SANDBOX_SLEEP_AFTER never doesn't work. My container is still sleeping!

Every night, and sometimes during the day, i will lose connection to my moltbot because the container shuts down/sleeps. Sometimes this happens mid conversation. Here is the conversation with my bot this morning diagnosing the issue. (the same convo we've had last couple days)

Me:
ok great, so, we need to debug something... last night i sent you a message on Telegram at 12:12 am, this was in the middle o9f our conversation about seting discord. Your last message before that was 12:11 am "Script works! Let me send a message there:"

You never replied to my message until this morning at 7:41 pm just now, this was AFTER i accessed the moltbot-sandbox gateway in my browser. So did you container go to sleep at 12:12 am and then not wake up until i hit the sandbox from my browser? Can you check your logs? (yesterday you set it up so that you would log container shutdowns)

BOT:
Restart log shows:
[2026-02-02 07:06:52 UTC] Container started - uptime: up 4 hours, 4 minutes
[2026-02-02 08:00:11 UTC] Container started | uptime: 17854s | location: qro02

So there were 2 restarts logged - but nothing after 08:00 UTC.

What likely happened:
After the 08:00 UTC restart, the container was running fine. We exchanged messages until ~08:12 UTC (your "Ok" message). Then... the container probably went idle and got paused by Cloudflare (not a full restart, just suspended).

When you hit /_admin/, that HTTP request woke the container back up.

The issue: Clawdbot's heartbeat crons were running internally, but if the container was suspended, the cron scheduler would also be frozen. When it woke up, everything resumed - including processing the queued "Ok" message.
The gap: ~7.5 hours of silence because the container was in a sleep state, not a crashed/restarted state.

Start with the SANDBOX_SLEEP_AFTER setting and the sandbox gateway's /_admin/ endpoint, then compare the reported Telegram gap with the startup and container-suspension logs. No source file or test is named; done means explaining why the container sleeps despite the setting and establishing the expected sleep behavior.

Written by the indexing model from the issue text.