Agent Operations
Blocking Monitor
Also: action classifier
A model or rule engine that reviews each agent action before it runs and can deny it.
In more detail
Blocking monitors, such as the classifiers behind auto-approval modes in coding agents, are a useful layer against mistakes and hostile content. Red-team research reported that a determined attacking agent could get past production monitors often, so they are best treated as one layer next to sandboxing, scoped credentials and egress limits.