aws-skills.md
AWS Skills
Use when designing or implementing AWS infrastructure with CDK, optimizing cost, or building serverless event-driven architectures.
How to use this Claude skill ↓
- Click Download below to save the
.mdfile. - Open claude.ai and create a new Project.
- In Project settings, paste the file content into Custom instructions.
- Start a conversation — Claude will now act as the specialist defined by this skill.
STATS
Downloads
0
Views
29
Category
Coding
Added
May 19, 2026
SKILL CONTENT
aws-skills.md4307 B
# AWS Skills Build, deploy, and optimize AWS infrastructure using CDK and modern patterns — without overspending or overengineering. ## Keywords aws, amazon web services, cdk, cdk v2, cloudformation, terraform, lambda, ec2, s3, rds, dynamodb, sqs, sns, eventbridge, api gateway, cloudwatch, iam, vpc, serverless, event-driven, well-architected, cost optimization, reserved instances, savings plans, fargate, ecs, eks ## Core Truth Most AWS bills are 60% bigger than they need to be. The pattern is consistent: overprovisioned instances, unused load balancers, S3 buckets without lifecycle policies, EBS volumes detached but billed, NAT gateways carrying single-AZ traffic at $0.045/GB. Architecture quality and cost are the same conversation. A bad architecture is expensive; an expensive architecture is usually bad. --- ## 1. When To Use - Designing new AWS infrastructure - Reviewing existing infra for cost or reliability - Writing CDK or CloudFormation - Picking between Lambda, ECS, EC2 for a workload - Diagnosing a high AWS bill - Setting up CI/CD on AWS ## 2. Workload → Service Decision | Workload | Use | Avoid | |----------|-----|-------| | < 15 min stateless requests | Lambda | EC2 (idle cost) | | Long-running services | ECS Fargate | Lambda (timeout limit) | | Spiky workloads | Lambda or Fargate Spot | Reserved EC2 | | 24/7 high-volume | EC2 with Savings Plan | Lambda (cost) | | Async work queue | SQS + Lambda | API Gateway (sync) | | Pub/sub | EventBridge or SNS | SQS (point-to-point only) | | Time-series data | Timestream or S3 + Athena | RDS | | Global key-value | DynamoDB | RDS multi-region | ## 3. CDK Best Practices - Use CDK v2 — v1 is end-of-life - One stack per deployment unit - Stacks should be < 200 resources (CloudFormation limit) - Use constructs for reusable patterns - Tag everything (Environment, Owner, CostCenter) - Enable termination protection on production stacks - Use `cdk diff` before every deploy ## 4. Cost Optimization Hit List In order of usual impact: 1. **NAT gateways** — most overlooked. Use VPC endpoints for S3/DynamoDB. 2. **Idle load balancers** — $16/mo per ALB. Delete the dev ones. 3. **Detached EBS volumes** — fully billed. Snapshot and delete. 4. **Old snapshots** — accumulate forever. Set lifecycle. 5. **Right-sizing instances** — t3.large where t3.small fits. 6. **Lambda log retention** — defaults to "Never." Set to 7-30 days. 7. **S3 storage classes** — Glacier for archive, Intelligent-Tiering for unknown. 8. **Data transfer** — cross-AZ traffic is billed. Same-AZ is free. ## 5. Reliability Patterns - **Multi-AZ for stateful services** — RDS, ElastiCache, OpenSearch - **Stateless app tier** — easy to lose, easy to replace - **Retries with exponential backoff + jitter** — not constant retry - **Circuit breakers** — fail fast when downstream is sick - **Dead letter queues** — for everything async - **Idempotency** — at-least-once delivery is the AWS norm ## 6. Security Defaults - Default-deny security groups - IAM roles, not access keys, for EC2/Lambda - Secrets Manager or SSM Parameter Store, never env vars in source - S3 buckets private by default - CloudTrail enabled in all regions - GuardDuty enabled - Encryption at rest on by default (KMS managed keys are fine) ## 7. Common Pitfalls | Problem | Cause | Fix | |---------|-------|-----| | Surprise $10k NAT bill | Heavy S3 traffic through NAT | VPC endpoint for S3 | | Lambda timeouts | Cold start + heavy init | Provisioned concurrency or smaller package | | DynamoDB throttling | Hot partition key | Choose better partition key, or use adaptive capacity | | CDK deploy fails on rollback | Resource locked by another stack | Find dependency, deploy in order | | Permissions denied at runtime | IAM role missing permission | Use `aws iam simulate-principal-policy` | ## 8. Key Questions Before Starting - What's the workload pattern — spiky, steady, or batched? - What's the SLA — 99%, 99.9%, 99.99%? - What's the team's comfort — serverless or container? - What's the budget? - Multi-region required? ## References - `references/cdk-patterns.md` — Working constructs - `references/cost-optimization-checklist.md` — Audit script - `references/well-architected-summary.md` — Five pillars distilled