HiAi-gg Investigate Performance (Sentry Trace)
Use this skill when an agent must find where a request is slow or which span failed.
Workflow
1. Locate the trace
- Identify the trace/transaction: from the event's trace id, the issue's samples, or the user's description of the slow request.
- Fetch the transaction and its span tree.
2. Find the bottleneck
- List the top spans by duration (self-time, not wall time).
- Flag spans whose time dominates the transaction, or that exceed a sensible budget for the operation type.
- Look for the classic offenders: external HTTP calls, database queries, serialization, I/O waits.
3. Trace failures
- If a span errored: which operation, what status/error, and which downstream system (service, DB, external API)?
- Correlate the failing span with the issue/exception if one exists.
4. Correlate with infrastructure
- Note span attributes: service name, DB statement type, HTTP status.
- If a downstream service is slow, say which one — do not assume it is the application's fault.
5. Report
Give: the transaction, the top spans with durations, the bottleneck or failing span, and the most likely fix (optimize the query, add a timeout/retry, cache, or escalate to the owning service). Tracing is read-only.