Observe Trace

You need to see how a distributed task actually executed across multiple agents, and this builds you a trace tree from span data stored in observability memory. It pulls all the spans for a task ID, reconstructs the parent-child hierarchy, calculates durations, and flags bottlenecks where things got slow or agents sat idle. The synthesis step is smart: it uses agentdb to turn raw span metadata into a readable narrative of what happened. Most useful when debugging coordination problems or hunting performance issues in multi-agent workflows. The trace tree shows you the critical path and total duration, so you can see exactly where time got burned.