Claude Code adds Monitor

Claude Code has added a Monitor tool, allowing the agent to directly create background scripts to watch for events and wake up only when needed.

The core change is shifting from repeated polling to event-driven monitoring. Instead of continuously checking status while consuming tokens, the agent now reacts only when an actual change occurs.

Usage examples include the following.

• Log monitoring: Call immediately when an error occurs
• PR tracking: Periodically check PR status via script
• Build failure detection: Wake the agent for follow-up action upon failure

In the comments, users mentioned experience building similar watchers directly with cron jobs, and noted that with Monitor, the agent can return while retaining the context of the monitoring setup as an advantage. In particular, there was anticipation about whether it could detect a build failure, trace the change that caused it, and even suggest a fix.