GKE Cost Optimization
Reduce GKE spending through rightsizing, quotas, Spot VMs, and capacity planning.
The document exposes Kubernetes read, apply, and patch capabilities and explicitly warns about Spot preemption; however, it does not define least-privilege permissions, user confirmation, sensitive-data handling, pre-change checks, rollback, or data flows, so points are deducted. No malware, credential theft, covert exfiltration, or other red-line behavior is shown.
The main paths include ResourceQuota, VPA, ComputeClass, nodeSelector, and CUD examples and are statically understandable; however, there is no test or execution evidence, no coverage of permission failures, invalid placeholders, version differences, resource conflicts, or rollback, and some tables are malformed. Static calibration therefore limits this to 8.
The name, use cases, and exclusions for general compute provisioning and GPU selection are reasonably clear, and Spot suitability boundaries are stated; however, required inputs, output format, cluster-version prerequisites, Chinese-language support, and mainland-China network reachability are unspecified, so points are deducted.
The skill has readable sections, examples, warnings, an Apache-2.0 license, an installation entry point, and official-organization provenance; however, it lacks skill-specific versioning, changelog, named maintenance responsibility, update path, FAQ, and systematic limitation/troubleshooting documentation, so points are deducted.
It provides actionable optimization guidance and Kubernetes snippets sufficient to cover the core task; however, monitoring depends on another skill, examples contain unresolved placeholders, discount and machine-type claims are not environment-verified, and changes still require substantial human review. Static calibration limits this to 7.
The evidence consists mainly of revision-pinned skill documentation, README material, and the license, providing limited auditability; however, there are no third-party citations, committed test suites, CI coverage, or reproducible experiment results, so key cost and configuration claims cannot be independently verified from the supplied files. Static calibration limits this to 4.
- Before using kubectl or other mutation tools, show the target cluster, namespace, permissions, and change plan and obtain user confirmation; rollback procedures are absent.
- Spot, VPA, quotas, CUDs, and cluster consolidation can affect availability, scheduling, and billing; validate workload fault tolerance, baseline costs, version compatibility, and monitoring data before applying the examples.
- The files do not address mainland-China reachability or Chinese-language support; workflows depending on Google Cloud Console, BigQuery, or related services may require additional network access.
What does this skill do, and when should you use it?
This skill is for teams reducing Google Kubernetes Engine costs while preserving a secure and reliable posture. It covers workload rightsizing, resource quotas, Spot VM placement, machine-type selection, committed use discounts, and cluster consolidation. It also documents cost-oriented GKE settings such as utilization-optimized autoscaling and vertical pod autoscaling. It is not intended for general compute-class provisioning or GPU selection.
Provides workflows and MCP operations for Kubernetes resources and clusters through get_k8s_resource, describe_k8s_resource, apply_k8s_manifest, patch_k8s_resource, and get_cluster. It shows how to apply ResourceQuota manifests, deploy VPA in recommendation mode, read recommendations with kubectl, configure ComputeClass or nodeSelector for Spot VMs, select machine families, use CUDs, and manage autoscaling and multi-tenancy. Cost-allocation setup, live utilization inspection, and detailed billing queries are delegated to the gke-cost-analysis skill.
- A GKE platform team wants to reduce oversized CPU and memory requests using VPA recommendations and P95-based rules.
- A multi-tenant cluster administrator needs namespace quotas to limit consumption and prevent runaway costs.
- A team running batch, data-processing, or development workloads wants to use interruptible Spot VM capacity.
- A Google Cloud organization with predictable baseline usage wants guidance on one-year or three-year CUDs.
- A platform engineering team wants to reduce idle capacity through autoscaling, node auto-provisioning, and shared clusters.
What are this skill's strengths and limitations?
- Covers major GKE cost levers: quotas, rightsizing, Spot VMs, machine types, CUDs, and cluster governance.
- Includes copyable kubectl and YAML examples, with explicit Spot preemption and replica guidance.
- Documents production versus development/test cost-oriented configuration differences.
- Distributed under Apache 2.0, in an actively developed repository.
- Requires Kubernetes, Google Cloud CLI, BigQuery CLI, and MCP tools; setup and authentication are not documented here.
- Spot VMs can be preempted at any time with 30 seconds’ notice and are unsuitable for databases or single-replica critical services.
- Cost analysis and detailed utilization or billing queries depend on the separate gke-cost-analysis skill.
- The source provides no test suite, performance benchmark, or verified savings results.
How do you install this skill?
This skill is part of the google/skills collection. Run npx skills add google/skills, then select the GKE Cost Optimization skill during installation. The source does not document a client-specific installation directory or standalone install command.
How do you use this skill?
Install skills/cloud/gke-cost-optimization/ and its SKILL.md in an Agent Skills-compatible client, then use a request such as “Optimize my GKE costs; review quotas, VPA, Spot VMs, and machine types.” Kubernetes, Google Cloud, and the relevant MCP tool access must be available for live operations. The source does not specify a client-specific invocation syntax.
How does this skill compare with similar options?
Unlike gke-cost-analysis, this skill focuses on applying cost-optimization configuration; gke-cost-analysis handles cost-allocation enablement, utilization inspection, and historical billing analysis. Unlike gke-compute-classes, it covers cost-oriented machine-type and Spot VM choices rather than general compute-class provisioning or GPU selection.