All guides

Original explainer

Schedule local project checks in Codex: an official-source guide

Test a focused prompt, choose an in-chat or standalone schedule, and verify project access, machine requirements and the first result.

AI Nav CityWritten Sources checked

Recurring checks need a clear scope and a reviewable completion standard. This original guide follows the official Scheduled tasks documentation checked on November 2, 2026. No paid model task, background wakeup, runtime or code-check outcome was tested.

Confirm three conditions first

  1. Scope

    Specify the project, checks and reporting conditions. In-chat schedules preserve context; standalone tasks use the saved prompt.[1]

  2. Availability

    Local-file tasks require the computer to stay on, the desktop app to keep running and the project to remain on disk.[1]

  3. Evidence

    Test manually and review the first few run records. A created schedule does not prove execution or completion.[1]

The documentation may use ChatGPT desktop app or Codex naming by version. Follow the Scheduled entry and workspace permissions visible in your current client.

1. Test the prompt manually

Start in a regular Codex chat with the project and scope. For example: “Check recent changes in this project, run its existing checks and report failures with relevant files.” Read the actual result before reusing the prompt daily.[1]

  • Specify a project or source the task can access.
  • State separately whether it may inspect, edit, commit or publish.
  • Save durable rules in the prompt or project documentation rather than relying on temporary context.

[1]

2. Choose cadence and execution location

In a desktop app supporting scheduled tasks, request creation from a ChatGPT or Codex chat. For example: “Return to this chat at 09:00 local time daily to check this project; report actionable findings.” Verify the name, timezone, time, prompt and active status. Entry locations may vary by version.[1]

Git projects can run directly in the local project or in an isolated worktree. Direct execution uses your working directory; worktree execution needs a clear route for reviewed results to return to the mainline. Projects without version control run in their project directory. The CLI and IDE extension lack the Scheduled management interface: test the prompt there, then manage it in a supported desktop or web interface. Web tasks cannot directly operate a folder on your computer.[1]

3. Inspect the first result

  • Review recent runs in Scheduled and confirm the actual time and specified project.
  • Keep the computer and desktop app running for local-file tasks, and ensure the project has not moved.
  • Review early outputs for scope coverage; narrow an overly broad prompt.
  • Archive completed worktree tasks when they are no longer needed.

[1]

The verifiable minimum is a real run record for the specified project with findings traceable to files or commands. Otherwise mark execution unverified rather than treating a planned time as evidence.[1]

Verification scope

Checked: official guidance on creation, in-chat and standalone schedules, local projects, worktrees, machine requirements, manual testing and run management. Unchecked: your account eligibility, organizational policies, actual background wakeup, model cost and code-repair outcomes. Example prompts are original; follow the capabilities of your current account.[1]

Article sources

Original editorial content; official descriptions and checking suggestions are distinguished.

  1. Codex · Official sources
    View details