[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7A7T024XHG6ME9F8TEJEPB7":3,"skill-same:01M7A7T024XHG6ME9F8TEJEPB7":147},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":43,"commitSha":44,"skillDirectory":45,"forksCount":46,"repositoryHtmlUrl":47,"repositoryHomepageUrl":6,"ownerHtmlUrl":48,"ownerBlogUrl":6,"ownerDescription":49,"updatedAt":50,"install":51,"resources":64},"01M7A7T024XHG6ME9F8TEJEPB7","eks-autoscaling-automator",null,"Designs and implements EKS cluster autoscaling automation. Compares and deploys Cluster Autoscaler vs Karpenter (provisioner node templates, consolidation policies, disruption budgets), configures node-group scaling policies, builds HPA (Horizontal Pod Autoscaler) with custom and external metrics, explains VPA (Vertical Pod Autoscaler) caveats and incompatibilities, deploys KEDA for event-driven scaling (Kafka, SQS, Prometheus), handles spot instance interruption (termination handler, graceful drain), configures overprovisioning pause-pods for fast scale-up, sets up pod disruption budgets to protect quorum, defines priority classes for workload preemption, and deploys descheduler for bin-packing optimization. Emits AUTOMATION_DEPLOYED with ready-to-apply Helm values + manifests or REVIEW_REQUIRED with the specific gap. Use when building EKS autoscaling, choosing Karpenter vs Cluster Autoscaler, or optimizing an existing scaling setup.","jhkchan\u002Faws-agent-skills","jhkchan","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F1083902?v=4","01M7A2SVQYP9GQTPTDXP5JKYQ4","Jacky Chan","01M7A2SVS117YWCTZABB75HZR1",4,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Feks-autoscaling-automator\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: eks-autoscaling-automator\ndescription: Designs and implements EKS cluster autoscaling automation. Compares and deploys Cluster Autoscaler vs Karpenter (provisioner node templates, consolidation policies, disruption budgets), configures node-group scaling policies, builds HPA (Horizontal Pod Autoscaler) with custom and external metrics, explains VPA (Vertical Pod Autoscaler) caveats and incompatibilities, deploys KEDA for event-driven scaling (Kafka, SQS, Prometheus), handles spot instance interruption (termination handler, graceful drain), configures overprovisioning pause-pods for fast scale-up, sets up pod disruption budgets to protect quorum, defines priority classes for workload preemption, and deploys descheduler for bin-packing optimization. Emits AUTOMATION_DEPLOYED with ready-to-apply Helm values + manifests or REVIEW_REQUIRED with the specific gap. Use when building EKS autoscaling, choosing Karpenter vs Cluster Autoscaler, or optimizing an existing scaling setup.\nlicense: Apache-2.0\ncompatibility: Agent runtime that reads SKILL.md (Claude Code, Cursor, Windsurf, Codex, Gemini). No AWS CLI or kubectl required for offline workflow design. Live deployment uses helm install\u002Fupgrade (Cluster Autoscaler, Karpenter, KEDA, descheduler), kubectl apply (HPA, VPA, PDB, priority classes, Karpenter NodePool\u002FNodeClaim), aws eks update-nodegroup-config (managed node group scaling) — AWS CLI v2, kubectl, helm v3.\nmetadata:\n  domain: aws-cloudops\n  complexity: high\n  requires_llm: 'true'\n  phase: '4'\n  supports_pipeline: 'true'\n  entry_point: 'false'\n  family: Compute\n  task_type: automate\n  skill_class: capability\n  lifecycle_status: active\n  verdict_shape: AUTOMATION_DEPLOYED | REVIEW_REQUIRED\n  when_to_use: Building EKS autoscaling (Karpenter or Cluster Autoscaler), configuring HPA\u002FVPA\u002FKEDA for workload scaling, handling spot instance interruptions, optimizing scale-up latency with overprovisioning, protecting workloads with PDBs and priority classes, or improving bin-packing with descheduler.\n  activation_triggers: EKS autoscaling, Karpenter setup, Cluster Autoscaler, horizontal pod autoscaler, KEDA event-driven scaling, spot interruption handling, overprovisioning pause pods, pod disruption budget, descheduler bin-packing, node group scaling policy\n  invocation_schema: 'Input: either (a) an EKS cluster description (version, node count, workload mix, scaling requirements, spot\u002Fon-demand ratio) plus observability metrics, OR (b) an autoscaling request (\"set up Karpenter with spot diversification\", \"configure HPA with custom metrics for the API deployment\"). Output: deterministic AUTOSCALING block — TOOLING\u002FPROVISIONING\u002FPOD_SCALING\u002FSPOT\u002FSAFETY\u002FVERDICT — where VERDICT is AUTOMATION_DEPLOYED (Helm values + manifests ready) or REVIEW_REQUIRED (specific gap cited).'\n  version: 0.1.0\n  author: Jacky Chan — AWS Community Builder\n  keywords: EKS autoscaling, Karpenter, Cluster Autoscaler, Horizontal Pod Autoscaler, HPA, Vertical Pod Autoscaler, VPA, KEDA, spot instance interruption, overprovisioning, pod disruption budget, priority class, descheduler, consolidation, disruption budget\n  tags: eks, karpenter, cluster-autoscaler, hpa, keda, spot, autoscaling, automate\n---\n\n# EKS Autoscaling Automator\n\n## Mindset\n\n**One-line takeaway:** every EKS autoscaling system operates at two\nlayers — **cluster autoscaling** (adding\u002Fremoving nodes: Karpenter or\nCluster Autoscaler) and **pod autoscaling** (adding\u002Fremoving pods: HPA\nor KEDA). A gap in either layer produces a silent failure: pods are\nPending because no node has capacity, or nodes sit idle because pod\ncount never scales with load.\n\n- **Cluster Autoscaler and Karpenter are mutually exclusive.** Running\n  both causes conflicting scaling decisions — CA provisions a node,\n  Karpenter de-provisions it, ad infinitum.\n- **HPA without custom metrics is just CPU autoscaling.** For request-\n  driven workloads (latency, queue depth), HPA needs custom\u002Fexternal\n  metrics via Prometheus Adapter or KEDA.\n- **Spot without interruption handling is an outage waiting to happen.**\n  Spot instances get a 2-minute warning. Without the termination handler\n  or Karpenter's native disruption, pods are killed ungracefully.