Codex 0.147.0 adds a flag for unattended approvals

Codex 0.147.0 replaces piping yes at a prompt with --approve-for-me, an auto-reviewed approval flag for unattended runs, per the changelog.

Anyone scripting Codex in CI or cron no longer has to fake consent by piping yes into a prompt built for humans. That workaround skipped every check the approval step exists for.

The new flag keeps some form of review in the loop instead of turning it off outright.

The same release tightened plugin isolation and network access enforcement, making both harder to break out of or slip past. It also raised the trust bar for local projects Codex hasn't seen before. None of those are optional the way approval gating is, and they apply whether or not you touch --approve-for-me.

Where the flag actually sits

--approve-for-me isn't the old model of a human clicking approve on every step, and it isn't approvals switched off. It's a middle setting. The changelog doesn't say what "automatically reviewed" actually checks, or what makes a review pass versus fail.

OpenAI didn't build this flag speculatively. Piping yes into Codex was already how people ran it in production, unsupported and never admitted. --approve-for-me is OpenAI catching up to what its own users were already doing.

Watch whether OpenAI documents what "automatically reviewed" evaluates. That detail decides whether this is a real safety boundary or a compliance-shaped wrapper around the same yes-pipe everyone was already running.