cloudthinker-ai/CloudSkills / skills/connections/managing-coreweave
managing-coreweave
UnverifiedUse when working with Coreweave — coreWeave GPU cloud management covering Kubernetes namespace inventory, GPU workload status, virtual server instances, persistent volume claims, node allocation, billing analysis, and network configuration. Use for comprehensive CoreWeave infrastructure assessment and GPU workload optimization.
CoreWeave Management
Analyze CoreWeave GPU workloads, virtual servers, storage, and Kubernetes resources.
Phase 1: Discovery
#!/bin/bash
# CoreWeave uses standard Kubernetes API with their kubeconfig
export KUBECONFIG="${COREWEAVE_KUBECONFIG:-$HOME/.kube/coreweave}"
echo "=== Namespaces ==="
kubectl get namespaces -o json \
| jq -r '.items[] | "\(.metadata.name)\t\(.status.phase)\t\(.metadata.creationTimestamp[0:10])"' \
| column -t | head -15
echo ""
echo "=== GPU Workloads ==="
kubectl get pods --all-namespaces -o json \
| jq -r '.items[] | select(.spec.containers[].resources.limits["nvidia.com/gpu"] != null) | "\(.metadata.namespace)\t\(.metadata.name[0:40])\t\(.status.phase)\t\(.spec.containers[0].resources.limits["nvidia.com/gpu"]) GPU(s)\t\(.spec.nodeName // "pending")"' \
| column -t | head -20
echo ""
echo "=== Virtual Servers ==="
kubectl get virtualservers --all-namespaces -o json 2>/dev/null \
| jq -r '.items[]? | "\(.metadata.namespace)\t\(.metadata.name)\t\(.status.state // "unknown")\t\(.spec.resources.gpu.type // "N/A")\t\(.spec.resources.gpu.count // 0) GPU(s)"' \
| column -t | head -15
echo ""
echo "=== Persistent Volume Claims ==="
kubectl get pvc --all-namespaces -o json \
| jq -r '.items[] | "\(.metadata.namespace)\t\(.metadata.name[0:30])\t\(.status.phase)\t\(.spec.resources.requests.storage)\t\(.spec.storageClassName)"' \
| column -t | head -20
Phase 2: Analysis
#!/bin/bash
export KUBECONFIG="${COREWEAVE_KUBECONFIG:-$HOME/.kube/coreweave}"
echo "=== Node GPU Summary ==="
kubectl get nodes -o json \
| jq -r '.items[] | select(.status.capacity["nvidia.com/gpu"] != null) | "\(.metadata.name[0:30])\t\(.metadata.labels["gpu.nvidia.com/class"] // "unknown")\tGPUs:\(.status.capacity["nvidia.com/gpu"])\tAllocatable:\(.status.allocatable["nvidia.com/gpu"])"' \
| column -t | head -20
echo ""
echo "=== Deployments ==="
kubectl get deployments --all-namespaces -o json \
| jq -r '.items[] | "\(.metadata.namespace)\t\(.metadata.name[0:30])\t\(.status.readyReplicas // 0)/\(.spec.replicas)\t\(.status.updatedReplicas // 0) updated"' \
| column -t | head -15
echo ""
echo "=== Services & Endpoints ==="
kubectl get services --all-namespaces -o json \
| jq -r '.items[] | select(.metadata.namespace != "kube-system") | "\(.metadata.namespace)\t\(.metadata.name[0:30])\t\(.spec.type)\t\(.spec.clusterIP)\t\(.status.loadBalancer.ingress[0].ip // "N/A")"' \
| column -t | head -15
echo ""
echo "=== Resource Requests Summary ==="
kubectl get pods --all-namespaces -o json \
| jq '{
total_gpu_requests: [.items[].spec.containers[].resources.requests["nvidia.com/gpu"] // "0" | tonumber] | add,
total_cpu_requests: [.items[].spec.containers[].resources.requests.cpu // "0" | gsub("m";"") | tonumber] | add,
total_memory_requests_gi: ([.items[].spec.containers[].resources.requests.memory // "0" | gsub("Gi";"") | gsub("Mi";"") | tonumber] | add / 1024 | . * 10 | round / 10),
pod_count: (.items | length)
}'
echo ""
echo "=== InferenceService (KServe) ==="
kubectl get inferenceservices --all-namespaces -o json 2>/dev/null \
| jq -r '.items[]? | "\(.metadata.namespace)\t\(.metadata.name[0:30])\t\(.status.conditions[-1].type // "unknown")\t\(.status.url // "N/A")"' \
| column -t | head -10
echo ""
echo "=== Recent Events ==="
kubectl get events --all-namespaces --sort-by='.lastTimestamp' -o json \
| jq -r '.items[-10:][] | "\(.metadata.namespace)\t\(.involvedObject.name[0:25])\t\(.type)\t\(.reason)\t\(.message[0:50])"' \
| column -t
Output Format
COREWEAVE ANALYSIS
====================
Namespace Workload GPUs Type Status Storage
──────────────────────────────────────────────────────────────────────────
ml-training train-job-large 8xA100 Pod Running 500Gi
inference llm-server 4xA40 VS Running 200Gi
dev experiment-1 1xRTX Pod Running 50Gi
GPUs: 13 allocated (A100:8, A40:4, RTX:1)
Pods: 12 running | Virtual Servers: 2 | PVCs: 8 (750Gi total)
Services: 5 (2 LoadBalancer) | Namespaces: 4 active
Safety Rules
- Read-only: Only use
kubectl get,describe, andlogscommands - Never create, delete, or scale workloads without confirmation
- Kubeconfig: Never output kubeconfig contents or tokens
- Secrets: Never read or output Kubernetes secret values
Anti-Hallucination Rules
- NEVER assume resource names — always discover via CLI/API in Phase 1 before referencing in Phase 2.
- NEVER fabricate metric names or dimensions — verify against the service documentation or
--helpoutput. - NEVER mix CLI commands between service versions — confirm which version/API you are targeting.
- ALWAYS use the discovery → verify → analyze chain — every resource referenced must have been discovered first.
- ALWAYS handle empty results gracefully — an empty response is valid data, not an error to retry.
Counter-Rationalizations
| Shortcut | Counter | Why |
|---|---|---|
| "I'll skip discovery and check known resources" | Always run Phase 1 discovery first | Resource names change, new resources appear — assumed names cause errors |
| "The user only asked for a quick check" | Follow the full discovery → analysis flow | Quick checks miss critical issues; structured analysis catches silent failures |
| "Default configuration is probably fine" | Audit configuration explicitly | Defaults often leave logging, security, and optimization features disabled |
| "Metrics aren't needed for this" | Always check relevant metrics when available | API/CLI responses show current state; metrics reveal trends and intermittent issues |
| "I don't have access to that" | Try the command and report the actual error | Assumed permission failures prevent useful investigation; actual errors are informative |
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deployment | Stars: 318k | Updated | Claude CodeCodex | Unverified | Install |
| xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |
| shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy. | Stars: 118k | Updated | Claude CodeCodex | Verified | Install |
| childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release. | Stars: 117k | Updated | Claude CodeCodex | Unverified | Install |