\n\n## Quick navigation\n\n| You want to... | Go to |\n|---|---|\n| Choose Karpenter vs Cluster Autoscaler | Step 1 |\n| Deploy Karpenter with NodePool + disruption budget | Step 2 |\n| Deploy Cluster Autoscaler with auto-discovery | Step 3 |\n| Configure managed node group scaling | Step 4 |\n| Set up HPA with custom metrics | Step 5 |\n| Understand VPA caveats | Step 6 |\n| Deploy KEDA for event-driven scaling | Step 7 |\n| Handle spot interruptions | Step 8 |\n| Configure overprovisioning pause-pods | Step 9 |\n| Set up PDBs and priority classes | Step 10 |\n| Deploy descheduler for bin-packing | Step 11 |\n\n## Critical rules at a glance (do NOT bury these)\n\n1. **Karpenter consolidation replaces nodes aggressively.**\n   `WhenEmptyOrUnderutilized` will DELETE underutilized nodes and\n   reschedule pods. For production, use `disruptionBudgets` to cap how\n   many nodes Karpenter disrupts at once. Without a budget, Karpenter\n   can drain 50% of nodes simultaneously.\n\n2. **Spot diversification across instance families is mandatory.** A\n   NodePool requesting only `c5.large` spot will be evicted en masse\n   when c5 capacity disappears. Diversify across at least 3-4 families\n   (c5, m5, c6i, m6i).\n\n3. **Overprovisioning pause-pods create headroom for fast scale-up.**\n   Without them, new pods wait 2-5 minutes for a node to boot. Pause-\n   pods (low priority) occupy scheduled capacity; real workloads preempt\n   them and schedule immediately.\n\n4. **HPA and VPA are incompatible for the same CPU\u002FMemory dimension.**\n   HPA scales replicas based on CPU; VPA adjusts CPU requests. Running\n   both on CPU causes oscillation. Use VPA in `Off` mode alongside HPA,\n   or VPA for memory only.\n\n5. **PDBs protect against voluntary disruption only.** A PDB with\n   `minAvailable: 2` prevents `kubectl drain` from evicting below 2\n   replicas. It does NOT protect against spot termination (involuntary).\n   For spot, combine PDB + termination handler + topology spread.\n\n## Pre-flight: data requirements\n\n| Input | Source | Why |\n|---|---|---|\n| EKS cluster version | `eks describe-cluster` | Karpenter v1 requires 1.27+ |\n| Current node count\u002Ftype | `kubectl get nodes -o wide` | Baseline fleet |\n| Workload resource requests | `kubectl get pods -A -o json` | Drives node sizing |\n| Current HPA\u002FVPA\u002FKEDA | `kubectl get hpa,vpa -A` | Don't duplicate |\n| Spot vs on-demand ratio | Business requirements | Cost target |\n| Metrics server health | `kubectl top pods` | Required for default HPA |\n| Prometheus (if custom metrics) | `kubectl get svc -n monitoring` | For custom-metric HPA |\n\n**If the input is malformed**, emit:\n```text\nAUTOSCALING: \u003Creference>\nVERDICT: ERROR\nREASON: Cannot design autoscaling — cluster version, node count, and workload requests are required.\nGAP: Run kubectl get nodes -o wide and kubectl get pods -A -o json.\n```\n\n## Process — Autoscaling design (apply in order)\n\n### Step 0: Expert knowledge — non-obvious autoscaling behaviors\n\nStep 0 expert-knowledge deep dive (Karpenter v1 API migration, consolidation modes, CA scale-down blockers, KEDA trigger semantics, IMDS spot notices, pause-pod priority, VPA Auto restarts, descheduler PodLifeTime, MNG desiredSize override) moved verbatim to [references\u002Fadvanced-patterns.md](references\u002Fadvanced-patterns.md).\nLoad on demand before choosing or tuning any autoscaling component.\n\n### Step 1: Choose Karpenter vs Cluster Autoscaler\n\n| Dimension | Karpenter | Cluster Autoscaler |\n|---|---|---|\n| Scaling speed | ~30s (direct EC2 API) | ~2-5 min (ASG lifecycle) |\n| Node selection | Flexible (any type, AZ, spot+OD mix) | Rigid (ASG instance types) |\n| Consolidation | Yes (deletes underutilized nodes) | Limited (empty nodes only) |\n| Spot handling | Native disruption + diversified | Via ASG mixed policy |\n| Maturity | v1 GA (2024) | Battle-tested, stable |\n| Best for | Dynamic, spot-heavy, fast scale | Stable, predictable capacity |\n\nChoose **Karpenter** if: spot priority, \u003C 1 min scale-up, consolidation\nneeded, heterogeneous types. Choose **CA** if: fixed types in managed\nnode groups, simplicity priority.\n\n**NEVER run both.** Migrate: uninstall CA first, then install Karpenter.\n\n### Step 2: Deploy Karpenter with NodePool and disruption budgets\n\n```bash\nhelm install karpenter oci:\u002F\u002Fpublic.ecr.aws\u002Fkarpenter\u002Fkarpenter \\\n  --namespace kube-system \\\n  --set clusterName=my-prod-cluster \\\n  --set interruptionQueue=true \\\n  --version 1.0.0\n```\n\nNodePool with spot diversification + disruption budget:\n\n```yaml\napiVersion: karpenter.sh\u002Fv1\nkind: NodePool\nmetadata:\n  name: default\nspec:\n  template:\n    spec:\n      requirements:\n        - key: karpenter.k8s.aws\u002Finstance-category\n          operator: In\n          values: [\"c\", \"m\"]\n        - key: karpenter.k8s.aws\u002Finstance-generation\n          operator: Gt\n          values: [\"5\"]\n        - key: karpenter.sh\u002Fcapacity-type\n          operator: In\n          values: [\"spot\", \"on-demand\"]\n      nodeClassRef:\n        group: karpenter.k8s.aws\n        kind: EC2NodeClass\n        name: default\n  limits:\n    cpu: 1000\n  disruption:\n    consolidationPolicy: WhenEmptyOrUnderutilized\n    consolidateAfter: 30s\n    budgets:\n      - nodes: \"20%\"\n      - schedule: \"0 9 * * Mon-Fri\"\n        duration: 8h\n        nodes: \"0\"\n```\n\nFull EC2NodeClass manifest (AMI alias, subnet\u002FSG selectors, role, block devices) moved verbatim to [references\u002Fkarpenter-configuration.md](references\u002Fkarpenter-configuration.md).\nLoad on demand when authoring the EC2NodeClass that pairs with the NodePool above.\n\n**Key configuration:**\n- `disruption.budgets`: First budget allows 20% disruption normally.\n  Second budget blocks ALL disruption Mon-Fri 09:00-17:00. Critical for\n  production stability.\n- Spot diversification: `instance-category: [c, m]` + `generation: >5`\n  allows 8+ instance types. If c5 spot disappears, Karpenter tries m5\n  or c6i immediately.