codex - 💡(How to fix) Fix Codex App Automations on Windows ARM64: “Run now” creates empty read-only session and never submits prompt [1 comments, 2 participants]

What version of the Codex App are you using (From “About Codex” dialog)?

Codex App for Windows, CLI version shown in local sessions: 0.126.0-alpha.8. Codex CLI also installed: 0.125.0. I can provide the exact “About Codex” app version after checking the dialog.

What subscription do you have?

ChatGPT Plus

What platform is your computer?

Windows on ARM64 Project path: C:\Users\ajago\Documents\Codex\Teaching PowerShell shell

What issue are you seeing?

Key Evidence

Automation config is valid and persisted: C:\Users\ajago.codex\automations\teste\automation.toml Run session is created: rollout-2026-04-28T20-52-37-019dd5a6-94ec-7143-a058-8be50a370e0b.jsonl The created run has: tokens_used = 0 first_user_message = "" model = null reasoning_effort = null sandbox_policy = {"type":"read-only"} The session contains no submitted user task, no model response, no tool calls, and no terminal/file activity. Same task succeeds in normal Codex App threads and via CLI/local execution. Most Precise Conclusion The “Run now” button appears to create the automation run/session record, but the automation prompt is never dispatched into that session. The UI then keeps showing the run as in progress because there is no agent turn to complete or fail.

This is likely an app-side automation dispatch bug, not a workspace permission, prompt, Python, or local filesystem issue.

What steps can reproduce the bug?

1. Open Codex App on Windows ARM64.
2. Create a local automation in project C:\Users\ajago\Documents\Codex\Teaching.
3. Set execution environment to Local.
4. Use a simple prompt that writes one diagnostic file into the project folder.
5. Save the automation.
6. Click “Run now”.
7. Observe that the UI creates a run under “Previous runs”, but it stays spinning indefinitely.
8. Inspect local Codex state/session files.
9. The run session jsonl exists, but contains metadata only.
10. The state row shows tokens_used = 0, first_user_message = "", model = null, reasoning_effort = null, sandbox_policy = read-only.
11. No model response, tool call, terminal command, or error is recorded.
12. The expected output file is not created.

What is the expected behavior?

Clicking “Run now” should submit the automation prompt to the agent, start a normal Codex run using the configured model/reasoning/workspace, execute the task, and mark the run as completed or failed with a visible error.

For this diagnostic automation, the expected result is that Codex creates the requested file in: C:\Users\ajago\Documents\Codex\Teaching\radars\GenAI_Updates\output\automation_relative_write_test.txt

If the automation cannot run, the UI should show a clear failure reason instead of leaving the run spinning indefinitely.

Additional information

No response