectors.rs:100
• codex-rs/chatgpt/src/chatgpt_client.rs:63

I also do not see any explicit handling for deactivated_workspace in the repo. The only clear auth-clearing path is
explicit logout:

• codex-rs/tui/src/app_server_session.rs:728
• codex-rs/app-server/src/codex_message_processor.rs:1756
• codex-rs/login/src/auth/manager.rs:1730

So I think this is best described as an invalid-auth/workspace state transition bug, not a true module circular
dependency (Correction to my earlier framing). Cached auth is enough to pass startup/login gating, but later workspace-backed startup calls fail. That failure does not clear auth or expose a recovery action, so the desktop shell can remain stuck in splash/loading until auth.json is removed or codex logout is run.

Suggested fix direction, not a full patch: when a startup request receives a permanent workspace/auth rejection like
deactivated_workspace, classify it as “saved auth is no longer usable”, clear or invalidate the cached auth state,
emit/update account state as unauthenticated, and route to login/workspace selection. A manual “logout/reset auth”
action on the splash screen would also be a good fallback, but the primary fix should probably be to break the stale-
auth state immediately rather than retrying it.

A retry/circuit-breaker limit would still be useful as a defensive guard against future splash loops, but for this
specific error retrying seems unnecessary because the workspace rejection is permanent until the user chooses a
different workspace/session/log-in method.

flowchart TD
    A[App starts] --> B[account/read]
    B --> C{Cached auth exists?}
    C -->|No| D[Show login/onboarding]
    C -->|Yes| E[Treat as authenticated]

    E --> F[Main UI / app boot continues]
    F --> G[app/list]
    G --> H[Load connector directory + codex_apps MCP]
    H --> I[ChatGPT backend rejects token/workspace]
    I --> J[402 deactivated_workspace returned as generic app/list error]

    J --> K[Desktop shell keeps splash/loading state]
    K --> L{Was auth cleared?}
    L -->|No| B
    L -->|Manual logout / codex logout| D

    style J fill:#ffd6d6,stroke:#c00
    style L fill:#ffd6d6,stroke:#c00

I can confirm I’m seeing this same failure mode.

In my case, the workspace was expired/deactivated. ChatGPT appears to have fixed this flow now by letting the user switch to/join a personal workspace instead of getting stuck, but Codex still opens to a blank/unusable screen with no in-app way to log out or switch workspace.

The workaround was to manually log out through the local Codex files, then reopen Codex. After that, the app became usable again.

The main requested fix is the same as this issue: when Codex detects deactivated_workspace / stale workspace auth state, it should show a recovery UI with “switch workspace” or “log out / sign in again” instead of rendering a blank screen.

Holy crap, this is exactly the same issue I ran into! I fixed the same problem using your method.