name	analyze-trace
description	Walk a Lensflare trace span-by-span and explain timing, errors, and parent/child structure. Use when the user pastes a trace id, asks "why is this slow", or wants a request-level post-mortem of a real production / dev request captured in Lensflare.

Analyze a trace in Lensflare

If the user supplied a traceId, use it. Otherwise call lensflare:queryTelemetry to find a relevant trace (filter by serviceName, time, or attributes). Pick the trace whose root span has the highest durationUs or whose status = "error".
Call lensflare:getTrace with the resolved datasetId and traceId. If it returns null, double-check the traceId with queryTelemetry (traceId = "<id>") before reporting back.
Read summary.spanCount and summary.erroredSpanCount. If errors exist, lead with them: name + service + durationUs.
Walk spans in order. Highlight the longest span, the deepest chain, and any span whose status = "error" or whose duration is more than 50% of its parent's.
Surface relatedEvents only when they materially explain a failure (e.g. an exception event on the failing span).
End with a one-line root-cause hypothesis and a suggested follow-up query. Example: "to see all requests with this error: attributes.http.status_code = 500 and serviceName = \"api\"".

name	analyze-trace
description	Walk a Lensflare trace span-by-span and explain timing, errors, and parent/child structure. Use when the user pastes a trace id, asks "why is this slow", or wants a request-level post-mortem of a real production / dev request captured in Lensflare.

If the user supplied a traceId, use it. Otherwise call lensflare:queryTelemetry to find a relevant trace (filter by serviceName, time, or attributes). Pick the trace whose root span has the highest durationUs or whose status = "error".
Call lensflare:getTrace with the resolved datasetId and traceId. If it returns null, double-check the traceId with queryTelemetry (traceId = "<id>") before reporting back.
Read summary.spanCount and summary.erroredSpanCount. If errors exist, lead with them: name + service + durationUs.
Walk spans in order. Highlight the longest span, the deepest chain, and any span whose status = "error" or whose duration is more than 50% of its parent's.
Surface relatedEvents only when they materially explain a failure (e.g. an exception event on the failing span).
End with a one-line root-cause hypothesis and a suggested follow-up query. Example: "to see all requests with this error: attributes.http.status_code = 500 and serviceName = \"api\"".

analyze-trace