aws-cost-cleanup – skill DevOps & Hạ tầng cho agent AI

AWS Cost Cleanup

Automate the identification and removal of unused AWS resources to eliminate waste.

When to Use This Skill

Use this skill when you need to automatically clean up unused AWS resources to reduce costs and eliminate waste.

Automated Cleanup Targets

Storage

• Unattached EBS volumes
• Old EBS snapshots (>90 days)
• Incomplete multipart S3 uploads
• Old S3 versions in versioned buckets

Compute

• Stopped EC2 instances (>30 days)
• Unused AMIs and associated snapshots
• Unused Elastic IPs

Networking

• Unused Elastic Load Balancers
• Unused NAT Gateways
• Orphaned ENIs

Cleanup Scripts

Safe Cleanup (Dry-Run First)

#!/bin/bash
# cleanup-unused-ebs.sh

echo "Finding unattached EBS volumes..."
VOLUMES=$(aws ec2 describe-volumes \
  --filters Name=status,Values=available \
  --query 'Volumes[*].VolumeId' \
  --output text)

for vol in $VOLUMES; do
  echo "Would delete: $vol"
  # Uncomment to actually delete:
  # aws ec2 delete-volume --volume-id $vol
done

#!/bin/bash
# cleanup-old-snapshots.sh

CUTOFF_DATE=$(date -d '90 days ago' --iso-8601)

aws ec2 describe-snapshots --owner-ids self \
  --query "Snapshots[?StartTime<='$CUTOFF_DATE'].[SnapshotId,StartTime,VolumeSize]" \
  --output text | while read snap_id start_time size; do
  
  echo "Snapshot: $snap_id (Created: $start_time, Size: ${size}GB)"
  # Uncomment to delete:
  # aws ec2 delete-snapshot --snapshot-id $snap_id
done

#!/bin/bash
# release-unused-eips.sh

aws ec2 describe-addresses \
  --query 'Addresses[?AssociationId==null].[AllocationId,PublicIp]' \
  --output text | while read alloc_id public_ip; do
  
  echo "Would release: $public_ip ($alloc_id)"
  # Uncomment to release:
  # aws ec2 release-address --allocation-id $alloc_id
done

S3 Lifecycle Automation

# Prepare a lifecycle proposal. This replaces bucket lifecycle configuration;
# retrieve/merge existing rules and scope the prefix before any authorized apply.
cat > lifecycle-policy.json <<EOF
{
  "Rules": [
    {
      "Id": "Archive old objects",
      "Status": "Enabled",
      "Transitions": [
        {
          "Days": 90,
          "StorageClass": "STANDARD_IA"
        },
        {
          "Days": 180,
          "StorageClass": "GLACIER"
        }
      ],
      "NoncurrentVersionExpiration": {
        "NoncurrentDays": 30
      },
      "AbortIncompleteMultipartUpload": {
        "DaysAfterInitiation": 7
      }
    }
  ]
}
EOF

# After explicit review only, apply the merged policy to the confirmed bucket.

Cost Impact Calculator

#!/usr/bin/env python3
# calculate-savings.py

import boto3
from datetime import datetime, timedelta

ec2 = boto3.client('ec2')

# Calculate EBS volume savings
volumes = ec2.describe_volumes(
    Filters=[{'Name': 'status', 'Values': ['available']}]
)

total_size = sum(v['Size'] for v in volumes['Volumes'])
monthly_cost = total_size * 0.10  # illustrative input, not a verified gp3 rate

print(f"Unattached EBS Volumes: {len(volumes['Volumes'])}")
print(f"Total Size: {total_size} GB")
print(f"Monthly Savings: ${monthly_cost:.2f}")

# Calculate Elastic IP savings
addresses = ec2.describe_addresses()
unused = [a for a in addresses['Addresses'] if 'AssociationId' not in a]

eip_cost = len(unused) * 3.65  # $0.005/hour * 730 hours
print(f"\nUnused Elastic IPs: {len(unused)}")
print(f"Monthly Savings: ${eip_cost:.2f}")

print(f"\nTotal Monthly Savings: ${monthly_cost + eip_cost:.2f}")
print(f"Annual Savings: ${(monthly_cost + eip_cost) * 12:.2f}")

Automated Cleanup Lambda

import boto3
from datetime import datetime, timedelta

def lambda_handler(event, context):
    ec2 = boto3.client('ec2')
    