\n- `interruptionQueue: true`: enables native spot interruption handling\n  via SQS + EventBridge. Do NOT install NTH when this is enabled.\n\nSee **references\u002Fkarpenter-configuration.md** for full API reference,\nrequirement keys, and consolidation troubleshooting.\n\n### Step 3: Deploy Cluster Autoscaler with auto-discovery\n\n```bash\nhelm install cluster-autoscaler autoscaler\u002Fcluster-autoscaler \\\n  --namespace kube-system \\\n  --set autoDiscovery.clusterName=my-prod-cluster \\\n  --set awsRegion=us-east-1 \\\n  --version 9.34.0\n```\n\nAuto-discovery tags (must be on node group ASGs):\n```\nk8s.io\u002Fcluster-autoscaler\u002Fmy-prod-cluster = owned\nk8s.io\u002Fcluster-autoscaler\u002Fenabled = true\n```\n\nKey tuning: `scaleDownUnneededTime: \"30m\"`,\n`scaleDownUtilizationThreshold: \"0.5\"`,\n`skipNodesWithLocalStorage: true`, `expander: \"least-waste\"`.\n\n### Step 4: Configure managed node group scaling\n\n```bash\naws eks update-nodegroup-config \\\n  --cluster-name my-prod-cluster --nodegroup-name my-ng \\\n  --scaling-config desiredSize=5,minSize=3,maxSize=20\n```\n\nFor EKS Auto Mode (2025+, simplest path for new clusters):\n```bash\naws eks update-cluster-config --name my-prod-cluster \\\n  --compute-config '{\"nodeRoleArn\":\"arn:...\",\"enabled\":true}'\n```\n\n### Step 5: Set up HPA with custom metrics\n\nHPA with CPU + custom metric + behavior tuning:\n\nFull HPA manifest (CPU + custom metric + scaleUp\u002FscaleDown behavior policies) moved verbatim to [references\u002Fhpa-keda-descheduler.md](references\u002Fhpa-keda-descheduler.md).\nLoad on demand when authoring the HPA object for a deployment.\n\n**Behavior tuning guide:** scale UP fast (users waiting), scale DOWN\nslow (avoid thrashing). `scaleDown.stabilizationWindowSeconds: 300`\nminimum for production.\n\nSee **references\u002Fhpa-keda-descheduler.md** for custom metrics setup,\nKEDA triggers, and descheduler strategies.\n\n### Step 6: VPA caveats and incompatibilities\n\n| Mode | Behavior | Production-safe? |\n|---|---|---|\n| `Auto` | Recreates pods to apply requests | No (causes restarts) |\n| `Initial` | Sets requests only on NEW pods | Yes |\n| `Off` | Recommendations only | Yes |\n\n**HPA + VPA CPU conflict:** VPA raises CPU request → HPA sees lower\nutilization → HPA scales down → loop. Resolution: VPA `Off` mode (apply\nmanually) or VPA for memory only.\n\n### Step 7: Deploy KEDA for event-driven scaling\n\n```bash\nhelm install keda kedacore\u002Fkeda --namespace keda-system --create-namespace --version 2.15.0\n```\n\nSQS-triggered ScaledObject (scale-to-zero):\n\nSQS-triggered ScaledObject manifest (scale-to-zero, queueLength trigger, polling\u002Fcooldown) moved verbatim to [references\u002Fhpa-keda-descheduler.md](references\u002Fhpa-keda-descheduler.md).\nLoad on demand when authoring a KEDA ScaledObject.\n\n**KEDA vs HPA:** use KEDA when you need scale-to-zero, external-system\ntriggers, or multiple trigger sources. Use HPA when CPU\u002FMemory suffices.\n\n### Step 8: Handle spot instance interruptions\n\n**Karpenter approach (recommended):** Set `interruptionQueue: true`.\nKarpenter watches SQS fed by EventBridge, receives 2-min notice, cordons\n+ drains + provisions replacement. No additional components.\n\n**Cluster Autoscaler approach (NTH):**\n```bash\nhelm install aws-node-termination-handler eks\u002Faws-node-termination-handler \\\n  --namespace kube-system \\\n  --set enableSpotInterruptionDraining=true \\\n  --set enableQueueProcessing=true \\\n  --set awsRegion=us-east-1\n```\n\n**NEVER run both** — they conflict on the same SQS queue.\n\n**Spot diversification rules:**\n\n| Rule | Why |\n|---|---|\n| 3+ instance families | Reduces mass-reclamation risk |\n| 2+ instance sizes | Same |\n| 2+ AZs | Capacity shortages are AZ-specific |\n| `capacity-type: [spot, on-demand]` | Fallback to on-demand |\n\n### Step 9: Configure overprovisioning pause-pods\n\nPriority class (MUST be lower than default):\n\n```yaml\napiVersion: scheduling.k8s.io\u002Fv1\nkind: PriorityClass\nmetadata:\n  name: overprovisioning\nvalue: -1\ndescription: \"Pause-pod priority (lowest).\"\n```\n\nPause-pod Deployment:\n\n```yaml\napiVersion: apps\u002Fv1\nkind: Deployment\nmetadata:\n  name: pause-pods\n  namespace: kube-system\nspec:\n  replicas: 10\n  strategy: {type: Recreate}\n  selector: {matchLabels: {app: pause-pod}}\n  template:\n    metadata: {labels: {app: pause-pod}}\n    spec:\n      priorityClassName: overprovisioning\n      containers:\n        - name: pause\n          image: registry.k8s.io\u002Fpause:3.9\n          resources:\n            requests: {cpu: 1, memory: 1Gi}\n```\n\nEach pause-pod occupies 1 CPU + 1 GiB. For 10 pods headroom, Karpenter\nprovisions 10 pods worth of extra nodes. When a real deployment scales\nup, it preempts pause-pods and schedules immediately.\n\n### Step 10: Pod disruption budgets and priority classes\n\n```yaml\napiVersion: policy\u002Fv1\nkind: PodDisruptionBudget\nmetadata:\n  name: api-server-pdb\nspec:\n  minAvailable: 2\n  selector: {matchLabels: {app: api-server}}\n```\n\nPriority classes:\n\n```yaml\n# production-critical: value 1000000\n# production-standard: value 100000\n# batch-default: value 10000\n# default (globalDefault): value 0\n# overprovisioning: value -1\n```\n\nPreemption flow: when a node is full, the scheduler checks if lower-\npriority pods can be evicted to make room for a higher-priority pod.\n\n### Step 11: Deploy descheduler for bin-packing\n\n```bash\nhelm install descheduler descheduler\u002Fdescheduler \\\n  --namespace kube-system \\\n  --set kind=Deployment \\\n  --set schedule=\"0 *\u002F2 * * *\" \\\n  --set deschedulingInterval=2h \\\n  --version 0.31.0\n```\n\nKey safety settings: `maxNoOfPodsToEvictPerNode: 3`,\n`maxNoOfPodsToEvictTotal: 50`, `evictSystemCriticalPods: false`,\n`nodeFit: true` (CRITICAL — prevents evicting pods with nowhere to go).\n\nRun in DryRun mode first: `--set dryDescheduling=true`.