Cloud routine (RemoteTrigger/schedule) sessions hang indefinitely with zero transcript activity

# Cloud routine (RemoteTrigger) sessions hang indefinitely with zero transcript activity

## Summary

Scheduled Claude Code cloud routines (created via the `/schedule` skill / RemoteTrigger API) on my account intermittently start provisioning correctly but then hang forever: the session log shows only infrastructure events (sandbox allocation, repo clone, "Claude Code process started") and then **nothing** — no user-prompt processing, no tool calls, no assistant output — for many hours. This has now happened on **6 out of 6** attempts for one specific routine, across both genuine cron-scheduled fires and manually-forced fires, including a brand-new trigger created from scratch. Two other independent routines on the same account have separately shown related but different connectivity/tooling issues over the same ~5-day window, suggesting a broader platform instability rather than one broken trigger.

## Account context

- Claude Pro account under `frans@micon-it.nl` (individual account, not an enterprise org)
- Cloud environment: a custom "Trusted"/Custom-network environment (`env_01NR1Hmi2tPKHetyEDj6xiQH`) with `api.open-meteo.com` and `opendata.dwd.de` on the Custom domain allowlist
- Git source: `FransM-ai/mainframe` (GitHub, private repo)

## Primary issue: routine hangs with zero transcript activity

**Routine:** "Bassano Skew-T-briefing" / "Bassano Skew-T-briefing v2"
**Trigger IDs:** `trig_01GtNGtkfnPKqUeAnnUiV8TB` (original, now disabled), `trig_016oPnCWagtAikmJtJaoTmz2` (recreated fresh to rule out per-trigger corruption — same result)

Failed sessions (all show: infra events complete normally, then `last_event_at == created_at` forever, no transcript):

| Date | Session ID | Trigger source | Result |
|---|---|---|---|
| 2026-08-21 | `cse_01PrrwuvWw627tGKttRU52KU` | one-off `run_once_at` test | stuck |
| 2026-08-22 | `cse_01S45pfZ1AEEMn3eKcsRhr2J` | one-off `run_once_at` test | stuck |
| 2026-08-22 | `cse_011QwxZYf7kX6t9LRWapUPmV` | manual `RemoteTrigger run` on newly-created recurring trigger | stuck |
| 2026-08-23 | `cse_01EyWFit1B2Cq3bRKwvEKaLH` | genuine cron fire, 04:50:50 UTC | stuck |
| 2026-08-24 | `cse_01J6Muu6WYQNUiHTyzrmfyRS` | genuine cron fire, 04:50:37 UTC | stuck |
| 2026-08-25 | `cse_01ARAm1P3cHBkANA6CiAW6Aj` | genuine cron fire on brand-new trigger v2, 04:50:39 UTC | stuck |

Sample log (identical pattern every time):
```
env[info]: Allocating sandbox
env[info]: Environment runner started
env[info]: Fetching repository FransM-ai/mainframe (x2)
env[info]: Finished processing sources
env[info]: No setup script configured
env[info]: Starting Claude Code
env[info]: Claude Code process started
[... nothing further, checked hours later ...]
```

The injected prompt was simply `/bassano-briefing` (a slash-skill invocation defined in the repo), same pattern used successfully by other routines on this account.

## Contrast: other routines on the same account/environment DO work

- **"Ochtendbriefing mail"** (`trig_01K8sW3AB2zEqCJyvk9PHjkt`), same custom environment, same git repo, fires daily and produces real transcript activity (multiple minutes of tool calls) and writes/commits files — but has separately shown an *intermittent* connectivity issue on the same days: `api.open-meteo.com` calls failing with `HTTP_STATUS:000`/exit 56, and `outlook_send_mail` missing from the session's tool list, on 5 consecutive days (2026-08-21 through 2026-08-25).
- **"T.C. Zuid agenda-sync"** (`trig_01PDaWnDgjsoEc76PwNG1SEM`) also fires and runs correctly, but `outlook_create_event` has been absent from the cloud-routine tool list for 5 consecutive daily runs (2026-08-21 through 2026-08-25), despite other Microsoft-365 tools (`outlook_calendar_search`, `outlook_modify_labels`, `outlook_create_label`, etc.) being present.

So: the Bassano routine hangs completely (0% success), while the other two routines run but hit partial/intermittent tool-availability or egress issues. All three problems appeared/worsened in the same ~5-day window (roughly since 2026-08-20/21), which is around when the account was migrated from an ITsmart-org Claude account to this individual micon-it.nl Pro account.

## Separately: interactive-session MCP tool-list staleness

In my interactive (non-routine) Claude Code session, `claude mcp list` reports `claude.ai Microsoft 365` as `✓ Connected`, but `ToolSearch` cannot find `outlook_send_mail` (or any `mcp__claude_ai_Microsoft_365__*` tool) for 3 consecutive days within the same long-running interactive session, even after the health check reports the server healthy. This may be a related symptom (stale tool registry not refreshing after a connector hiccup) or a separate bug.

## What I'd like

1. Confirmation of whether this is a known issue with the routine-scheduling infrastructure (stuck sessions with zero transcript activity).
2. Any guidance on why a brand-new trigger would reproduce the exact same hang as the original one (rules out trigger-specific corruption in my mind, but maybe there's something about my account/environment config that's the common factor).
3. Whether the `outlook_send_mail`/`outlook_create_event` tool-list gaps and the `api.open-meteo.com` egress flakiness are related to the same root cause or separate issues worth filing separately.

Happy to provide more session IDs, timestamps, or account details as needed.

Start by reviewing the /schedule skill and RemoteTrigger API flow for the listed session IDs, then compare the stuck Bassano sessions with the working routines in the same custom environment. Check where the infrastructure log stops after "Claude Code process started." Done means identifying a reproducible cause or confirming the account/environment-specific failure and documenting whether the tool-list and egress issues are related.

Written by the indexing model from the issue text.