Skip to main content

AWS operations plugin

Diagnose AWS and EKS incidents across compute, networking, identity, observability, storage, databases, analytics, and cost.

Specialists​

  • EKS agent: Cluster and node health, add-on lifecycle, upgrades, pod scheduling, CrashLoopBackOff, ImagePullBackOff, eviction, and resource pressure.
  • Network agent: VPC CNI, ENI/IP capacity, pod connectivity, load balancers, DNS, routes, network policies, security groups, and VPC endpoints.
  • IAM agent: IRSA, EKS Pod Identity, IAM trust and permissions, Kubernetes RBAC, EKS access entries, and legacy aws-auth mappings.
  • Observability agent: CloudWatch/Container Insights, Logs Insights, alarms, AMP, AMG, ADOT, and self-managed Prometheus/Grafana.
  • Storage agent: EBS/EFS, CSI drivers, persistent volumes/claims, StorageClasses, access modes, topology, attachment failures, throughput, and lifecycle.
  • Database agent: RDS/Aurora connectivity and performance, DynamoDB throttling/capacity, and ElastiCache connectivity, memory, and latency.
  • Analytics agent: OpenSearch and OpenSearch Serverless, ClickHouse, Athena, QuickSight, and Kinesis ingestion/query pipelines.
  • Cost agent: Service-level spending, EKS allocation, idle resources, utilization, storage lifecycle, and commitment/right-sizing opportunities.
  • Ops coordinator agent: Incident severity, five-minute triage, domain routing, cross-domain evidence, mitigation, verification, and postmortems.
  • Well-Architected agent: Review operational excellence, security, reliability, performance efficiency, cost optimization, and sustainability.

Workflows​

Use ops-troubleshoot for a concrete failure, ops-health-check for a general assessment, ops-network-diagnosis for connectivity, ops-observability for telemetry, ops-security-audit for security evidence, and ops-wellarchitected-review for a scored six-pillar review.

Start with scope and read-only evidence. Diagnose before changing resources, apply only authorized remediation, and verify against the original symptom. Single-domain work goes to a specialist; severe or multi-domain incidents can use the coordinator and parallel specialists where supported by the host.

The Claude manifest declares ten agents and six skills. Codex uses generated skill overlays and host-specific tools. The bundled MCP configuration is limited to the servers actually declared in the package.