What problem does it solve? When ECS tasks get stuck in PENDING or ACTIVATING, Docker NAT breaks, or awslogs fails on specific nodes, operators need a controlled, evidence-backed way to recover one node at a time without destabilizing the whole cluster. ## Core Features & Use Cases - Single-node restart with verification: Restarts docker and ecs via SSM on one EC2 instance, then confirms docker=active, ecs=active, and that the Docker NAT chain returned. - Optional drain before restart: Sets a container instance to DRAINING when the caller has ecs:UpdateContainerInstancesState, and treats missing permission as a hard stop. - Durable evidence capture: Writes before/after command output and a summary to a timestamped directory under ~/.AGENTS-temp for audit and review. - Use Case: A service is flapping because one ECS node's Docker NAT chain is missing. Resolve the container instance ARN from the EC2 ID, drain it if permitted, restart Docker and ECS, and verify NAT and service counts before moving on. ## Quick Start Ask the agent to restart Docker and ECS on EC2 instance i-0123456789abcdef0 in cluster production using your AWS profile and region, then verify the Docker NAT chain returned.
