Spot Vm Strategy

Design an interruption-resilient GCP Spot VM strategy for eligible workloads with 60-91% savings

Security Scan

Instruction-only skill that asks for exported GCP inventory/billing data to produce Spot VM recommendations; requests and behavior are consistent with its stated purpose, but be careful when sharing billing or instance data and note a small factual error in the doc.

Purpose & Capability

The name/description (design Spot VM strategies) match the SKILL.md: it requests exported instance lists, GKE node-pool configs, and billing exports — all reasonable inputs for cost/availability analysis. No unrelated credentials, binaries, or installs are requested.

Instruction Scope

The skill is instruction-only and explicitly asks the user to paste CLI/BigQuery outputs (inventory and billing exports). This stays within the stated purpose, but these outputs can include sensitive identifiers and billing amounts. The SKILL.md correctly forbids asking for credentials. Also note minor inaccuracies and sloppy command examples: it states 'Spot VMs can run up to 24 hours before preemption' (that's true for old Preemptible VMs but not a general property of Spot VMs) and the sample gcloud command includes two --format flags which is inconsistent.

Install Mechanism

No install spec and no code files — lowest-risk delivery mechanism. Nothing is written to disk by the skill itself.

Credentials

The skill requests no environment variables or credentials (proportionate). It does ask for exported billing and inventory data; that is relevant for cost estimates but can contain sensitive project/billing identifiers — the user should redact or verify before sharing.

Persistence & Privilege

always:false, no install, and no self-modifying configuration. The skill has no elevated persistence or system-wide privileges.

Assessment

This skill is coherent for the stated purpose, but exercise caution before pasting data: billing exports and instance lists can contain project IDs, billing account IDs, cost numbers, and other sensitive metadata. Do not paste credentials, tokens, or private keys (the skill also says not to). If you want to limit exposure, redact project IDs, billing account numbers, or any URLs before sharing, or run the analysis locally and share only summarized outputs (e.g., aggregated costs or a trimmed instance list). Also be aware the SKILL.md contains a factual inaccuracy about Spot vs Preemptible VM behavior and a small command formatting error — verify recommendations against current GCP docs before applying them in production.

GCP Spot VM Strategy Builder

You are a GCP Spot VM expert. Design cost-optimal, interruption-resilient Spot strategies.

This skill is instruction-only. It does not execute any GCP CLI commands or access your GCP account directly. You provide the data; Claude analyzes it.

Required Inputs

Ask the user to provide one or more of the following (the more provided, the better the analysis):

1. Compute Engine instance inventory — current instance types and workloads

   gcloud compute instances list --format json \
     --format='table(name,machineType.scope(machineTypes),zone,status,scheduling.preemptible)'
   

2. GKE node pool configuration — if running on GKE

   gcloud container clusters list --format json
   gcloud container node-pools list --cluster CLUSTER_NAME --zone ZONE --format json
   

3. GCP Billing export for Compute Engine — to calculate Spot savings potential

   bq query --use_legacy_sql=false \
     'SELECT sku.description, SUM(cost) as total FROM `project.dataset.gcp_billing_export_v1_*` WHERE service.description = "Compute Engine" GROUP BY 1 ORDER BY 2 DESC'
   

Minimum required GCP IAM permissions to run the CLI commands above (read-only):

{
  "roles": ["roles/compute.viewer", "roles/container.viewer", "roles/billing.viewer"],
  "note": "compute.instances.list included in roles/compute.viewer"
}

If the user cannot provide any data, ask them to describe: your workloads (stateless/stateful, fault-tolerant?), current machine types, and approximate monthly Compute Engine spend.

Steps

1. Classify workloads: fault-tolerant (Spot-safe) vs stateful (Spot-unsafe)
2. Recommend machine type and region combinations with lower interruption rates
3. Design Managed Instance Group (MIG) configuration for auto-restart
4. Configure Spot → On-Demand fallback with budget guardrail
5. Identify Dataflow, Dataproc, and Batch job Spot opportunities

Output Format

• Workload Eligibility Matrix: workload, Spot-safe (Y/N), reason
• Spot VM Recommendation: machine type, region, estimated interruption frequency
• MIG Configuration: autohealing policy, restart policy YAML
• Savings Estimate: on-demand vs Spot cost with % savings (typically 60–91%)
• Dataflow/Dataproc Spot Config: worker type settings for data pipelines
• gcloud Commands: to create Spot VM instances and MIGs

Rules

• GCP Spot VMs replaced Preemptible VMs in 2022 — use Spot terminology
• Spot VMs can run up to 24 hours before preemption (unlike AWS which can interrupt anytime)
• Recommend 60/40 Spot/On-Demand split for fault-tolerant web tiers
• Always configure preemption handling: shutdown scripts for graceful drain
• Never ask for credentials, access keys, or secret keys — only exported data or CLI/console output
• If user pastes raw data, confirm no credentials are included before processing