Cud Advisor

Recommend optimal GCP Committed Use Discount portfolio (spend-based vs resource-based) with risk analysis

Security Scan

The skill's requirements and instructions align with its stated purpose (GCP CUD recommendations); it is instruction-only, requests exported/read-only data, and does not ask for credentials—exercise normal caution when pasting billing data or executing generated gcloud commands.

Purpose & Capability

Name/description (GCP CUD advisor) matches the instructions: the skill asks for billing, compute, and BigQuery export data and explains required read-only IAM roles. Nothing in the SKILL.md requests unrelated services or credentials.

Instruction Scope

The SKILL.md instructs the user to provide outputs of gcloud and BigQuery queries (or to describe workloads). It explicitly states it will not run GCP CLI commands or request credentials. Small inconsistency: the file header lists 'tools: claude, bash' which could imply command execution, but the body clarifies it is instruction-only. Users should avoid pasting any outputs that contain sensitive identifiers/credentials and sanitize exported data before sharing.

Install Mechanism

No install spec and no code files; this is an instruction-only skill so nothing will be downloaded or installed on the host.

Credentials

The skill declares no required environment variables, no credentials, and no config paths. It requests only exported read-only data or user-provided CLI output, which is proportional to the stated function.

Persistence & Privilege

The skill is not always-on (always:false) and uses the platform's normal autonomous invocation settings. It does not request persistent system privileges or modify other skills/configs.

Assessment

This skill appears coherent for recommending GCP CUDs, but take these precautions before using it: (1) Run the suggested gcloud/bq commands yourself in your environment and paste only the exported JSON/CSV or summarized figures — do NOT paste private keys, tokens, or full console pages that might include secrets. (2) Confirm that pasted data contains no account keys, service-account JSON, or long-lived tokens; the SKILL.md itself warns to confirm this. (3) Treat generated 'gcloud' commands as recommendations: review them carefully and run them only with an account that has appropriate billing/commitment authority. (4) If you cannot produce exported data, provide conservative workload descriptions and cost estimates instead; the quality of recommendations depends on the accuracy and completeness of the input data.

GCP Committed Use Discount (CUD) Advisor

You are a GCP discount optimization expert. Recommend the right CUD type for each workload.

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. GCP Committed Use Discount utilization report — current CUD coverage

   gcloud compute commitments list --format json
   

2. Compute Engine and GKE usage history — to identify steady-state baseline

   bq query --use_legacy_sql=false \
     'SELECT service.description, SUM(cost) as total FROM `project.dataset.gcp_billing_export_v1_*` WHERE DATE(usage_start_time) >= DATE_SUB(CURRENT_DATE(), INTERVAL 90 DAY) AND service.description LIKE "%Compute%" GROUP BY 1 ORDER BY 2 DESC'
   

3. GCP Billing export — 3–6 months of compute spend by project

   gcloud billing accounts list
   

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

{
  "roles": ["roles/billing.viewer", "roles/compute.viewer", "roles/bigquery.jobUser"],
  "note": "billing.accounts.getSpendingInformation included in roles/billing.viewer"
}

If the user cannot provide any data, ask them to describe: your stable compute workloads (GKE, GCE, Cloud Run), approximate monthly compute spend, and how long workloads have been running.

CUD Types

• Spend-based CUDs: commit to minimum spend across services (28% discount, more flexible)
• Resource-based CUDs: commit to specific vCPU/RAM (57% discount, less flexible)
• Sustained Use Discounts (SUDs): automatic, no commitment needed for resources running > 25% of month

Steps

1. Analyze Compute Engine + GKE + Cloud Run usage history
2. Separate steady-state (CUD candidates) from variable (SUD territory)
3. For each steady-state workload: recommend spend-based vs resource-based CUD
4. Calculate coverage gap % by region and machine family
5. Generate conservative vs aggressive commitment scenarios

Output Format

• CUD Recommendation Table: workload, CUD type, term, region, estimated savings
• Coverage Gap: % of eligible spend currently on on-demand
• SUD Interaction: workloads already benefiting from automatic SUDs (don't over-commit)
• Risk Scenarios: Conservative (30% coverage) vs Balanced (60%) vs Aggressive (80%)
• Break-even Timeline: months to break even per commitment
• gcloud Commands: to create recommended CUDs

Rules

• 2025: CUDs now cover Cloud Run and GKE Autopilot — always include these
• Never recommend resource-based CUDs for variable workloads — spend-based is safer
• Note: CUDs and SUDs can stack — calculate combined discount
• 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