Observability Llm Obs

It knows to check both OpenTelemetry traces (where instrumentation like EDOT or OpenLLMetry sends span data with token counts, model names, and call chains) and integration metrics from providers like OpenAI, Bedrock, or Vertex AI. The skill teaches Claude to discover what data streams actually exist before querying, then use ES|QL and Elasticsearch APIs to pull performance, cost, token usage, and agentic workflow details. It explicitly works without Kibana UI and handles the fact that deployments mix ingestion paths. Use this when you need Claude to investigate LLM costs, trace multi-step agent calls, or debug response quality issues from observability data you're already collecting.