Problem

On Windows, the /restart command (and config-change triggered restarts) frequently fail to relaunch the gateway. The gateway shuts down but never comes back, requiring manual openclaw gateway restart.

Root Cause

Commit 6d26609 (#52487) added a "Running" state check to the cmd.exe handoff script to avoid duplicate schtasks /Run calls. However, this introduces a race condition:

1. /restart fires → gateway process exits
2. Handoff script spawns and checks Get-ScheduledTask.State
3. Windows Task Scheduler still reports "Running" due to delayed state updates after process exit
4. Handoff script sees "Running" → assumes another path already restarted → jumps to cleanup
5. No new instance is launched — gateway stays offline

The cmd script also spawns powershell.exe | findstr.exe, which creates visible console popup windows even with windowsHide: true, because the piped child processes aren't hidden.

Fix

Replace the cmd.exe handoff script with a detached Node.js subprocess that:

1. Writes a temporary .mjs restart launcher
2. schtasks /End — signals the Task Scheduler to stop the old task
3. The launcher polls schtasks /Query until the task is confirmed NOT running
4. Force-kills the process on port 18789 if the task won't stop (timeout fallback)
5. schtasks /Run — starts the new instance once the old one is confirmed gone
6. Falls back to launching gateway.cmd directly if schtasks /Run fails

Why Node.js instead of cmd script?

• ✅ No console popups — windowsHide: true works correctly on Node.js child processes
• ✅ No race condition — explicitly waits for task to stop before launching new instance
• ✅ Better logging — structured log lines with timestamps
• ✅ Force-kill fallback — handles stuck processes that refuse to die

Real behavior proof (required for external PRs)

• Behavior or issue addressed: Windows /restart should relaunch the managed OpenClaw Gateway Scheduled Task without leaving the gateway offline and without visible cmd / findstr console popups.

• Real environment tested: Real Windows 10 machine (Build 19045), OpenClaw 2026.6.6, managed Windows Scheduled Task named OpenClaw Gateway, WebChat /restart command.

• Exact steps or command run after this patch:

  1. Reproduced the issue with the published Windows task handoff implementation.
  2. Observed /restart shut down the gateway and leave it offline until manual openclaw gateway restart.
  3. Patched the local installed OpenClaw restart handoff to use the Node subprocess strategy from this PR.
  4. Restarted the gateway so the patched code was loaded.
  5. Ran /restart twice from WebChat.
  6. Observed the gateway relaunch both times with no black console windows.

• Evidence after fix (screenshot, recording, terminal capture, console output, redacted runtime log, linked artifact, or copied live output):

  2026/6/13 23:06:25 openclaw restart attempt source=node-handoff target=OpenClaw Gateway
  2026/6/13 23:06:25 launching task via schtasks /Run
  2026/6/13 23:06:25 schtasks /Run exit=0 out=成功: 尝试运行计划任务 "OpenClaw Gateway"�?  2026/6/13 23:06:25 openclaw restart finished source=node-handoff
  2026/6/13 23:08:21 openclaw restart attempt source=node-handoff target=OpenClaw Gateway
  2026/6/13 23:08:21 launching task via schtasks /Run
  2026/6/13 23:08:21 schtasks /Run exit=0 out=成功: 尝试运行计划任务 "OpenClaw Gateway"�?  2026/6/13 23:08:21 openclaw restart finished source=node-handoff
  

• Observed result after fix: /restart completed successfully on both attempts, the gateway came back immediately, and no visible findstr / cmd console windows appeared.

• What was not tested: Windows 11 and non-default gateway ports.

• Proof limitations or environment constraints: The proof used copied redacted runtime logs instead of screenshots/recordings. The task name tested was the default OpenClaw Gateway; custom task names are covered by unit tests but not by the live Windows proof.

• Before evidence (optional but encouraged):

  [周六 2026/06/13 22:30:07.92] openclaw restart log initialized
  [周六 2026/06/13 22:30:07.92] openclaw restart attempt source=windows-task-handoff target="OpenClaw Gateway"
  [周六 2026/06/13 22:31:11.95] openclaw restart finished source=windows-task-handoff
  [周六 2026/06/13 22:40:37.72] openclaw restart log initialized
  [周六 2026/06/13 22:40:37.72] openclaw restart attempt source=windows-task-handoff target="OpenClaw Gateway"
  [周六 2026/06/13 22:41:35.34] openclaw restart finished source=windows-task-handoff
  

  Those 60s+ gaps corresponded to the old handoff path; the gateway did not reliably relaunch and the user had to manually recover it. During this path, visible black console windows appeared with findstr involved.

Tests and validation

Which commands did you run?

• GitHub CI for the latest pushed commit: check-prod-types, check-test-types, check-lint, and the affected infra test shard passed.
• Live Windows /restart command was run twice from WebChat after applying the local patch.

What regression coverage was added or updated?

• Updated src/infra/windows-task-restart.test.ts to cover the Node handoff launcher, custom task names, fallback content, and windowsHide: true.