\n\n## Output format\n\n```text\nAUTOSCALING: \u003Creference>\nCLUSTER: \u003CEKS version + node count>\nTOOLING:\n  - Cluster autoscaler: Karpenter \u003Cver> | CA \u003Cver> | EKS Auto Mode\n  - Pod autoscaler: HPA | KEDA | VPA (Off)\n  - Metrics: metrics-server | Prometheus Adapter\nPROVISIONING:\n  - NodePool: \u003Cname> \u002F Instance types: \u003Cdiversified list>\n  - Spot ratio: \u003C%> \u002F Disruption budget: \u003Cnodes per window>\nPOD_SCALING:\n  - HPA: \u003Cdeployment> min=N max=N metric=\u003Ctype>\n  - KEDA: \u003Cdeployment> trigger=\u003Ctype> min=N max=N\n  - VPA: \u003Cdeployment> mode=\u003COff|Initial>\nSPOT:\n  - Interruption handling: Karpenter native | NTH\n  - Diversification: \u003Cfamilies> \u002F Graceful drain: enabled|disabled\nSAFETY:\n  - Overprovisioning: \u003CN> x \u003Ccpu\u002Fmem>\n  - PDB: \u003Cdeployment> minAvailable=N\n  - Priority classes: \u003Clist> \u002F Descheduler: enabled|disabled\nVERDICT: AUTOMATION_DEPLOYED | REVIEW_REQUIRED\nGAP: \u003Cif REVIEW_REQUIRED, the specific missing piece>\nTEMPLATE: \u003CHelm values + kubectl manifests>\n```\n\n### Worked example — AUTOMATION_DEPLOYED\n\n```text\nAUTOSCALING: prod-cluster-autoscaling\nCLUSTER: EKS 1.30, 25 nodes\nTOOLING: Karpenter 1.0.0 (interruptionQueue) | HPA + KEDA | metrics-server + Prom Adapter\nPROVISIONING: default NodePool \u002F c5,m5,c6i,m6i \u002F 70% spot 30% OD \u002F 20% budget, freeze biz hrs\nPOD_SCALING: HPA api-server 3-50 (CPU+custom) | KEDA sqs-consumer 1-30 (SQS) | VPA Off\nSPOT: Karpenter native \u002F 4 families \u002F graceful drain\nSAFETY: 10 pause-pods x 1CPU\u002F1Gi \u002F PDB api-server minAvail=2 \u002F 4 priority classes \u002F descheduler on\nVERDICT: AUTOMATION_DEPLOYED\nGAP: None\n```\n\n### Worked example — REVIEW_REQUIRED\n\nREVIEW_REQUIRED worked example (staging cluster with metrics-server missing, no HPA\u002FPDBs, on-demand only) moved verbatim to [references\u002Fworked-examples.md](references\u002Fworked-examples.md).\nLoad on demand when the verdict is REVIEW_REQUIRED; the primary AUTOMATION_DEPLOYED example stays inline.\n\n## Anti-Patterns — NEVER do these things\n\n- NEVER run Karpenter and Cluster Autoscaler simultaneously. They\n  conflict: CA provisions, Karpenter deletes, CA re-provisions.\n\n- NEVER use `WhenEmptyOrUnderutilized` without a disruption budget in\n  production. Karpenter can simultaneously disrupt a large fraction of\n  nodes during consolidation.\n\n- NEVER use spot from a single instance family. Diversify across at\n  least 3-4 families to survive capacity reclamation events.\n\n- NEVER run NTH alongside Karpenter's interruption handling. Both watch\n  the same SQS queue, causing double-drain race conditions.\n\n- NEVER set HPA `scaleDown.stabilizationWindowSeconds: 0`. Zero causes\n  immediate scale-down on transient dips, then scale-back-up — thrashing.\n  Set at least 300s.\n\n- NEVER use VPA `Auto` on single-replica Deployments. VPA recreates the\n  pod, causing an outage. Use `Off` or `Initial`.\n\n- NEVER run HPA on CPU AND VPA on CPU simultaneously. They oscillate:\n  VPA raises request → HPA sees lower utilization → HPA scales down.\n\n- NEVER set pause-pod priority equal to or higher than real workloads.\n  Must be lower (e.g., `-1`) so real pods can preempt them.\n\n- NEVER omit PDBs for multi-replica production workloads. Without PDB,\n  voluntary disruption can evict all replicas simultaneously.\n\n- NEVER run descheduler without `maxNoOfPodsToEvictPerNode` and\n  `maxNoOfPodsToEvictTotal`. Without caps, mass eviction overwhelms the\n  scheduler. Start with 3\u002Fnode and 50 total.\n\n- NEVER set CA `scaleDownUtilizationThreshold` above 0.7. Too high\n  prevents scale-down; too low causes churn. 0.5 is the sweet spot.\n\n- NEVER rely on PDB alone for spot workloads. PDB protects against\n  voluntary disruption only. Combine with Karpenter\u002FNTH + topology\n  spread constraints.\n\n- NEVER enable KEDA without `cooldownPeriod`. Default is 300s; setting\n  to 0 causes rapid scale-up\u002Fscale-down oscillation for bursty workloads.\n\n## Pre-flight safety checks\n\nPre-flight safety checks (CONFIRM gate, CA-to-Karpenter migration, WhenEmptyOrUnderutilized graduation, descheduler DryRun, spot SQS verification) moved verbatim to [references\u002Fdiagnostic-commands.md](references\u002Fdiagnostic-commands.md).\nLoad on demand before executing any state-changing autoscaling operation.\n\n## Appendix A — Karpenter vs CA decision tree\n\nKarpenter-vs-CA decision tree appendix moved verbatim to [references\u002Fadvanced-patterns.md](references\u002Fadvanced-patterns.md); the Step 1 comparison table stays inline.\nLoad on demand when deciding the cluster-autoscaler tool.\n\n## Appendix B — HPA behavior quick reference\n\nHPA behavior quick-reference appendix (API\u002Fworker\u002Fbatch scale policies) moved verbatim to [references\u002Fadvanced-patterns.md](references\u002Fadvanced-patterns.md).\nLoad on demand when tuning HPA behavior policies.\n\n## Appendix C — Spot instance diversification\n\nSpot diversification appendix (per-workload family tables + Karpenter flexible selector) moved verbatim to [references\u002Fadvanced-patterns.md](references\u002Fadvanced-patterns.md).\nLoad on demand when planning spot instance diversification.\n\n## Recent AWS features (2024-2026)\n\nRecent AWS features (Karpenter v1 GA, disruption budgets, EKS Auto Mode, KEDA v2.15, HPA v2 behavior, VPA in-place resize) moved verbatim to [references\u002Fadvanced-patterns.md](references\u002Fadvanced-patterns.md).\nLoad on demand when picking component versions.\n\n## Expert heuristic: autoscaling blast radius\n\nAutoscaling blast-radius heuristic (conservative-first rule, component interaction cascade, 4-phase wave-enabling pattern, failure-detection alarms) moved verbatim to [references\u002Fadvanced-patterns.md](references\u002Fadvanced-patterns.md).\nLoad on demand when planning phased rollout or setting failure alarms.