[BUG] Scheduled task push notifications report "failed — lost network" for runs that completed successfully on-device; one task also fires duplicate notices

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

Desktop app, two daily local scheduled tasks (cron, both fire ~7 AM). On mornings when the machine is asleep at the scheduled time, and I wake it after, both tasks run successfully and complete their actual work on the device. Despite that, I then get a push notification on my phone reporting the run failed because the computer lost contact with the internet before it could complete, for both tasks, every time. The Desktop app itself shows the run completed normally; the failure notification does not reflect the run's actual outcome at all.

Separately, one of the two tasks (crostini-bwrap-pr776-check, whose steps include browsing via Claude in Chrome) also shows duplication: 2 "can't run, not connected to the internet" notices pre-wake instead of 1, 2 false "failed — lost network" notices post-completion instead of 1, and 2 Chrome tab groups instead of 1 — the phantom duplicate run opens its own separate tab group in addition to the one the legitimate run creates. This phantom entry also briefly appears in the Desktop app's own task list, rendered in what looks like a "Run History" view distinct from the normal chat-transcript view, then disappears on its own. daily-receipt-processing, which has no browser step, shows none of this — just the single false-failure notification described above.

What Should Happen?

A push notification's reported status should match the run's actual outcome. A run that completes successfully in the Desktop app should not generate a "failed — lost network" notification. Each logical run should produce exactly one notification, not a variable number depending on task type.

Error Messages/Logs

Steps to Reproduce

Step 1: Have two or more daily local scheduled tasks configured, at least one of which uses a Chrome-dependent tool/step.

Step 2: Let the machine sleep through the scheduled time; wake it afterward so the tasks run late.

Step 3: Observe push notifications before and after wake, and compare against the Desktop app's own run status for the same tasks.

Claude Model

None

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.7032.0 (aaddrick/claude-desktop-debian unofficial package 3.2.4, ChromeOS/Crostini bwrap backend)

Platform

Other

Operating System

Ubuntu/Debian Linux

Terminal/Shell

Other

Additional Information

Related but distinct: issue #95450's author separately flagged an "extra trigger firing" data point as likely a different bug from that issue's main regression, so this report may be that bug.

Also related in the notification pipeline generally, though not the same symptom: issue #82728 (scheduler misreports run outcome status) and issue #84645 (notification channel-specific delivery failures).