[BUG] [BUG] Routines (scheduled tasks) stuck in PENDING forever — sessions never execute

### Preflight Checklist

- [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) 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?

Multiple Routines (Claude Code Remote scheduled tasks), fired three different ways (manual fire_trigger, a run_once_at one-shot, and a normal recurring cron), all remain in ROUTINE_RUN_STATUS_PENDING for 30-300+ minutes with zero execution — not even the first tool call in the prompt runs.

A completely unrelated pre-existing Routine (different topic, no relation to the ones below) is also stuck in PENDING since its normal cron fire ~5 hours earlier, ruling out anything specific to my own trigger config.

Evidence:
- trig_01AHeM7YFgitwkFfhYPvLAiy, fired 2026-09-08T17:16:22Z via fire_trigger, session cse_01Au4AThYexzdxdSXcsU5HF2, still PENDING, 0 files written by its task.
- trig_01XW4kM2xVUeAdKLsvQEHWxT, fired 2026-09-08T17:30:55Z via fire_trigger, session cse_01QuUuKfMfZ1wBLsz18ALPn6, still PENDING, 0 files written.
- trig_01ESXPE9SnHBC78teZpWCxhR, a run_once_at fired by the scheduler itself at 2026-09-08T17:56:47Z, session cse_01S3NTBqt9ARaetLEuf7AcC6, still PENDING, 0 files written.
- An unrelated Routine unconnected to any of the above ("Check quotidien Bugs & Idées") stuck in PENDING since its normal cron fire at 13:09 UTC the same day (~5h).

Ruled out: not an Airtable issue (Airtable is never reached — first step is a Dropbox call), not a Dropbox auth issue (same connector worked fine in an interactive session same evening), not specific to fire_trigger (the run_once scheduler fire shows the same symptom).

### What Should Happen?

I should be able to save data on airtable without tedious work using Claude pro

### Error Messages/Logs

```shell
as per lower
```

### Steps to Reproduce

Summary

Multiple Routines (Claude Code Remote scheduled tasks), triggered three different ways (manual fire_trigger, a run_once_at one-shot, and a normal recurring cron), all remain in ROUTINE_RUN_STATUS_PENDING for 30–300+ minutes with zero observable execution — not even the first tool call in the prompt runs. A completely unrelated pre-existing Routine (topic: "Check quotidien Bugs & Idées", unrelated to any of the below) is also stuck in PENDING since its normal cron fire ~5 hours ago, which rules out anything specific to our own trigger configuration and points to a platform-side execution/session-scheduling failure.

Environment

• Product: Cowork / Claude Code Remote scheduled tasks ("Routines"), MCP server claude-code-remote


• Account: lecreatif1@gmail.com


• Date/time window observed: 2026-09-08, 17:16 UTC through at least 18:33 UTC (evening, US/Eastern-adjacent workflow)


• Model configured on affected Routines: claude-sonnet-5

Evidence (session/trigger IDs, all reproducible via list_triggers)

For the three Alfred triggers, the target folder in the connected Dropbox account (/innovation/Alfred data/2026-09/) shows zero new files, not even the very first step of the prompt (a dropbox create_folder call), which confirms the underlying session never actually began executing tool calls — this is not a slow/long-running task, it is a task that never started.

What we ruled out

• Not an Airtable API issue: no Airtable call was ever reached (first step is Dropbox create_folder).


• Not a Dropbox connector/auth issue: the same Dropbox MCP connector worked normally in an interactive session earlier the same evening (12 files written successfully).


• Not specific to our trigger's prompt/config: an unrelated, previously-existing Routine with a different prompt and no Airtable/Dropbox connectors is stuck identically.


• Not specific to the fire_trigger manual-trigger path: a run_once_at fire driven by the scheduler itself shows the same symptom.

Historical related data point (may or may not be the same root cause — flagging for correlation)

On 2026-09-01, a different (now-disabled) Routine trig_01BF7T6yAc9iQq9poV2C8MUj (14-table Airtable→Dropbox export in a single run) reported last_run.status: ROUTINE_RUN_STATUS_ABANDONED rather than PENDING — i.e. a session did start and later was abandoned, as opposed to tonight's sessions which never leave PENDING at all. We originally attributed that failure purely to the prompt design (too much data fetched/written in one unattended session). Given tonight's platform-wide PENDING-forever pattern, we now suspect Routine execution reliability may have a broader, possibly longer-standing issue, and the September 1st ABANDONED result may be a related (though not identical) symptom of the same underlying execution infrastructure, not purely a prompt-design problem. We are not certain of this and would appreciate Anthropic's own visibility into it.

