Observe Metrics

Pulls metrics from the observability namespace, aggregates them by type (counters, gauges, histograms), and compares against baselines stored in ReasoningBank to flag anomalies beyond two standard deviations. You get p50/p95/p99 percentiles for durations, sums for task completions and errors, and a color-coded health score. It follows the dual-path pattern, storing both typed snapshots via pattern-store and timestamped records in memory. Useful when you need a quick read on swarm health or want to catch degradation before it cascades. Default period is one hour but you can pass --period to adjust. The anomaly detection is straightforward statistics, nothing fancy, but often that's exactly what you need.