Build and inspect Nebius cloud infrastructure using Python SDK patterns and scoped inspector scripts.
npx skills add https://github.com/nebius/nebius-ps-services --skill nebius-nebius
Please help me install this Agent Skill. Skill: nebius Source: https://github.com/nebius/nebius-ps-services/tree/main/skills/nebius Command: npx skills add https://github.com/nebius/nebius-ps-services --skill nebius-nebius
scripts/inspect_*.py commands report compute, Kubernetes, storage, VPC topology/routes, and quota state with a common {scope, data, complete, errors} JSON envelope and explicit exit codes. - Reusable SDK assets: Importable assets/ modules for SDK runtime, request builders, IAM workflows, and bucket management with bounded pagination, operation waiting, and reconciliation-safe error handling. - Category guidance: References covering compute, storage, networking, observability, IAM, API/SDK usage, MK8s GPU setup, quotas, and active diagnostics, each with official source links and review dates. - Use Case: Ask the agent to inspect a project's quota headroom and VPC route tables before provisioning a GPU node group; it runs the scoped inspectors, reports incomplete collections honestly, and only mutates resources after explicit authorization. ## Quick Start Ask the agent to inspect the Nebius compute and quota state for your project ID using the nebius skill, then plan an authorized GPU node group deployment.High-intent search queries and answers about installing and using this skill.
Run the bundled inspector scripts such as inspect_compute.py, inspect_vpc_topology.py, or inspect_quotas.py with an explicit --project-id and optional --profile. Each emits a JSON envelope with scope, data, complete flag, and sanitized errors, exiting 0 only on complete collection.
Use inspect_quotas.py with a tenant or project ID to get effective available quota per name and region, merging tenant and project allowances. The tool treats unknown usage as unresolved rather than zero and never equates quota with actual capacity.
Yes, references cover MK8s compatibility matrices, drivers_preset selection, GPU Operator and Network Operator ownership, and GPUDirect RDMA validation. It requires live compatibility and preset checks before configuring InfiniBand fabrics or proof pods.
No mutation happens implicitly; inspectors are read-only and asset builders only construct typed requests. Writes require explicit user authorization, submit once with retained operation IDs, and reconcile ambiguous outcomes before any retry.
Python 3.10 or newer is required, with the offline validation baseline pinned to Python 3.12 and nebius==0.6.7. Helpers never install dependencies, log in, or repair credentials on their own.
Use Terraform when you need declarative resource lifecycle and state management, since Terraform owns apply and state for declarative resources. The SDK assets are callable examples for imperative inspection and authorized one-off operations, not an orchestrator.
FenixAllianceManage Alliance Business Suite accounting entities via the absuite CLI.
ricableDeploy Python machine learning workloads on serverless GPU infrastructure via Modal.
NVIDIADiagnose NVIDIA Network Operator and Kubernetes networking failures from sosreport files.
kaynetikImplement infrastructure as code with Terraform, Kubernetes, and GitOps workflows.