Silent AI Monitoring: Zero Output on Success, Alerts Only on Failure

A developer applies Unix philosophy to AI agent monitoring, designing hooks that produce no output on success and only communicate via exit codes. The system uses two layers: config-health for advisory checks and quality-gate for hard blocks on stale outputs or low disk space. After two weeks, the silent approach caught a critical disk space alert that would have been missed in noisy systems.