GitHub Integration: Surface Codex Cloud PR Reviews as GitHub Check Runs
What variant of Codex are you using?
App, CLI
What feature would you like to see?
Feature Request
Problem
When Codex Cloud automatically reviews a pull request through the GitHub integration, there is currently no visible GitHub Check or Status Check indicating that the review is in progress.
As a PR author, I cannot easily tell whether:
- Codex has started reviewing
- Codex is still reviewing
- Codex is queued
- Codex timed out
- Codex failed
- Codex completed successfully
The only reliable way to determine review status today is to wait for review comments to appear or navigate to the Codex Cloud review dashboard.
Proposed Solution
Create a GitHub Check Run for Codex Cloud reviews.
Suggested states:
- Pending: Review queued or running
- Success: Review completed
- Failure: Review failed
- Cancelled: Review aborted
- Timed Out: Review exceeded execution limits
Benefits
- Eliminates uncertainty about whether a review is still running.
- Allows developers to stay entirely within GitHub.
- Enables branch protection rules that require Codex review completion.
- Makes Codex Cloud reviews behave consistently with other asynchronous CI systems.
- Reduces the need to repeatedly check the Codex dashboard.
Additional Notes
This request is specifically for the built-in Codex Cloud GitHub review integration and not the Codex GitHub Action.
The GitHub Action already provides CI status because it runs inside GitHub Actions. The gap is that the managed Codex Cloud review workflow currently has no corresponding status indicator in GitHub.
Additional information
_No response_
2 Comments
I can confirm a related blocker for the managed Codex Cloud GitHub review integration.
Observed output is asymmetric:
For a fail-closed merge policy, the clean outcome therefore cannot be safely bound to the exact current PR head. A reaction, comment timestamp, timeline order, or abbreviated SHA should not be treated as merge authority.
Please make any proposed Check Run head-SHA-bound and structured enough to distinguish queued/running/completed/failed and clean/finding outcomes. A clean formal PullRequestReview with full commit_id would also solve this.
This is from a private-repository reproduction; no private repository or PR identifiers are included here.
This is a great enhancement and makes automated process of feedback easier. Now, the AI looks at the thread and assumes nothing is happening. I have instructions to wait up to 45 minutes, but the review does not always run and therefore the 45 minutes may be a waste of time. It would be great if we only have to wait for an action to complete.