What problem does it solve?
Kubernetes incidents like CrashLoopBackOff, OOMKilled pods, stuck rollouts, and node pressure are hard to root-cause because signals are scattered across metrics, events, logs, and APM. This Skill gives an AI agent a structured investigation methodology to correlate OTel telemetry collected via EDOT and pinpoint the failure mode with calibrated confidence.
Core Features & Use Cases
- Failure-mode taxonomy: Classifies incidents across workload, node, control-plane, autoscaling, and networking layers using pivotal signals like
k8s.container.status.last_terminated_reason and k8s.pod.cpu_limit_utilization.
- Guided investigation flow: Walks through orient, characterize, classify, corroborate, and synthesize phases with ready-to-use ES|QL query recipes against OTel receiver indices.
- Confidence-calibrated synthesis: Produces structured incident reports with hypothesis, evidence, confidence level, and recommended next steps, explicitly avoiding confabulation from empty results.
- Use Case: When a pod alert fires for CrashLoopBackOff, the agent checks restart counts and termination reasons, distinguishes OOMKilled from CPU throttling from probe misconfiguration, corroborates with K8s events and baselines, and delivers a root-cause hypothesis.
Quick Start
Ask the agent to investigate why a specific pod in a given namespace is CrashLoopBackOff-ing using the Kubernetes investigation skill.
