Codex App update breaks GitHub/network access in existing working threads

Open 💬 1 comment Opened Jun 25, 2026 by tomlucero

What version of the Codex App are you using (From “About Codex” dialog)?

Version 26.623.30605

What subscription do you have?

Plus

What platform is your computer?

Darwin 25.5.0 arm64 arm

What issue are you seeing?

After updating the Codex App on June 25, 2026, an existing thread that had previously been able to use GitHub lost network access. Earlier in the same thread, GitHub auth worked and I was able to push a branch successfully. After the update, the thread began showing CODEX_SANDBOX_NETWORK_DISABLED=1, could no longer resolve github.com, and GitHub operations failed.

This appears to be a regression caused by the Codex App update or by how updates affect already-running thread sessions. The local repo and branch were fine, and the same work could still be pushed from a different thread. The issue is that the existing thread/runtime lost GitHub/network capability after the update.

What steps can reproduce the bug?

Feedback ID: 019effeb-7e31-71a0-a11d-59f9b16fea95

What is the expected behavior?

Open a Codex App thread that currently has working GitHub/network access.

Use GitHub successfully in that thread. In my case, I was able to authenticate and push a branch before the update.

Update the Codex App while continuing to use the same thread.

Return to that same thread and try GitHub/network operations again.

Observed results:
CODEX_SANDBOX_NETWORK_DISABLED=1 appears in the thread environment.
gh auth status fails or reports invalid auth in that thread.
git fetch, git push, or direct access to github.com fail.
github.com cannot be resolved from that same thread after the update.

This happened more than once on June 25, 2026. The same thread worked before the update and broke after the update.

Updating the Codex App should not break GitHub/network access for a thread that was already working. If a thread had active GitHub/network capability before the update, it should keep that capability after the update, or Codex should clearly recover/re-request the needed permissions instead of silently leaving the thread in a broken network-disabled state.

Additional information

This had a real cost because I spent a significant amount of my limited token budget diagnosing a Codex runtime/session issue instead of doing actual work. The failure was especially disruptive because the affected thread contained the project context and prior work history, so I could not simply continue normally in the same place.

Important detail: this was not a repo or GitHub configuration problem on my side. The local branch and repo were fine, and I was able to push the same work from a different thread when that thread had working network access.

View original on GitHub ↗

This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