aws-specialist
AWS cloud infrastructure, services, and architecture including CloudFormation, CDK, and the AWS Well-Architected Framework. Use when asked to deploy to AWS, set up EC2 or ECS or EKS, configure IAM rol
SKILL.md.claude/skills/aws-specialist/SKILL.md
GitHubAWS Specialist
Activation criteria
- User language explicitly matches trigger phrases such as
deploy to AWS,set up EC2,configure IAM. - The requested work fits this skill's lane: EC2, ECS, EKS, Lambda, RDS, S3, CloudWatch, IAM, VPC, cost optimization, AWS security.
- The task stays inside this skill's boundary and avoids adjacent areas called out as out of scope: Azure or GCP work; Terraform HCL writing (use terraform-specialist); application code.
First actions
Glob('**/*.tf', '**/*.cfn.yml', '**/*.cfn.json', '**/cdk.json', '**/samconfig.toml')— find existing IaCGlob('**/.aws/config', '**/aws-config*')— find AWS config files that reveal region and profile context- Read any existing CloudFormation templates or CDK stacks to understand current resource patterns
- Identify: target region, whether this is greenfield or existing infrastructure, AWS account structure (single vs multi-account)
Decision rules
- If IaC tooling is already established (Terraform files present): use terraform-specialist for provisioning; use this skill for AWS-specific service knowledge
- If asked about IAM: default to least-privilege; never recommend wildcard (
*) actions in production policies - If the task requires cross-account access: use IAM roles with trust policies, not access keys
- If cost optimization is involved: check for Reserved Instance/Savings Plan opportunities, right-sizing, and S3 lifecycle policies
- If multi-cloud scope appears: route Azure to azure-specialist, GCP to gcp-specialist
Steps
Step 1: Assess current AWS context
Understand what already exists: existing VPCs, IAM structure, compute patterns, and IaC approach in use.
Step 2: Design the solution
Apply Well-Architected Framework pillars: Operational Excellence, Security, Reliability, Performance Efficiency, Cost Optimization.
Step 3: Write the code
For CloudFormation: write YAML with proper Outputs and Parameters. For CDK: use TypeScript or Python constructs with L2+ where available. For CLI-based tasks: write bash scripts using aws CLI with proper error handling.
Step 4: Validate
For CloudFormation: aws cloudformation validate-template. For CDK: cdk synth. For IAM policies: use IAM policy simulator logic to check permissions.
Output contract
- Primary artifact: CloudFormation YAML, CDK stack file, or AWS CLI bash script
- Required: IAM least-privilege analysis for any IAM resources created; cost estimate note for significant resources
- Security: all S3 buckets must have
BlockPublicAcls: trueby default; all IAM roles must have scoped conditions
Constraints
- NEVER create IAM users with long-lived access keys for workloads — use IAM roles
- NEVER use
Effect: Allow, Action: *, Resource: *in production - NEVER hardcode AWS credentials in any file
- Scope boundary: application code belongs to language-specific skills; Terraform module writing belongs to terraform-specialist
Examples
Example 1: Lambda + API Gateway setup
User says: "Set up a Lambda function with an API Gateway endpoint for a Python REST API" Actions:
- Glob for existing CloudFormation/SAM templates
- Write SAM template with Lambda function, API Gateway, and IAM execution role
- Include environment variables, memory/timeout configuration, and CloudWatch log group Result: Complete SAM template with deployment instructions
Example 2: AWS account scan
User says: "Scan my AWS account for security issues" Actions:
- Write Python script using boto3 to check: public S3 buckets, overly permissive security groups, IAM users without MFA, unencrypted EBS volumes, CloudTrail status
- Output findings as JSON grouped by severity Result: Security audit script with prioritized findings
Troubleshooting
"Access Denied" on AWS API calls
Cause: IAM permissions missing or wrong region
Fix: Check effective permissions with aws iam simulate-principal-policy; verify --region flag matches resource location
CloudFormation stack stuck in UPDATE_ROLLBACK_FAILED
Cause: Manual resource changes outside CloudFormation
Fix: Use aws cloudformation continue-update-rollback --skip-resources to skip conflicting resources
Reference
references/legacy-agent.md: full service reference — EC2, ECS, EKS, Lambda, RDS, S3, CloudWatch, IAM, VPC, cost optimization patterns, MCP code examples (note: MCP stubs are non-functional)