Regression in codex-cli 0.116.0: agent cannot work normally in Debian 12, downgrading to 0.115.0 restores functionality

What version of Codex CLI is running?

codex-cli 0.116.0

What subscription do you have?

business

Which model were you using?

gpt-5.4

What platform is your computer?

Debian 12 (WSL)

What terminal emulator and version are you using (if applicable)?

Windows Terminal WSL

What issue are you seeing?

Summary

codex-cli 0.116.0 appears to regress normal repo-based usage for me on Debian 12. On the same machine, in the same repository, downgrading to 0.115.0 immediately restored normal behavior.

Environment

• OS: Debian 12
• Installation method: npm global install
• Working version: codex-cli 0.115.0
• Regressed version: codex-cli 0.116.0
• Likely relevant environment detail: bwrap / sandboxing may be involved, but I have not isolated root cause

Version proof

Before downgrade:

• codex --version returned codex-cli 0.116.0

Downgrade command:

• npm install -g @openai/codex@0.115.0

After downgrade:

• codex --version returned codex-cli 0.115.0

Observed behavior

With 0.116.0, Codex was not working normally in a real repo workflow. The practical outcome was that the agent became unusable for the task at hand in this workspace.

After downgrading to 0.115.0, the same environment immediately became usable again. For example, Codex could again access the repository, inspect local state, and respond normally.

Expected behavior

0.116.0 should work at least as reliably as 0.115.0 in the same Debian 12 environment and repo workflow, with no regression in normal agent execution.

Actual behavior

0.116.0 behaved as a regression severe enough that I had to downgrade to continue working. 0.115.0 resolved the issue immediately without any other environment changes.

Reproduction

1. Install 0.116.0:

npm install -g @openai/codex@0.116.0

1. Start Codex in an existing repository workspace on Debian 12.

1. Attempt a normal coding / agent workflow.

1. Observe that 0.116.0 does not function normally in that workflow.

1. Downgrade:

npm install -g @openai/codex@0.115.0

1. Retry the same workflow in the same repo and environment.

1. Observe that normal behavior is restored.

Notes

• I suspect Debian 12 sandboxing interaction may be relevant.
• bwrap seems like a plausible area to inspect - Debian 12 has v0.8.0 and at a guess from the erros I suspect 0.116.0 now assumed >=0.9.0 - but I haven't proved this absolutely.
• I am intentionally not claiming confirmed root cause; I am only reporting the version-to-version regression.
• I do not currently have a clean minimal stderr snippet to attach, but the regression was clear enough and reproducible enough in practice that downgrading was necessary.

What steps can reproduce the bug?

Try anything that results in a file write - codex is unable to write, claims it's blocked by sandbox malfunction.

What is the expected behavior?

Continue to work as per previous version - codex is able to update files in repo.

Additional information

_No response_