Purpose

Monitor, analyze, and reduce cloud infrastructure spending — implement cost allocation tagging, configure budget alerts and anomaly detection, right-size compute and storage resources, evaluate reserved instance and spot/preemptible pricing, detect idle and orphaned resources, and build cost visibility dashboards.

When to use this skill

• Analyzing a cloud bill to identify the top cost drivers by service, team, or environment.
• Setting up cost allocation tags across AWS, GCP, or Azure resources.
• Configuring budget alerts and spending thresholds with automated notifications.
• Right-sizing compute instances based on utilization metrics (CPU, memory, network).
• Evaluating reserved instances, savings plans, or committed use discounts vs on-demand pricing.
• Identifying idle resources: unattached EBS volumes, unused Elastic IPs, idle load balancers, stopped instances with attached storage.
• Setting up spot/preemptible instance strategies for fault-tolerant workloads.
• Building cost dashboards in AWS Cost Explorer, GCP Billing, or third-party tools (Infracost, Vantage).

Do not use this skill when

• The task is application performance tuning without a cost reduction objective — prefer performance-specific skills.
• The task is capacity planning focused on scaling up to meet demand rather than reducing spend.
• The focus is on a specific provider's CLI or console workflows — prefer

  aws

  or

  gcp

  for provider-specific commands.
• The task involves billing disputes or account-level commercial negotiations.

Operating procedure

1. Gather the current cost baseline. Pull the last 3 months of billing data from the provider's cost management tool (AWS Cost Explorer, GCP Billing Reports, Azure Cost Management). Export to CSV if a third-party tool is used.
2. Identify top cost drivers. Sort spending by service, then by resource tag (team, environment, project). Flag any single resource or service consuming >30% of total spend.
3. Audit cost allocation tags. List all active resources and check for the required tag set (e.g.,

   team

   ,

   environment

   ,

   project

   ,

   cost-center

   ). Flag untagged resources and generate a remediation list.
4. Apply missing tags. Use the provider CLI to tag untagged resources:

   aws resourcegroupstaggingapi tag-resources

   ,

   gcloud resource-manager tags bindings create

   , or Terraform

   default_tags

   .
5. Configure budget alerts. Create a monthly budget at the account or project level. Set alert thresholds at 50%, 80%, and 100% of the budget. Wire notifications to Slack, email, or PagerDuty. For AWS: use AWS Budgets. For GCP: use Budget Alerts in Billing.
6. Enable anomaly detection. Turn on AWS Cost Anomaly Detection or GCP billing anomaly alerts. Set the sensitivity to flag daily spend increases >20% above the trailing 7-day average.
7. Right-size compute instances. Pull CPU and memory utilization metrics for the last 14 days. Flag instances with average CPU <20% and memory <30% as over-provisioned. Recommend downsizing by one instance family step (e.g., m5.xlarge → m5.large).
8. Evaluate reserved vs spot pricing. For steady-state workloads running >80% of the month, calculate savings from 1-year no-upfront reserved instances. For fault-tolerant batch workloads, calculate savings from spot/preemptible instances with interruption handling.
9. Detect idle and orphaned resources. Scan for: unattached EBS volumes, unused Elastic IPs, load balancers with zero targets, stopped instances with attached storage, unused NAT gateways, and empty S3 buckets with lifecycle policies disabled.
10. Build the cost dashboard. Create a dashboard showing: monthly spend trend, spend by service, spend by team/environment tag, top 10 costliest resources, and budget vs actual.
11. Document recommendations. Produce a prioritized list of cost savings actions ranked by estimated monthly savings. Include the specific resource, current cost, recommended action, and projected savings.

Decision rules

• Prioritize savings actions by dollar impact — address the largest waste first.
• Only recommend reserved instances for workloads with >80% steady-state utilization over 3+ months.
• Use spot instances only for workloads that can tolerate interruption (batch, CI runners, stateless workers).
• If a resource is untagged and the owner cannot be identified, flag it for review before terminating.
• Never delete resources without confirming they have no active consumers — check network connections, DNS references, and dependent services.
• Treat cost allocation tagging as a prerequisite for all other optimization work — without tags, attribution is impossible.
• Set budget alerts before optimizing — you need visibility into whether changes are working.

Output requirements

1. Cost baseline — current monthly spend, top 5 services, top 5 resources by cost.
2. Tag audit report — count of tagged vs untagged resources, remediation list.
3. Budget alert configuration — thresholds, notification channels, alert policy names.
4. Right-sizing recommendations — instance ID, current size, utilization metrics, recommended size, estimated savings.
5. Idle resource report — resource type, ID, age, last-used date, estimated monthly waste.
6. Savings action plan — prioritized list with estimated monthly savings per action.

References

• AWS Cost Explorer: https://docs.aws.amazon.com/cost-management/latest/userguide/ce-what-is.html
• GCP Billing Reports: https://cloud.google.com/billing/docs/how-to/reports
• AWS Cost Anomaly Detection: https://docs.aws.amazon.com/cost-management/latest/userguide/manage-ad.html
• FinOps Foundation principles: https://www.finops.org/framework/principles/
• Infracost (IaC cost estimation): https://www.infracost.io/docs/

• references/preflight-checklist.md

Related skills

• aws

  — AWS-specific billing tools, Cost Explorer, Budgets API.

• gcp

  — GCP Billing, committed use discounts, recommender API.

• cloud-deploy

  — deployment decisions that affect cost (instance types, regions, scaling).

Anti-patterns

• Optimizing costs without a tagging strategy — savings cannot be attributed to teams.
• Purchasing reserved instances for workloads with unpredictable or declining usage.
• Using spot instances for stateful services without data persistence strategies.
• Setting a single budget alert at 100% — by the time it fires, the overspend has already happened.
• Terminating idle resources without checking for scheduled or seasonal usage patterns.
• Focusing on small savings (pennies/month) while ignoring large waste (hundreds/month).

Failure handling

• If billing data is inaccessible, verify IAM permissions for the cost management APIs (e.g.,

  ce:GetCostAndUsage

  for AWS,

  billing.viewer

  for GCP).
• If cost allocation tags are not appearing in billing reports, check that tag activation is enabled (AWS: activate user-defined tags in Billing; GCP: export labels to BigQuery billing export).
• If budget alerts are not firing, verify the notification channel is configured and test with a threshold below current spend.
• If right-sizing recommendations conflict with performance requirements, defer to the application team and document the trade-off.
• If the provider is unknown or multi-cloud, produce provider-agnostic recommendations and flag items that need provider-specific follow-up.