Version 0.42.0 - new required approvals breaks workflows

Resolved 💬 4 comments Opened Sep 27, 2025 by olliecrow Closed Jan 10, 2026

💡 Likely answer: A maintainer (etraut-openai, contributor) responded on this thread — see the highlighted reply below.

Safer command execution: dangerous command checks in core and Windows-specific safety (https://github.com/openai/codex/pull/4211, https://github.com/openai/codex/pull/4119).

From: https://github.com/openai/codex/releases/tag/rust-v0.42.0

Version 0.42.0 enforces approvals for some commands (eg rm -f, git reset, etc).

These approvals seem to be enforced always, irrespective of whether various bypass mechanisms are enabled or not (eg --dangerously-bypass-approvals-and-sandbox, approval_policy="never", etc).

I couldn't seem to find a way to permit all commands with the new 0.42.0.

This breaks several of our automated workflows.

Is there a way to bypass the requirement for these approvals?

For now, I'm pinning 0.41.0.

View original on GitHub ↗

4 Comments

moxi000 · 11 months ago

same

zbynekdrlik · 10 months ago

Same here, strong reason to switch back to claude. Main positive of codex was that was working without interruptions. Now unusable as in codex previously. Simple rm on highly safe vm with dangerously-bypass-approvals-and-sandbox and it not work without interruption any more

dimfeld · 10 months ago

If you're comfortable compiling your own, it's pretty easy to modify this behavior. For example in https://github.com/openai/codex/commit/44624bd0fbf8bba068e0a9136498c86378235c35 I changed it to only flag rm commands where one of the arguments starts with /, and if you really want you could replace the whole function body with false to disable it completely.

etraut-openai contributor · 7 months ago

Closing old bug.