≈644 tokens on demand. Your agent reads this file only when SKILL.md points to it.
AWS Infrastructure Specialist Reference
Purpose: Use this file when AWS-specific networking, compute, database, messaging, IAM, or cost decisions materially affect the design.
Contents:
- Networking decisions
- Compute service selection
- Database and storage choices
- Messaging and orchestration
- IAM and account strategy
- AWS-specific cost optimization
Networking Decisions
| Decision |
Preferred choice |
3 or fewer VPCs |
VPC Peering |
4+ VPCs or on-prem integration |
Transit Gateway |
| S3 / DynamoDB private access |
Gateway Endpoint (free) |
| Other private AWS service access |
Interface Endpoint (~$7.5/month + transfer) |
Compute Service Selection
| Use case |
Recommended |
Alternative |
Avoid |
| Always-on web API |
ECS Fargate |
EKS if Kubernetes is required |
Direct EC2 management |
| Low-traffic API |
Lambda or App Runner |
Fargate |
EC2 |
| Short batch jobs |
Lambda |
Step Functions + Lambda |
EC2 |
| Long-running batch |
ECS Tasks |
Step Functions + ECS |
Lambda |
| Existing Kubernetes platform |
EKS |
Fargate for simpler services |
unmanaged clusters |
Database And Storage Choices
| Need |
Recommended choice |
| General relational DB |
Aurora Serverless v2 or RDS |
| Unpredictable key-value workload |
DynamoDB On-Demand |
| Stable predictable DynamoDB workload |
Provisioned + Auto Scaling |
| Lifecycle-based object storage savings |
S3 lifecycle + Intelligent Tiering / IA / Glacier |
| CDN + static/private origin |
CloudFront + OAC |
Messaging And Orchestration
| Need |
Recommended |
| General async queue |
SQS Standard |
| Strict ordering / dedupe |
SQS FIFO |
| Event routing |
EventBridge |
Workflow with 3+ steps and branching/retry |
Step Functions |
IAM And Account Strategy
- Prefer multi-account separation for management, security, logs, shared services, and workloads.
- Use permission boundaries for delegated administration.
- Use cross-account roles with
ExternalId.
- Enforce region boundaries with SCPs where appropriate.
- No
0.0.0.0/0 ingress by default.
AWS-Specific Cost Optimization
- Prefer
Compute Savings Plans before stricter EC2-only commitments.
- Graviton can reduce compute cost by about
20%.
- Monitor CPU and memory for at least
2+ weeks.
- Average utilization under
40% is a strong downsizing signal.
- Consider scheduled shutdowns for
dev and staging.
- Use VPC endpoints to reduce NAT Gateway cost.
