BUG: --full-auto does not set 'on-failure' APPROVAL_POLICY as per --help
Resolved 💬 3 comments Opened Sep 15, 2025 by ben-vargas Closed Nov 3, 2025
What version of Codex is running?
0.36.0
Which model were you using?
gpt-5-codex
What platform is your computer?
Linux 6.8.0-79-generic x86_64 x86_64
What steps can reproduce the bug?
Start codex with codex --full-auto and once TUI loads run /status and note the Approval Mode:
• Approval Mode: on-request
• Sandbox: workspace-write
Running codex --help the --full-auto is supposed to be setting APPROVAL_POLICY to on-failure:
--full-auto
Convenience alias for low-friction sandboxed automatic execution (-a on-failure, --sandbox workspace-write)
What is the expected behavior?
As documented in codex --help the --full-auto flag should run codex as follows -
• Approval Mode: on-failure
• Sandbox: workspace-write
What do you see instead?
Approval mode runs in on-request instead of on-failure in --full-auto.
Additional information
codex-rs/tui/src/lib.rs:92 should use AskForApproval::OnFailure rather than AskForApproval::OnRequest
3 Comments
Maybe this was actually an intended change but
--helpwasn't properly updated? It appears #3608 changed this but didn't fix documentation/help and wasn't communicated in release notes?I've been working around this issue when using codex by adding the following to my
~/.codex/config.tomland not usingcodex --full-autoanymore. Since an OAI employee changed it on purpose in PR to 0.36.0, but probably forgot to change help output and documentation - it may be working as OAI intends.But, just putting this default in the
config.tomlor setting up equivalent flags as an alias, is a reasonable workaround if they do intend to change--full-autoto useon-requestapprovals going forward.closed the PR sitting untouched for a month and a half, i guess someone else can fix it if its an issue anyone cares about