\n\n## References (load on demand)\n\n- [references\u002Fadvanced-patterns.md](references\u002Fadvanced-patterns.md) — Step 0 expert behaviours, Appendices A-C, Recent AWS features (2024-2026), and the blast-radius heuristic, moved from SKILL.md\n- [references\u002Fworked-examples.md](references\u002Fworked-examples.md) — secondary REVIEW_REQUIRED worked example, moved from SKILL.md (primary AUTOMATION_DEPLOYED example stays inline)\n- [references\u002Fdiagnostic-commands.md](references\u002Fdiagnostic-commands.md) — pre-flight safety checks, moved from SKILL.md\n- [references\u002Fkarpenter-configuration.md](references\u002Fkarpenter-configuration.md) — Karpenter API reference — extended with the EC2NodeClass manifest moved from SKILL.md\n- [references\u002Fhpa-keda-descheduler.md](references\u002Fhpa-keda-descheduler.md) — HPA\u002FKEDA\u002Fdescheduler detail — extended with the HPA and KEDA ScaledObject manifests moved from SKILL.md\n\n## Domain\n\nAWS CloudOps \u002F Compute Automation — EKS autoscaling.\n\n## AWS documentation\n\n- **Karpenter** — https:\u002F\u002Fkarpenter.sh\u002Fdocs\u002F\n- **Cluster Autoscaler on EKS** — https:\u002F\u002Fdocs.aws.amazon.com\u002Feks\u002Flatest\u002Fuserguide\u002Fcluster-autoscaler.html\n- **Horizontal Pod Autoscaler** — https:\u002F\u002Fkubernetes.io\u002Fdocs\u002Ftasks\u002Frun-application\u002Fhorizontal-pod-autoscale\u002F\n- **KEDA** — https:\u002F\u002Fkeda.sh\u002Fdocs\u002F\n- **EKS Auto Mode** — https:\u002F\u002Fdocs.aws.amazon.com\u002Feks\u002Flatest\u002Fuserguide\u002Fauto-mode.html\n","# EKS Autoscaling Automator\n\n## Mindset\n\n**One-line takeaway:** every EKS autoscaling system operates at two\nlayers — **cluster autoscaling** (adding\u002Fremoving nodes: Karpenter or\nCluster Autoscaler) and **pod autoscaling** (adding\u002Fremoving pods: HPA\nor KEDA). A gap in either layer produces a silent failure: pods are\nPending because no node has capacity, or nodes sit idle because pod\ncount never scales with load.\n\n- **Cluster Autoscaler and Karpenter are mutually exclusive.** Running\n  both causes conflicting scaling decisions — CA provisions a node,\n  Karpenter de-provisions it, ad infinitum.\n- **HPA without custom metrics is just CPU autoscaling.** For request-\n  driven workloads (latency, queue depth), HPA needs custom\u002Fexternal\n  metrics via Prometheus Adapter or KEDA.\n- **Spot without interruption handling is an outage waiting to happen.**\n  Spot instances get a 2-minute warning. Without the termination handler\n  or Karpenter's native disruption, pods are killed ungracefully.\n\n## Quick navigation\n\n| You want to... | Go to |\n|---|---|\n| Choose Karpenter vs Cluster Autoscaler | Step 1 |\n| Deploy Karpenter with NodePool + disruption budget | Step 2 |\n| Deploy Cluster Autoscaler with auto-discovery | Step 3 |\n| Configure managed node group scaling | Step 4 |\n| Set up HPA with custom metrics | Step 5 |\n| Understand VPA caveats | Step 6 |\n| Deploy KEDA for event-driven scaling | Step 7 |\n| Handle spot interruptions | Step 8 |\n| Configure overprovisioning pause-pods | Step 9 |\n| Set up PDBs and priority classes | Step 10 |\n| Deploy descheduler for bin-packing | Step 11 |\n\n## Critical rules at a glance (do NOT bury these)\n\n1. **Karpenter consolidation replaces nodes aggressively.**\n   `WhenEmptyOrUnderutilized` will DELETE underutilized nodes and\n   reschedule pods. For production, use `disruptionBudgets` to cap how\n   many nodes Karpenter disrupts at once. Without a budget, Karpenter\n   can drain 50% of nodes simultaneously.\n\n2. **Spot diversification across instance families is mandatory.** A\n   NodePool requesting only `c5.large` spot will be evicted en masse\n   when c5 capacity disappears. Diversify across at least 3-4 families\n   (c5, m5, c6i, m6i).\n\n3. **Overprovisioning pause-pods create headroom for fast scale-up.**\n   Without them, new pods wait 2-5 minutes for a node to boot. Pause-\n   pods (low priority) occupy scheduled capacity; real workloads preempt\n   them and schedule immediately.\n\n4. **HPA and VPA are incompatible for the same CPU\u002FMemory dimension.**\n   HPA scales replicas based on CPU; VPA adjusts CPU requests. Running\n   both on CPU causes oscillation. Use VPA in `Off` mode alongside HPA,\n   or VPA for memory only.\n\n5. **PDBs protect against voluntary disruption only.** A PDB with\n   `minAvailable: 2` prevents `kubectl drain` from evicting below 2\n   replicas. It does NOT protect against spot termination (involuntary).\n   For spot, combine PDB + termination handler + topology spread.\n\n## Pre-flight: data requirements\n\n| Input | Source | Why |\n|---|---|---|\n| EKS cluster version | `eks describe-cluster` | Karpenter v1 requires 1.27+ |\n| Current node count\u002Ftype | `kubectl get nodes -o wide` | Baseline fleet |\n| Workload resource requests | `kubectl get pods -A -o json` | Drives node sizing |\n| Current HPA\u002FVPA\u002FKEDA | `kubectl get hpa,vpa -A` | Don't duplicate |\n| Spot vs on-demand ratio | Business requirements | Cost target |\n| Metrics server health | `kubectl top pods` | Required for default HPA |\n| Prometheus (if custom metrics) | `kubectl get svc -n monitoring` | For custom-metric HPA |\n\n**If the input is malformed**, emit:\n```text\nAUTOSCALING: \u003Creference>\nVERDICT: ERROR\nREASON: Cannot design autoscaling — cluster version, node count, and workload requests are required.\nGAP: Run kubectl get nodes -o wide and kubectl get pods -A -o json.\n```\n\n## Process — Autoscaling design (apply in order)\n\n### Step 0: Expert knowledge — non-obvious autoscaling behaviors\n\nStep 0 expert-knowledge deep dive (Karpenter v1 API migration, consolidation modes, CA scale-down blockers, KEDA trigger semantics, IMDS spot notices, pause-pod priority, VPA Auto restarts, descheduler PodLifeTime, MNG desiredSize override) moved verbatim to [references\u002Fadvanced-patterns.md](references\u002Fadvanced-patterns.md).\nLoad on demand before choosing or tuning any autoscaling component.