Hermes Agent follows up on Claude Code's /loop: The Agent can wake up on its own schedule to do work

According to Motion Beating monitoring, Nous Research has added /loop to the Hermes Agent. Users can instruct the Agent to repeatedly perform the same task within the current session, such as monitoring CI, checking deployment status, or continuously fixing code until tests pass. Each trigger is a complete Agent execution, re-reading the latest state at that time.

Users can specify the interval for each execution themselves, or delegate the frequency to Hermes. The adaptive mode defaults to checking every minute. If the result remains unchanged for several consecutive checks, the interval decreases to 2, 4, or 8 minutes, with a maximum interval of 15 minutes. Once the result changes, the frequency reverts to every minute.

The loop feature also allows for setting the number of iterations and stop conditions. The Agent can autonomously determine when the task is complete, or users can use /loop --until to have an additional Judge check if certain conditions are met. By default, it pauses after running 100 times to prevent excessive token consumption in unmanned supervision scenarios.

This feature is a direct response to Claude Code's /loop, with full compatibility even with the /proactive alias. /loop runs only in the current session, making it more suitable for short-term tasks such as deployment monitoring and task completion tracking. For long-running tasks that require overnight execution or resumption after a terminal restart, the official recommendation is still to use Cron.