GKE Cluster Creation Advisor
Plan, provision, and audit GKE clusters against production best practices.
The skill requires the project, location, and cluster name and instructs the agent to present the configuration and obtain confirmation before creation. It recommends private nodes, authorized networks, Secret Manager, and Workload Identity, showing some least-privilege and security awareness. Points are deducted because cluster creation has external infrastructure and cost effects but lacks rollback, deletion recovery, failed-resource cleanup, and complete data-flow disclosure; some examples use the broad cloud-platform OAuth scope; and authorized-network examples do not specify restrictive CIDRs.
The workflow covers discovery, input collection, creation, operation tracking, and post-creation verification, and names the MCP tools. Points are deducted because this is static documentation with no tests or reproducibility evidence; examples are not fully consistent with the stated best practices, with some omitting private-cluster, Workload Identity, Shielded Nodes, or complete networking settings; systematic handling of quota, permission, partial-creation, and abnormal-input failures is absent. The mode table also contains visibly corrupted formatting.
The description clearly targets GKE cluster creation, mode selection, GPU/AI scenarios, and production-readiness audits, while excluding application onboarding; it also requires asking for missing essential parameters. Points are deducted because non-fit boundaries, required permissions, existing-network constraints, and audit inputs/outputs are not sufficiently defined; Chinese-language support is not addressed, nor is reachability from mainland-China networks for gcloud, MCP, or dependent Google Cloud services.
The skill has front matter, a layered workflow, a mode table, best practices, templates, and repository-level Apache-2.0 licensing, maintenance links, and active-development context. Points are deducted because the skill itself lacks versioning, a changelog, named maintenance responsibility, and a concrete update path; referenced adjacent documents and tool contracts are not included in the supplied evidence; troubleshooting, parameter constraints, FAQs, and template-difference guidance are thin, and the mode table is malformed.
It supplies gcloud and MCP configuration skeletons for Autopilot, Standard, GPU, and Hypercompute scenarios, with confirmation before creation and read-back verification afterward, so it plausibly supports the core planning task. Points are deducted because static files cannot establish that the commands and JSON are directly runnable against the current API; several templates are inconsistent or incomplete and may require manual networking, identity, and permission work; and no verified representative output or audit-report format is provided.
The material provides concrete commands, MCP payloads, configuration fields, and references to adjacent files, offering some auditability; repository provenance, license, and revision are supplied in the review context. Points are deducted because there is no committed test suite, CI coverage, third-party execution record, or corroborating source; commands and best-practice claims were not executed in this static review, so conclusions rely mainly on one author-provided source.
- GKE creation provisions billable cloud resources; before confirmation, verify the project, region, quotas, network CIDRs, permissions, and expected cost.
- Do not apply the cloud-platform OAuth scope or unrestricted authorized-network examples directly to production; tighten permissions and ingress for the workload.
- The gcloud and MCP examples do not consistently contain equivalent security settings; compare and validate every parameter against the current API before execution.
- This assessment is static only; no commands, MCP calls, or cluster verification were executed.
What does this skill do, and when should you use it?
This skill plans and executes Google Kubernetes Engine cluster creation, configuration, and production-readiness audits. It defaults to the production Autopilot golden path and also provides Standard Regional, GPU inference, and AI Hypercompute templates. Its workflow covers context discovery, mode selection, networking, configuration approval, creation, operation tracking, and verification. It does not handle application onboarding or deployment configuration.
Inspects existing clusters and project context, gathers the project ID, location, cluster name, and environment type, and explains trade-offs between Autopilot and Standard modes. It produces a gcloud command or create_cluster JSON payload, then uses create_cluster or gcloud after user confirmation. It tracks progress with get_operation and verifies settings with get_cluster using readMask="*". Referenced MCP tools include list_clusters, create_cluster, get_cluster, list_operations, and get_operation.
- A cloud platform engineer needs a production GKE cluster but has not chosen between Autopilot and Standard.
- A production team wants private nodes, VPC-native networking, Workload Identity, and a release channel configured.
- An ML platform team needs an L4-based inference cluster and guidance on quota requirements.
- An infrastructure team requires custom kernel, node OS, or privileged workload support that Autopilot cannot provide.
- An operator needs to audit an existing GKE cluster against the recommended golden-path settings.
What are this skill's strengths and limitations?
- Provides explicit Autopilot, Standard Regional, GPU inference, and AI Hypercompute templates.
- Includes guidance for private clusters, VPC-native networking, Workload Identity, Shielded Nodes, autoscaling, and regional availability.
- Requires configuration review and confirmation before creation.
- Supports creation and verification through MCP tools or the gcloud CLI.
- Requires GKE, the gcloud CLI, kubectl, or the corresponding MCP tools; no test suite or platform validation record is provided.
- GPU and AI Hypercompute templates require specific quotas and can be expensive.
- Standard clusters require manual management of node pools, upgrades, and autoscaling.
- Networking and private-node choices are identified as Day-0 decisions that can be difficult to change later.
How do you install this skill?
Run npx skills add google/skills, then select GKE Cluster Creation from the interactive skill selection. The source does not document another installation method or a fixed installation directory.
How do you use this skill?
Give a concrete request to an Agent Skills-compatible client, for example: “Create a production GKE cluster named prod-cluster in us-central1 for project my-project.” The skill requests missing required details, proposes Autopilot by default, presents a gcloud command or JSON payload, and creates and verifies the cluster after confirmation.
How does this skill compare with similar options?
The skill directly compares Autopilot and Standard: Autopilot uses Google-managed nodes and pod-resource pricing, while Standard uses self-managed nodes and VM-based node pricing. It also distinguishes cluster creation from gke-app-onboarding, which is identified as the skill for application onboarding or deployment configuration.