\n\n### Step 1: Choose Karpenter vs Cluster Autoscaler\n\n| Dimension | Karpenter | Cluster Autoscaler |\n|---|---|---|\n| Scaling speed | ~30s (direct EC2 API) | ~2-5 min (ASG lifecycle) |\n| Node selection | Flexible (any type, AZ, spot+OD mix) | Rigid (ASG instance types) |\n| Consolidation | Yes (deletes underutilized nodes) | Limited (empty nodes only) |\n| Spot handling | Native disruption + diversified | Via ASG mixed policy |\n| Maturity | v1 GA (2024) | Battle-tested, stable |\n| Best for | Dynamic, spot-heavy, fast scale | Stable, predictable capacity |\n\nChoose **Karpenter** if: spot priority, \u003C 1 min scale-up, consolidation\nneeded, heterogeneous types. Choose **CA** if: fixed types in managed\nnode groups, simplicity priority.\n\n**NEVER run both.** Migrate: uninstall CA first, then install Karpenter.\n\n### Step 2: Deploy Karpenter with NodePool and disruption budgets\n\n```bash\nhelm install karpenter oci:\u002F\u002Fpublic.ecr.aws\u002Fkarpenter\u002Fkarpenter \\\n  --namespace kube-system \\\n  --set clusterName=my-prod-cluster \\\n  --set interruptionQueue=true \\\n  --version 1.0.0\n```\n\nNodePool with spot diversification + disruption budget:\n\n```yaml\napiVersion: karpenter.sh\u002Fv1\nkind: NodePool\nmetadata:\n  name: default\nspec:\n  template:\n    spec:\n      requirements:\n        - key: karpenter.k8s.aws\u002Finstance-category\n          operator: In\n          values: [\"c\", \"m\"]\n        - key: karpenter.k8s.aws\u002Finstance-generation\n          operator: Gt\n          values: [\"5\"]\n        - key: karpenter.sh\u002Fcapacity-type\n          operator: In\n          values: [\"spot\", \"on-demand\"]\n      nodeClassRef:\n        group: karpenter.k8s.aws\n        kind: EC2NodeClass\n        name: default\n  limits:\n    cpu: 1000\n  disruption:\n    consolidationPolicy: WhenEmptyOrUnderutilized\n    consolidateAfter: 30s\n    budgets:\n      - nodes: \"20%\"\n      - schedule: \"0 9 * * Mon-Fri\"\n        duration: 8h\n        nodes: \"0\"\n```\n\nFull EC2NodeClass manifest (AMI alias, subnet\u002FSG selectors, role, block devices) moved verbatim to [references\u002Fkarpenter-configuration.md](references\u002Fkarpenter-configuration.md).\nLoad on demand when authoring the EC2NodeClass that pairs with the NodePool above.\n\n**Key configuration:**\n- `disruption.budgets`: First budget allows 20% disruption normally.\n  Second budget blocks ALL disruption Mon-Fri 09:00-17:00. Critical for\n  production stability.\n- Spot diversification: `instance-category: [c, m]` + `generation: >5`\n  allows 8+ instance types. If c5 spot disappears, Karpenter tries m5\n  or c6i immediately.\n- `interruptionQueue: true`: enables native spot interruption handling\n  via SQS + EventBridge. Do NOT install NTH when this is enabled.\n\nSee **references\u002Fkarpenter-configuration.md** for full API reference,\nrequirement keys, and consolidation troubleshooting.\n\n### Step 3: Deploy Cluster Autoscaler with auto-discovery\n\n```bash\nhelm install cluster-autoscaler autoscaler\u002Fcluster-autoscaler \\\n  --namespace kube-system \\\n  --set autoDiscovery.clusterName=my-prod-cluster \\\n  --set awsRegion=us-east-1 \\\n  --version 9.34.0\n```\n\nAuto-discovery tags (must be on node group ASGs):\n```\nk8s.io\u002Fcluster-autoscaler\u002Fmy-prod-cluster = owned\nk8s.io\u002Fcluster-autoscaler\u002Fenabled = true\n```\n\nKey tuning: `scaleDownUnneededTime: \"30m\"`,\n`scaleDownUtilizationThreshold: \"0.5\"`,\n`skipNodesWithLocalStorage: true`, `expander: \"least-waste\"`.\n\n### Step 4: Configure managed node group scaling\n\n```bash\naws eks update-nodegroup-config \\\n  --cluster-name my-prod-cluster --nodegroup-name my-ng \\\n  --scaling-config desiredSize=5,minSize=3,maxSize=20\n```\n\nFor EKS Auto Mode (2025+, simplest path for new clusters):\n```bash\naws eks update-cluster-config --name my-prod-cluster \\\n  --compute-config '{\"nodeRoleArn\":\"arn:...\",\"enabled\":true}'\n```\n\n### Step 5: Set up HPA with custom metrics\n\nHPA with CPU + custom metric + behavior tuning:\n\nFull HPA manifest (CPU + custom metric + scaleUp\u002FscaleDown behavior policies) moved verbatim to [references\u002Fhpa-keda-descheduler.md](references\u002Fhpa-keda-descheduler.md).\nLoad on demand when authoring the HPA object for a deployment.\n\n**Behavior tuning guide:** scale UP fast (users waiting), scale DOWN\nslow (avoid thrashing). `scaleDown.stabilizationWindowSeconds: 300`\nminimum for production.\n\nSee **references\u002Fhpa-keda-descheduler.md** for custom metrics setup,\nKEDA triggers, and descheduler strategies.\n\n### Step 6: VPA caveats and incompatibilities\n\n| Mode | Behavior | Production-safe? |\n|---|---|---|\n| `Auto` | Recreates pods to apply requests | No (causes restarts) |\n| `Initial` | Sets requests only on NEW pods | Yes |\n| `Off` | Recommendations only | Yes |\n\n**HPA + VPA CPU conflict:** VPA raises CPU request → HPA sees lower\nutilization → HPA scales down → loop. Resolution: VPA `Off` mode (apply\nmanually) or VPA for memory only.\n\n### Step 7: Deploy KEDA for event-driven scaling\n\n```bash\nhelm install keda kedacore\u002Fkeda --namespace keda-system --create-namespace --version 2.15.0\n```\n\nSQS-triggered ScaledObject (scale-to-zero):\n\nSQS-triggered ScaledObject manifest (scale-to-zero, queueLength trigger, polling\u002Fcooldown) moved verbatim to [references\u002Fhpa-keda-descheduler.md](references\u002Fhpa-keda-descheduler.md).\nLoad on demand when authoring a KEDA ScaledObject.\n\n**KEDA vs HPA:** use KEDA when you need scale-to-zero, external-system\ntriggers, or multiple trigger sources. Use HPA when CPU\u002FMemory suffices.\n\n### Step 8: Handle spot instance interruptions\n\n**Karpenter approach (recommended):** Set `interruptionQueue: true`.\nKarpenter watches SQS fed by EventBridge, receives 2-min notice, cordons\n+ drains + provisions replacement. No additional components.\n\n**Cluster Autoscaler approach (NTH):**\n```bash\nhelm install aws-node-termination-handler eks\u002Faws-node-termination-handler \\\n  --namespace kube-system \\\n  --set enableSpotInterruptionDraining=true \\\n  --set enableQueueProcessing=true \\\n  --set awsRegion=us-east-1\n```\n\n**NEVER run both** — they conflict on the same SQS queue.