[BUG] Permission "Allow" prompt appeared during an unattended Cloud Routine run

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?

During a scheduled, unattended Cloud Routine run (claude.ai/code/routines), a permission "Allow" confirmation prompt appeared and blocked execution until I manually approved it on my phone. This contradicts documented behavior that Cloud Routines run fully autonomously with no approval prompts during execution.

• Routine: "سلطان - تقرير التحليلات الأسبوعي" (trigger id trig_012eqsHPrBFUf7LRxZhN5b4h)
• Time: 2026-09-13, roughly 4:07-4:19 PM (Riyadh time, UTC+3)
• I received a push notification from the Claude app; opening it took me directly to the live transcript of this scheduled routine run, confirming the prompt occurred inside a genuine unattended Cloud Routine, not an interactive session.
• I recall approximately 3 such prompts during this run.

What Should Happen?

Cloud Routines should run fully unattended with no permission/approval prompts at all, as documented. The scheduled report should have completed and delivered on its own without requiring me to be available to tap "Allow" on my phone. If I'm asleep or unavailable when a routine fires, execution should not stall waiting for approval that may never come.

Error Messages/Logs

Steps to Reproduce

This occurred during an unattended, scheduled Cloud Routine run (claude.ai/code/routines) — not a manual/interactive CLI session, so I cannot reliably force it to reproduce on demand.

1. A Cloud Routine ("سلطان - تقرير التحليلات الأسبوعي") fired on its cron schedule (Sunday 4:00 PM Riyadh / 13:00 UTC) with no human present, per documented unattended-execution design.
2. During the run, the Claude app sent me a push notification.
3. Opening the notification took me directly to the live transcript of this scheduled routine run at claude.ai/code/routines.
4. A permission "Allow" confirmation dialog was shown there and blocked the routine's progress until I manually tapped Allow.
5. This happened approximately 3 times during the same run (4:07-4:19 PM Riyadh time, 2026-09-13).

I don't have a minimal reproducible example since this is intermittent and tied to an autonomous scheduled run, not a command I invoked directly.

Claude Model

Sonnet (default)

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

N/A — occurred in a claude.ai Cloud Routine (web/cloud), not a local CLI session

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

Note: several required fields above (Claude Model, Claude Code Version, Terminal/Shell, Operating System) don't cleanly apply — this bug occurred inside a claude.ai Cloud Routine (a scheduled, unattended cron-triggered session running in Anthropic's cloud), not in a local CLI/terminal session on my machine. I filled them with my local device's details only because the form requires an answer.

Routine trigger id: trig_012eqsHPrBFUf7LRxZhN5b4h
Routine name: "سلطان - تقرير التحليلات الأسبوعي"
Run time: 2026-09-13, ~4:07-4:19 PM Riyadh time (UTC+3)