Interrupted video transcription leaves generated transcript file locked by Python on Windows

What version of Codex is running?

Unknown from UI; observed on May 5, 2026 in the Codex desktop app on Windows.

Which model were you using?

GPT-5 based Codex coding agent in the desktop app.

What platform is your computer?

Windows

What happened?

After asking Codex to transcribe a local video from the project docs folder, I interrupted the transcription while it was running. Codex had created a transcript file named 2026-05-05 21-39-46_PARTE_1.transcript.txt inside the project docs directory.

When I tried to delete that generated file from Windows Explorer, Windows showed that the file was still open in Python and could not be removed until the Python process was terminated.

The app had already uploaded feedback and showed this feedback/thread ID:

019df9f1-7d90-7950-8e06-b329535708db

Steps to reproduce

1. In Codex desktop on Windows, ask Codex to transcribe a local video file.
2. Let Codex start a Python-based transcription workflow that writes a transcript file into the working directory.
3. Interrupt the task before transcription completes.
4. Try to delete the generated transcript file from Windows Explorer.
5. Observe that Windows reports the file is still in use by Python.

Expected behavior

Interrupting the task should cleanly release file handles and stop background Python processes so generated files can be deleted immediately.

Actual behavior

The generated transcript file remains locked by Python after interruption, and Windows blocks deletion until the Python process is killed.

Additional context

The generated file involved here was:

docs/2026-05-05 21-39-46_PARTE_1.transcript.txt

A related temporary audio extraction file was also created during troubleshooting, but the locking issue was specifically visible on the transcript file.

I initially tried to inspect the current in-app browser tab for the feedback modal, but direct browser automation in this environment was blocked by an outdated local Node runtime. The feedback ID above comes from the UI screenshot/link and should help correlate the uploaded report.