observability-telemetry

OpenTelemetry tracing propagates the W3C traceparent header (version-trace_id-parent_id-trace_flags) across all HTTP and gRPC boundaries. Each service parses incoming headers, starts child spans, and injects the context into outbound requests so the full request lifecycle appears as one trace.

Structured logging emits JSON events with high-cardinality fields like user_id, order_id, and trace_id instead of concatenated strings. This enables ad-hoc querying across arbitrary tags in backends like Honeycomb or Datadog, unlike flat text logs that require manual parsing.

Burn rate alerts fire only when the error budget consumption speed indicates imminent SLO violation, such as a 14.4x burn over 1 hour. This avoids alerting on transient spikes or symptom thresholds like CPU usage that do not affect users.

Yes, the skill detects the project's existing telemetry stack such as Datadog, Sentry, Winston, or Bunyan and applies observability principles to those tools. It explicitly never forces migration to a different platform.

This skill is not suited for database index tuning, auth configuration, or layout adjustments. Core web performance optimization tasks are routed to a dedicated performance skill instead.