Codex App: 'Approve for me' no longer displays Auto-review rationale
What version of the Codex App are you using (From "About Codex" dialog)?
26.707.8479.0 (bundled Codex CLI 0.144.2)
What subscription do you have?
ChatGPT Pro
What platform is your computer?
Microsoft Windows NT 10.0.26200.0 x64
What issue are you seeing?
When I use Approve for me (Auto-review), the automatic-review item no longer shows the reviewer's textual decision rationale. I see the result/status, such as Approved or Denied, but not the explanation of why the request was allowed or blocked.
This affects both approved and denied requests. Earlier desktop versions displayed the reviewer text, and I have not intentionally changed any approval or Auto-review policy configuration.
This makes it difficult to audit automated approval decisions, especially when an unexpected request is approved or a legitimate request is denied.
What steps can reproduce the bug?
- In the Codex App, enable Auto-review under Settings > General > Permissions.
- Start a task using Approve for me.
- Ask Codex to perform an action that requires crossing the active sandbox boundary, so an automatic approval review is triggered.
- Wait for the review to complete as Approved or Denied.
- Inspect the automatic-review item in the conversation.
The item shows the review status, but the reviewer rationale text is absent.
Example task/session containing automatic-review activity: 019f5f3f-770e-7c31-a732-60deed5bc796.
What is the expected behavior?
For both approved and denied requests, the automatic-review item should display the reviewer's rationale, along with the risk level and user-authorization assessment when those fields are available. This is important for understanding and auditing why an action was permitted or blocked.
If the rationale is unavailable for a particular review, the UI should say that explicitly rather than silently showing only the status.
Additional information
The current app-server protocol still defines Auto-review completion data with status, optional rationale, riskLevel, and userAuthorization. It marks this payload as unstable, but the data path still supports a rationale.
I searched the public issue tracker and did not find an exact duplicate. The closest related report is #21975, which requests that denials/aborts be surfaced with the Auto-review rationale and risk metadata. This report is specifically about a regression in the desktop UI where the textual rationale that previously appeared is no longer displayed for either approval outcome.