\n\n**Spot diversification rules:**\n\n| Rule | Why |\n|---|---|\n| 3+ instance families | Reduces mass-reclamation risk |\n| 2+ instance sizes | Same |\n| 2+ AZs | Capacity shortages are AZ-specific |\n| `capacity-type: [spot, on-demand]` | Fallback to on-demand |\n\n### Step 9: Configure overprovisioning pause-pods\n\nPriority class (MUST be lower than default):\n\n```yaml\napiVersion: scheduling.k8s.io\u002Fv1\nkind: PriorityClass\nmetadata:\n  name: overprovisioning\nvalue: -1\ndescription: \"Pause-pod priority (lowest).\"\n```\n\nPause-pod Deployment:\n\n```yaml\napiVersion: apps\u002Fv1\nkind: Deployment\nmetadata:\n  name: pause-pods\n  namespace: kube-system\nspec:\n  replicas: 10\n  strategy: {type: Recreate}\n  selector: {matchLabels: {app: pause-pod}}\n  template:\n    metadata: {labels: {app: pause-pod}}\n    spec:\n      priorityClassName: overprovisioning\n      containers:\n        - name: pause\n          image: registry.k8s.io\u002Fpause:3.9\n          resources:\n            requests: {cpu: 1, memory: 1Gi}\n```\n\nEach pause-pod occupies 1 CPU + 1 GiB. For 10 pods headroom, Karpenter\nprovisions 10 pods worth of extra nodes. When a real deployment scales\nup, it preempts pause-pods and schedules immediately.\n\n### Step 10: Pod disruption budgets and priority classes\n\n```yaml\napiVersion: policy\u002Fv1\nkind: PodDisruptionBudget\nmetadata:\n  name: api-server-pdb\nspec:\n  minAvailable: 2\n  selector: {matchLabels: {app: api-server}}\n```\n\nPriority classes:\n\n```yaml\n# production-critical: value 1000000\n# production-standard: value 100000\n# batch-default: value 10000\n# default (globalDefault): value 0\n# overprovisioning: value -1\n```\n\nPreemption flow: when a node is full, the scheduler checks if lower-\npriority pods can be evicted to make room for a higher-priority pod.\n\n### Step 11: Deploy descheduler for bin-packing\n\n```bash\nhelm install descheduler descheduler\u002Fdescheduler \\\n  --namespace kube-system \\\n  --set kind=Deployment \\\n  --set schedule=\"0 *\u002F2 * * *\" \\\n  --set deschedulingInterval=2h \\\n  --version 0.31.0\n```\n\nKey safety settings: `maxNoOfPodsToEvictPerNode: 3`,\n`maxNoOfPodsToEvictTotal: 50`, `evictSystemCriticalPods: false`,\n`nodeFit: true` (CRITICAL — prevents evicting pods with nowhere to go).\n\nRun in DryRun mode first: `--set dryDescheduling=true`.\n\n## Output format\n\n```text\nAUTOSCALING: \u003Creference>\nCLUSTER: \u003CEKS version + node count>\nTOOLING:\n  - Cluster autoscaler: Karpenter \u003Cver> | CA \u003Cver> | EKS Auto Mode\n  - Pod autoscaler: HPA | KEDA | VPA (Off)\n  - Metrics: metrics-server | Prometheus Adapter\nPROVISIONING:\n  - NodePool: \u003Cname> \u002F Instance types: \u003Cdiversified list>\n  - Spot ratio: \u003C%> \u002F Disruption budget: \u003Cnodes per window>\nPOD_SCALING:\n  - HPA: \u003Cdeployment> min=N max=N metric=\u003Ctype>\n  - KEDA: \u003Cdeployment> trigger=\u003Ctype> min=N max=N\n  - VPA: \u003Cdeployment> mode=\u003COff|Initial>\nSPOT:\n  - Interruption handling: Karpenter native | NTH\n  - Diversification: \u003Cfamilies> \u002F Graceful drain: enabled|disabled\nSAFETY:\n  - Overprovisioning: \u003CN> x \u003Ccpu\u002Fmem>\n  - PDB: \u003Cdeployment> minAvailable=N\n  - Priority classes: \u003Clist> \u002F Descheduler: enabled|disabled\nVERDICT: AUTOMATION_DEPLOYED | REVIEW_REQUIRED\nGAP: \u003Cif REVIEW_REQUIRED, the specific missing piece>\nTEMPLATE: \u003CHelm values + kubectl manifests>\n```\n\n### Worked example — AUTOMATION_DEPLOYED\n\n```text\nAUTOSCALING: prod-cluster-autoscaling\nCLUSTER: EKS 1.30, 25 nodes\nTOOLING: Karpenter 1.0.0 (interruptionQueue) | HPA + KEDA | metrics-server + Prom Adapter\nPROVISIONING: default NodePool \u002F c5,m5,c6i,m6i \u002F 70% spot 30% OD \u002F 20% budget, freeze biz hrs\nPOD_SCALING: HPA api-server 3-50 (CPU+custom) | KEDA sqs-consumer 1-30 (SQS) | VPA Off\nSPOT: Karpenter native \u002F 4 families \u002F graceful drain\nSAFETY: 10 pause-pods x 1CPU\u002F1Gi \u002F PDB api-server minAvail=2 \u002F 4 priority classes \u002F descheduler on\nVERDICT: AUTOMATION_DEPLOYED\nGAP: None\n```\n\n### Worked example — REVIEW_REQUIRED\n\nREVIEW_REQUIRED worked example (staging cluster with metrics-server missing, no HPA\u002FPDBs, on-demand only) moved verbatim to [references\u002Fworked-examples.md](references\u002Fworked-examples.md).\nLoad on demand when the verdict is REVIEW_REQUIRED; the primary AUTOMATION_DEPLOYED example stays inline.\n\n## Anti-Patterns — NEVER do these things\n\n- NEVER run Karpenter and Cluster Autoscaler simultaneously. They\n  conflict: CA provisions, Karpenter deletes, CA re-provisions.\n\n- NEVER use `WhenEmptyOrUnderutilized` without a disruption budget in\n  production. Karpenter can simultaneously disrupt a large fraction of\n  nodes during consolidation.\n\n- NEVER use spot from a single instance family. Diversify across at\n  least 3-4 families to survive capacity reclamation events.\n\n- NEVER run NTH alongside Karpenter's interruption handling. Both watch\n  the same SQS queue, causing double-drain race conditions.\n\n- NEVER set HPA `scaleDown.stabilizationWindowSeconds: 0`. Zero causes\n  immediate scale-down on transient dips, then scale-back-up — thrashing.\n  Set at least 300s.\n\n- NEVER use VPA `Auto` on single-replica Deployments. VPA recreates the\n  pod, causing an outage. Use `Off` or `Initial`.\n\n- NEVER run HPA on CPU AND VPA on CPU simultaneously. They oscillate:\n  VPA raises request → HPA sees lower utilization → HPA scales down.\n\n- NEVER set pause-pod priority equal to or higher than real workloads.\n  Must be lower (e.g., `-1`) so real pods can preempt them.\n\n- NEVER omit PDBs for multi-replica production workloads. Without PDB,\n  voluntary disruption can evict all replicas simultaneously.