Codex Desktop (Linux): in-app Browser control unavailable to tasks despite enabled setting

What variant of Codex are you using?

Codex Desktop on Linux.

What version of the Codex App are you using?

Not captured from the About dialog. The bundled in-app Browser plugin visible in the affected session was version 26.814.41957.

What platform is your computer?

Linux desktop. The user also has Brave installed with the ChatGPT extension, but the primary reproduction below uses Codex's built-in in-app Browser, not the external-browser extension path.

What issue are you seeing?

Codex Desktop's Settings → Browser → “Let ChatGPT control the built-in browser” toggle is enabled, but a task that explicitly requests the in-app Browser cannot attach to or control it.

The task was asked to inspect the currently open Facebook page, navigate/scroll it, and capture screenshots. The in-app Browser skill/plugin was available, but the agent-side browser-control execution interface required to initialize the browser was not exposed to the task. Consequently, the agent could not list tabs, inspect the page, navigate, click, scroll, or take screenshots.

The failure persisted after starting a fresh task.

This is not explained by the Browser preference being disabled: a screenshot taken during troubleshooting shows the toggle enabled.

What steps can reproduce the bug?

1. In Codex Desktop, open Settings → Browser.
2. Enable Let ChatGPT control the built-in browser.
3. Open a website in the built-in browser (an authenticated Facebook page was used in this report).
4. Start a task with an explicit in-app Browser request, for example:

``
Inspect the currently open Facebook page and capture screenshots.
``

1. Ask the task to use the in-app Browser plugin/skill.
2. Observe that the task cannot initialize browser control because its browser-control execution interface is unavailable, even though the Browser setting is enabled.
3. Start a fresh task and repeat; the same failure occurs.

What is the expected behavior?

When the built-in Browser control toggle is enabled and a task explicitly requests the in-app Browser, Codex should provide the task with the browser-control capability so it can attach to the currently open browser tab and perform permitted UI actions such as inspecting, navigating, scrolling, and taking screenshots.

If browser control cannot be attached, Codex should surface a user-visible diagnostic that identifies the missing capability and offers a recovery path, rather than leaving the task unable to distinguish a platform issue from a settings problem.

What is the actual behavior?

The Browser setting is enabled, but the in-app Browser control capability is absent from the task. The agent cannot make the initial browser connection and therefore cannot perform any browser actions.

Additional information

• The user confirmed the Browser toggle was enabled for the entire troubleshooting sequence.
• The user has a ChatGPT browser extension installed in Brave, but this report concerns the enabled built-in Browser path.
• The issue occurred in more than one task, including a newly created task.
• The failure was encountered while trying to inspect an already-open Facebook page; no attempt was made to read cookies, passwords, local storage, or other sensitive browser data.
• The closest related public report found was #24040, but it concerns a missing Windows native-messaging registration for the Chrome extension, not this Linux built-in Browser attachment failure.

A screenshot of the enabled Browser setting is available and can be attached to this issue.