›_ Codex Issues
Codex Issues / #38137

[Windows][Scheduled Tasks] ACTIVE scheduled task skipped its 08:40 run

Open 💬 2 comments Opened Aug 12, 2026 by joon42066667-maker

Related issue: #38130 (/feedback upload failure)

Summary

An ACTIVE weekday Scheduled Task failed to run at 2026-08-12 08:40 KST.

The task remained ACTIVE, but no heartbeat Submission was created, no model execution started, and the configured command and Collector did not run.

Environment

  • OS: Windows
  • Component: ChatGPT desktop app / Codex Scheduled Tasks
  • Time zone: Asia/Seoul (UTC+9)
  • Schedule: Monday–Friday at 08:40 KST
  • Scheduled run: 2026-08-12 08:40 KST
  • Session/thread identifier: Withheld from this public issue and available privately if required

Expected behavior

The ACTIVE Scheduled Task should start automatically at 08:40 KST, create a heartbeat Submission, start the model, and then execute the configured command and Collector.

Actual behavior

  • The Scheduled Task did not start.
  • No heartbeat Submission was created.
  • No model execution occurred.
  • No command or Collector execution occurred.
  • No OAuth or external API requests occurred.
  • The task remained ACTIVE.
  • The missed run was not automatically retried.

Comparison

  • The same Scheduled Task ran successfully on the previous day.
  • An independent scheduled test also ran successfully later on 2026-08-12 at approximately 12:49:57 KST.
  • This therefore appears to be an intermittent missed scheduled run rather than a permanently broken task.

Additional information

The ChatGPT desktop app had been running since approximately 06:20 KST.

A PowerShell shell-snapshot warning was recorded around the scheduled time:

Failed to create shell snapshot for powershell: Shell snapshot not supported yet for PowerShell

However, the same warning has also occurred during successful scheduled runs, so it is not confirmed as the direct cause.

Requested investigation

Please check:

  1. Whether the 08:40 scheduled trigger was created and dispatched.
  2. Whether target-thread preparation or resume failed.
  3. Whether an internal exception occurred before heartbeat Submission creation.
  4. Why the missed scheduled run was not automatically retried.
  5. The internal difference between this missed run and the successful runs.
  6. Whether the PowerShell shell-snapshot warning can become fatal in a particular execution path.

I originally attempted to report this through /feedback, but the feedback uploader repeatedly failed. That separate problem is reported in issue #38130.

View original on GitHub ↗

2 Comments

boombx403-byte · 9 days ago

Hi @joon42066667-maker, the frozen history/resume behavior you described is one of the persistence patterns I've been studying. Codex Rescue Alpha5 includes a read-only projection consistency check that compares canonical rollout progress against stored projection cursors without touching SQLite.

If you still have the affected pre-repair session locally, you can test it with:

pip install codex-rescue==0.1.0a5
codex-rescue doctor --latest --json

(or point doctor directly to the rollout file). Please sanitize/redact any sensitive paths or names before sharing diagnostic output; raw session files are never needed.

quiezent · 3 days ago

Additional Windows reproduction on a newer Codex Desktop build — 2026-08-24

I observed the same failure signature on a newer Windows/Codex Desktop build.

Environment:

  • Windows NT 10.0.26200.0
  • Codex Desktop package 26.818.5229.0
  • Local timezone: Asia/Kuala_Lumpur (UTC+8)
  • Automation kind: existing-thread heartbeat
  • Stored schedule: Monday 21:30 local time

Expected:

  • At 2026-08-24 21:30 MYT, the ACTIVE heartbeat should submit a turn to its existing target task.

Observed:

  • No heartbeat input or task run appeared at 21:30 or 21:31.
  • No model turn started and no downstream task dispatch occurred.
  • The automation remained ACTIVE and retained the still-due 21:30 rule.
  • At 21:32 the user manually opened/prompted the target task.
  • A manual recovery read at 21:33:39 still found the heartbeat ACTIVE, correctly targeted, and still set to the missed 21:30 rule.
  • The manual task turn and its downstream task messages then worked normally.
  • The same heartbeat was manually advanced at 21:33:54 for the next checkpoint.

Controls checked:

  • Codex Desktop had been continuously running since 2026-08-22 19:42 MYT and was responsive.
  • Windows recorded no sleep/resume event during 21:20–21:40 MYT.
  • Exactly one local automation record existed; there was no duplicate schedule.
  • The target task was not already running at the due time.
  • No heartbeat Submission exists in the target task rollout at the scheduled time.

This appears to fail before target-task invocation. It is therefore not attributable to the automation prompt, target task logic, downstream messaging, laptop sleep, or a duplicate automation. The local task/automation identifiers are withheld here but can be supplied privately if maintainers need them.