ECS / Fargate deep-dive on Kuberly clusters via the Platform MCP aws workflow: diagnose tasks that won't start, crash, OOM (RESOURCE:MEMORY), fail image pulls (CannotPullContainerError), can't assume a role, or sit behind unhealthy ALB targets. Walks aws_ecs_* service/task tools, CloudWatch logs and metrics, ALB health, CloudTrail audit, and VPC/SG correlation. Use after the router picks the ECS path.
