Codex code review silently fails on initial trigger (no 👀) but succeeds on retry minutes later

What issue are you seeing?

On the attempt to trigger a code review using @codex after a period of inactivity, the chatgpt-codex-connector[bot] completely ignores the request. There is no 👀 reaction (the synchronous webhook ACK) and no review posted. The event silently drops.

However, if I wait 2 to 5 minutes and post the exact same @codex comment again, the bot instantly adds the 👀 reaction and successfully generates the review. The missing 👀 on the first attempt confirms the failure is happening at the webhook ingress layer before the job ever reaches the review generation queue.

What steps can reproduce the bug?

Leave the Codex GitHub App idle on a repository for an extended period (allowing backend compute/containers to spin down).

Go to a Pull Request and comment @codex review.

Observe: The bot adds no 👀 reaction and generates no review.

Wait 2 to 5 minutes.

Post the exact same @codex review comment on the same PR.

Observe: The bot immediately adds the 👀 reaction and posts a successful review shortly after.

What is the expected behavior?

The bot must acknowledge the GitHub webhook with the 👀 reaction and process the review on the first trigger, regardless of how long the app has been idle.

Additional information

Ruled out: Repository permissions, token scopes, GitHub App installation state, or PR diff size/complexity. If any of these were the root cause, the secondary retry a few minutes later would also fail.

Suspected cause: Serverless cold start or API Gateway timeout at the Codex webhook receiver layer. GitHub enforces a strict 10-second timeout for webhook deliveries. Because the bot fails to post the synchronous ACK (👀) on the first try, it strongly indicates the Codex ingress container is taking longer than 10 seconds to spin up from a cold state. GitHub permanently severs the connection and drops the event. By the second attempt, your compute instance is warm, the payload is ingested within 10 seconds, and the review succeeds.

Requested check: Please investigate your webhook ingress telemetry for 504 Gateway Timeout or 502 Bad Gateway errors resulting from GitHub's 10-second delivery limit. Your webhook receiver warming/scaling policies likely need adjustment to prevent these transient cold-start drops.