1. Open the host's server health page
From the monitoring board, View history on the agent's card. The Alert thresholds section shows three rules: CPU, memory and disk, each marked Default rule or Custom rule.
2. Set the line and the sustained window
A rule reads as at or above N% for M minutes, clears under N% for M minutes. It fires only once the reading has held past the line for the whole window, not on one spike, and clears the same way. The recovery line must sit below the alert line. Windows run from 1 to 60 minutes.
3. Turn CPU alerting on if you want it
CPU is off by default on every host because it is the flap-happiest reading. The sustained window is what makes it safe: 90% for 5 minutes rides out a build or a backup. Tick Alert when CPU is busy and save.
4. Read the defaults, and reset when you are done experimenting
Disk alerts at 90% for 2 minutes and clears under 88% for 5; memory at 90% used for 5 minutes, clears under 88% for 5. Reset to default deletes your custom rule and the host follows the defaults again, including any future change to them.
Go deeper
The full reference lives in the docs: Server health documentation. Error codes named above are each explained in the error-code reference.