observability

Scope

Covers telemetry as a production capability: the OpenTelemetry data model and how traces, metrics and logs are joined; SDK instrumentation (automatic versus manual, context propagation, span naming, kind, status and attributes, error recording); semantic conventions — which are stable, which still move, and how an instrumentation library switches between them; Collector pipelines (receivers, processors, exporters, connectors, ordering, memory limits, agent versus gateway deployment, head and tail sampling); metric cardinality as a cost and a correctness problem; structured logging and its correlation to traces; RED and USE metrics; SLOs, error budgets and burn-rate alerting; and the workflow that gets from an alert back to a line of code.

Where a local failure can be reproduced, use the debugging skill; where only production signals exist (logs, metrics, traces, alerts), this is the observability skill's job.

Collector facts in this skill are written against otelcol-contrib 0.160.0 and semantic conventions v1.44.0; rules that depend on a newer component than that carry a gate such as (Collector 0.156+).