Clarify scheduling limits for ChatGPT Scheduled Tasks vs Codex automations

Summary

The current documentation appears ambiguous about the minimum scheduling interval for tasks/automations.

The ChatGPT Scheduled Tasks documentation says that tasks cannot run more than once per hour.

At the same time, the Codex automation documentation discusses minute-based intervals, including use cases such as periodically checking a PR and addressing new review feedback.

There are also public Codex issues/examples involving automations configured at intervals such as 1, 2, or 5 minutes.

It is therefore unclear whether:

1. the one-hour minimum applies only to ChatGPT Scheduled Tasks;
2. Codex automations use a separate scheduler and may run every few minutes;
3. there are additional differences depending on whether an automation runs in ChatGPT, Codex app, Codex Cloud, or another environment.

Why this matters

A particularly important use case is the documented “babysit a pull request” workflow.

For PR monitoring, the difference between a 5-minute interval and a 1-hour interval is fundamental. A user deciding whether Codex can monitor review comments and CI failures needs to know the actual expected polling frequency before relying on this workflow.

It also affects how Codex compares with event-driven PR agents. For example, Claude's PR-following workflow appears to react to new GitHub events typically within seconds, with minutes being more of an upper bound. An hourly Codex poll would therefore be a very different workflow, while a minute-level Codex automation would be considerably closer.

Suggested documentation change

It would be helpful to explicitly document the distinction, for example:

ChatGPT Scheduled Tasks: minimum interval is 1 hour.
Codex automations: use a separate scheduling mechanism and support intervals down to X minutes.

If the limits vary by environment, plan, or automation type, a small table listing those limits would make this much clearer.

It would also be useful for the PR-babysitting example to state the expected minimum polling interval and whether it is polling-based or event-driven.

Start by comparing the current ChatGPT Scheduled Tasks and Codex automation documentation, especially the PR-babysitting example. Verify the minimum intervals and whether limits differ by environment, plan, or automation type. Done means the documentation clearly distinguishes the schedulers, states applicable limits, and explains whether PR monitoring is polling-based or event-driven.