Summary

Multiple Routines (Claude Code Remote scheduled tasks), triggered three different ways (manual fire_trigger, a run_once_at one-shot, and a normal recurring cron), all remain in ROUTINE_RUN_STATUS_PENDING for 30–300+ minutes with zero observable execution — not even the first tool call in the prompt runs. A completely unrelated pre-existing Routine (topic: "Check quotidien Bugs & Idées", unrelated to any of the below) is also stuck in PENDING since its normal cron fire ~5 hours ago, which rules out anything specific to our own trigger configuration and points to a platform-side execution/session-scheduling failure.

Environment
Product: Cowork / Claude Code Remote scheduled tasks ("Routines"), MCP server claude-code-remote
Account: [lecreatif1@gmail.com](mailto:lecreatif1@gmail.com)
Date/time window observed: 2026-09-08, 17:16 UTC through at least 18:33 UTC (evening, US/Eastern-adjacent workflow)
Model configured on affected Routines: claude-sonnet-5
Evidence (session/trigger IDs, all reproducible via list_triggers)
Trigger Trigger ID Fire mechanism Fired at (UTC) Session ID Status after 10–300+ min
Alfred — Backup Contributions trig_01AHeM7YFgitwkFfhYPvLAiy manual fire_trigger 2026-09-08T17:16:22Z cse_01Au4AThYexzdxdSXcsU5HF2 still PENDING, 0 files written
Alfred — Backup Chambres trig_01XW4kM2xVUeAdKLsvQEHWxT manual fire_trigger 2026-09-08T17:30:55Z cse_01QuUuKfMfZ1wBLsz18ALPn6 still PENDING, 0 files written
TEST — Cron réel (Chambres) trig_01ESXPE9SnHBC78teZpWCxhR run_once_at fired by scheduler itself (not manual) 2026-09-08T17:56:47Z cse_01S3NTBqt9ARaetLEuf7AcC6 still PENDING, 0 files written
Check quotidien Bugs & Idées (unrelated pre-existing Routine, no Airtable/Dropbox involvement) trig_01AMhTRc… (truncated in our tooling) normal recurring cron 2026-09-08T13:09Z (see list_triggers) still PENDING after ~5 hours

For the three Alfred triggers, the target folder in the connected Dropbox account (/innovation/Alfred data/2026-09/) shows zero new files, not even the very first step of the prompt (a dropbox create_folder call), which confirms the underlying session never actually began executing tool calls — this is not a slow/long-running task, it is a task that never started.

What we ruled out
Not an Airtable API issue: no Airtable call was ever reached (first step is Dropbox create_folder).
Not a Dropbox connector/auth issue: the same Dropbox MCP connector worked normally in an interactive session earlier the same evening (12 files written successfully).
Not specific to our trigger's prompt/config: an unrelated, previously-existing Routine with a different prompt and no Airtable/Dropbox connectors is stuck identically.
Not specific to the fire_trigger manual-trigger path: a run_once_at fire driven by the scheduler itself shows the same symptom.
Historical related data point (may or may not be the same root cause — flagging for correlation)

On 2026-09-01, a different (now-disabled) Routine trig_01BF7T6yAc9iQq9poV2C8MUj (14-table Airtable→Dropbox export in a single run) reported last_run.status: ROUTINE_RUN_STATUS_ABANDONED rather than PENDING — i.e. a session did start and later was abandoned, as opposed to tonight's sessions which never leave PENDING at all. We originally attributed that failure purely to the prompt design (too much data fetched/written in one unattended session). Given tonight's platform-wide PENDING-forever pattern, we now suspect Routine execution reliability may have a broader, possibly longer-standing issue, and the September 1st ABANDONED result may be a related (though not identical) symptom of the same underlying execution infrastructure, not purely a prompt-design problem. We are not certain of this and would appreciate Anthropic's own visibility into it.

### Claude Model

Sonnet (default)

### Is this a regression?

No, this never worked

### Last Working Version

_No response_

### Claude Code Version

5.0

### Platform

Anthropic API

### Operating System

Windows

### Terminal/Shell

Cursor

### Additional Information

Ask
Confirm whether there was a known incident/degradation in Routine session-scheduling execution around 2026-09-08 17:00–19:00 UTC.
Advise whether ROUTINE_RUN_STATUS_PENDING has a defined timeout/retry

No repository file or test is identified in the report. Start by using the mentioned list_triggers entry point to verify the affected trigger and session statuses, then trace the Routine session-scheduling path; done means scheduled runs leave PENDING and execute their first tool call.

由索引模型根据 Issue 内容生成。