aws-systems-cost-reduction overview

End-to-end AWS cost-reduction workflow. All tasks are agent-driven; no inter-task filesystem contract — agents self-manage state under their task-scoped directory and pass typed results forward as JSON. Caller controls the report destination via `inputs.outputDir`. Risk tiers: 1 - cosmetic (Route53 dangling, untagged, log retention=Never) 2 - rightsize (Compute Optimizer EC2; gp2->gp3; S3 lifecycle) 3 - reservations / Savings Plans (irreversible 1y/3y commit) 4 - decommission (REQUIRES backup unless waived) 5 - cluster-level (EKS shrink/delete) AWS-specific waste patterns to detect: - NAT Gateways at $32+/mo + per-GB processed - gp2 EBS (gp3 always cheaper, same perf) - Old AMI snapshots from terminated instances (often 100s GB) - Idle ALBs (zero ProcessedBytes 30d) - Lambda provisioned concurrency on funcs w/ no invocations - VPC Interface endpoints at $7.20/mo each - CloudWatch Logs retention=Never on high-ingestion groups - Unattached EIPs at $3.60/mo each