@cloudflare/think@0.8.8

cloudflare/agents@cloudflare/think@0.8.8Jun 8, 2026by github-actions[bot]

AI Summary

A patch release fixing a bug where agents fired alarms every 30 seconds indefinitely when they did not use workflow notifications.

Key Highlights

Fixes alarm firing loop for agents without workflow notifications
Agents now correctly hibernate after the fix
`_startWorkflowNotificationDrain()` now returns early when no notifications exist

New Features

Bug fix for alarm behavior
Improved hibernation behavior

Full Release Notes

### Patch Changes

-   [#1705](https://github.com/cloudflare/agents/pull/1705) [`611e8c3`](https://github.com/cloudflare/agents/commit/611e8c3417b06ec6b284e38c4a1ba73730530a10) Thanks [@threepointone](https://github.com/threepointone)! - Fix Think agents firing an alarm every 30s forever when they don't use workflow notifications ([#1703](https://github.com/cloudflare/agents/issues/1703)).

    `alarm()` called `_startWorkflowNotificationDrain()` unconditionally, which wrapped its work in `keepAliveWhile(...)`. Acquiring the keepAlive lease armed the 30s keepAlive heartbeat alarm even though there was nothing to drain, and releasing the lease did not pull the alarm back — so the DO re-scheduled itself every 30s and never hibernated.

    `_startWorkflowNotificationDrain()` now returns early when there are no pending notifications, matching its other call sites. Affected DOs self-heal on their next alarm fire after upgrading: `super.alarm()` reschedules to the next legitimate task (or clears the alarm entirely) and the drain no longer re-arms the heartbeat.