    # Discover candidates only; creation age is not detachment age.
    volumes = ec2.describe_volumes(
        Filters=[{'Name': 'status', 'Values': ['available']}]
    )
    
    cutoff = datetime.now() - timedelta(days=7)
    deleted = 0
    
    for vol in volumes['Volumes']:
        create_time = vol['CreateTime'].replace(tzinfo=None)
        if create_time < cutoff:
            try:
                print(f"Candidate only (not deleted): {vol['VolumeId']}")
            except Exception as e:
                print(f"Error deleting {vol['VolumeId']}: {e}")
    
    return {
        'statusCode': 200,
        'body': 'Dry run only; no volumes deleted'
    }

Cleanup Workflow

1. Discovery Phase (Read-only)

• Run all describe commands
• Generate cost impact report
• Review with team

2. Validation Phase

• Verify resources are truly unused
• Check for dependencies
• Notify resource owners

3. Execution Phase (Dry-run first)

• Run cleanup scripts with dry-run
• Review proposed changes
• Execute actual cleanup

4. Verification Phase

• Confirm deletions
• Monitor for issues
• Document savings

Safety Checklist

• [ ] Run in dry-run mode first
• [ ] Verify resources have no dependencies
• [ ] Check resource tags for ownership
• [ ] Notify stakeholders before deletion
• [ ] Create snapshots of critical data
• [ ] Test in non-production first
• [ ] Have rollback plan ready
• [ ] Document all deletions

Example Prompts

Discovery

• "Find all unused resources and calculate potential savings"
• "Generate a cleanup report for my AWS account"
• "What resources can I safely delete?"

Execution

• "Create a script to cleanup unattached EBS volumes"
• "Delete all snapshots older than 90 days"
• "Release unused Elastic IPs"

Automation

• "Set up automated cleanup for old snapshots"
• "Create a Lambda function for weekly cleanup"
• "Schedule monthly resource cleanup"

Integration with AWS Organizations

# Run cleanup across multiple accounts
for account in $(aws organizations list-accounts \
  --query 'Accounts[*].Id' --output text); do
  
  echo "Checking account: $account"
  aws ec2 describe-volumes \
    --filters Name=status,Values=available \
    --profile account-$account
done

Monitoring and Alerts

# Create CloudWatch alarm for cost anomalies
aws cloudwatch put-metric-alarm \
  --alarm-name high-cost-alert \
  --alarm-description "Alert when daily cost exceeds threshold" \
  --metric-name EstimatedCharges \
  --namespace AWS/Billing \
  --statistic Maximum \
  --period 86400 \
  --evaluation-periods 1 \
  --threshold 100 \
  --comparison-operator GreaterThanThreshold

Best Practices

• Schedule cleanup during maintenance windows
• Always create final snapshots before deletion
• Use resource tags to identify cleanup candidates
• Implement approval workflow for production
• Log all cleanup actions for audit
• Set up cost anomaly detection
• Review cleanup results weekly

Risk Mitigation

Medium Risk Actions:

• Deleting unattached volumes (ensure no planned reattachment)
• Removing old snapshots (verify no compliance requirements)
• Releasing Elastic IPs (check DNS records)

Always:

• Maintain 30-day backup retention
• Use AWS Backup for critical resources
• Test restore procedures
• Document cleanup decisions

Kiro CLI Integration

# Analyze and cleanup in one command
kiro-cli chat "Use aws-cost-cleanup to find and remove unused resources"

# Generate cleanup script
kiro-cli chat "Create a safe cleanup script for my AWS account"

# Schedule automated cleanup
kiro-cli chat "Set up weekly automated cleanup using aws-cost-cleanup"

Additional Resources

• [AWS Resource Cleanup Best Practices](https://aws.amazon.com/blogs/mt/automate-resource-cleanup/)
• [AWS Systems Manager Automation](https://docs.aws.amazon.com/systems-manager/latest/userguide/systems-manager-automation.html)
• [AWS Config Rules for Compliance](https://docs.aws.amazon.com/config/latest/developerguide/managed-rules-by-aws-config.html)

Limitations

• Use this skill only when the task clearly matches the scope described above.
• Do not treat the output as a substitute for environment-specific validation, testing, or expert review.
• Stop and ask for clarification if required inputs, permissions, safety boundaries, or success criteria are missing.

An unattached/stopped/old resource can still be a required recovery asset. Establish account/region, ownership, explicit candidate IDs, dependency checks, retention requirements and a tested backup/restore plan before deletion. Creation time does not show when a volume became unused. Keep dry-run as default; a separate approved execution must recheck state/tags and scope IAM. Paginate discovery across intended regions, use current region/type/IOPS/throughput/IPv4 pricing and distinguish estimated avoided cost from realized billed savings. GNU date -d is not portable to macOS.