\n\n- NEVER run descheduler without `maxNoOfPodsToEvictPerNode` and\n  `maxNoOfPodsToEvictTotal`. Without caps, mass eviction overwhelms the\n  scheduler. Start with 3\u002Fnode and 50 total.\n\n- NEVER set CA `scaleDownUtilizationThreshold` above 0.7. Too high\n  prevents scale-down; too low causes churn. 0.5 is the sweet spot.\n\n- NEVER rely on PDB alone for spot workloads. PDB protects against\n  voluntary disruption only. Combine with Karpenter\u002FNTH + topology\n  spread constraints.\n\n- NEVER enable KEDA without `cooldownPeriod`. Default is 300s; setting\n  to 0 causes rapid scale-up\u002Fscale-down oscillation for bursty workloads.\n\n## Pre-flight safety checks\n\nPre-flight safety checks (CONFIRM gate, CA-to-Karpenter migration, WhenEmptyOrUnderutilized graduation, descheduler DryRun, spot SQS verification) moved verbatim to [references\u002Fdiagnostic-commands.md](references\u002Fdiagnostic-commands.md).\nLoad on demand before executing any state-changing autoscaling operation.\n\n## Appendix A — Karpenter vs CA decision tree\n\nKarpenter-vs-CA decision tree appendix moved verbatim to [references\u002Fadvanced-patterns.md](references\u002Fadvanced-patterns.md); the Step 1 comparison table stays inline.\nLoad on demand when deciding the cluster-autoscaler tool.\n\n## Appendix B — HPA behavior quick reference\n\nHPA behavior quick-reference appendix (API\u002Fworker\u002Fbatch scale policies) moved verbatim to [references\u002Fadvanced-patterns.md](references\u002Fadvanced-patterns.md).\nLoad on demand when tuning HPA behavior policies.\n\n## Appendix C — Spot instance diversification\n\nSpot diversification appendix (per-workload family tables + Karpenter flexible selector) moved verbatim to [references\u002Fadvanced-patterns.md](references\u002Fadvanced-patterns.md).\nLoad on demand when planning spot instance diversification.\n\n## Recent AWS features (2024-2026)\n\nRecent AWS features (Karpenter v1 GA, disruption budgets, EKS Auto Mode, KEDA v2.15, HPA v2 behavior, VPA in-place resize) moved verbatim to [references\u002Fadvanced-patterns.md](references\u002Fadvanced-patterns.md).\nLoad on demand when picking component versions.\n\n## Expert heuristic: autoscaling blast radius\n\nAutoscaling blast-radius heuristic (conservative-first rule, component interaction cascade, 4-phase wave-enabling pattern, failure-detection alarms) moved verbatim to [references\u002Fadvanced-patterns.md](references\u002Fadvanced-patterns.md).\nLoad on demand when planning phased rollout or setting failure alarms.\n\n## References (load on demand)\n\n- [references\u002Fadvanced-patterns.md](references\u002Fadvanced-patterns.md) — Step 0 expert behaviours, Appendices A-C, Recent AWS features (2024-2026), and the blast-radius heuristic, moved from SKILL.md\n- [references\u002Fworked-examples.md](references\u002Fworked-examples.md) — secondary REVIEW_REQUIRED worked example, moved from SKILL.md (primary AUTOMATION_DEPLOYED example stays inline)\n- [references\u002Fdiagnostic-commands.md](references\u002Fdiagnostic-commands.md) — pre-flight safety checks, moved from SKILL.md\n- [references\u002Fkarpenter-configuration.md](references\u002Fkarpenter-configuration.md) — Karpenter API reference — extended with the EC2NodeClass manifest moved from SKILL.md\n- [references\u002Fhpa-keda-descheduler.md](references\u002Fhpa-keda-descheduler.md) — HPA\u002FKEDA\u002Fdescheduler detail — extended with the HPA and KEDA ScaledObject manifests moved from SKILL.md\n\n## Domain\n\nAWS CloudOps \u002F Compute Automation — EKS autoscaling.\n\n## AWS documentation\n\n- **Karpenter** — https:\u002F\u002Fkarpenter.sh\u002Fdocs\u002F\n- **Cluster Autoscaler on EKS** — https:\u002F\u002Fdocs.aws.amazon.com\u002Feks\u002Flatest\u002Fuserguide\u002Fcluster-autoscaler.html\n- **Horizontal Pod Autoscaler** — https:\u002F\u002Fkubernetes.io\u002Fdocs\u002Ftasks\u002Frun-application\u002Fhorizontal-pod-autoscale\u002F\n- **KEDA** — https:\u002F\u002Fkeda.sh\u002Fdocs\u002F\n- **EKS Auto Mode** — https:\u002F\u002Fdocs.aws.amazon.com\u002Feks\u002Flatest\u002Fuserguide\u002Fauto-mode.html\n",{"name":5,"license":15,"metadata":24,"description":7,"compatibility":42},{"tags":25,"phase":26,"author":27,"domain":28,"family":29,"version":30,"keywords":31,"task_type":32,"complexity":33,"entry_point":34,"skill_class":35,"when_to_use":36,"requires_llm":37,"verdict_shape":38,"lifecycle_status":39,"invocation_schema":40,"supports_pipeline":37,"activation_triggers":41},"eks, karpenter, cluster-autoscaler, hpa, keda, spot, autoscaling, automate","4","Jacky Chan — AWS Community Builder","aws-cloudops","Compute","0.1.0","EKS autoscaling, Karpenter, Cluster Autoscaler, Horizontal Pod Autoscaler, HPA, Vertical Pod Autoscaler, VPA, KEDA, spot instance interruption, overprovisioning, pod disruption budget, priority class, descheduler, consolidation, disruption budget","automate","high","false","capability","Building EKS autoscaling (Karpenter or Cluster Autoscaler), configuring HPA\u002FVPA\u002FKEDA for workload scaling, handling spot instance interruptions, optimizing scale-up latency with overprovisioning, protecting workloads with PDBs and priority classes, or improving bin-packing with descheduler.","true","AUTOMATION_DEPLOYED | REVIEW_REQUIRED","active","Input: either (a) an EKS cluster description (version, node count, workload mix, scaling requirements, spot\u002Fon-demand ratio) plus observability metrics, OR (b) an autoscaling request (\"set up Karpenter with spot diversification\", \"configure HPA with custom metrics for the API deployment\"). Output: deterministic AUTOSCALING block — TOOLING\u002FPROVISIONING\u002FPOD_SCALING\u002FSPOT\u002FSAFETY\u002FVERDICT — where VERDICT is AUTOMATION_DEPLOYED (Helm values + manifests ready) or REVIEW_REQUIRED (specific gap cited).","EKS autoscaling, Karpenter setup, Cluster Autoscaler, horizontal pod autoscaler, KEDA event-driven scaling, spot interruption handling, overprovisioning pause pods, pod disruption budget, descheduler bin-packing, node group scaling policy","Agent runtime that reads SKILL.md (Claude Code, Cursor, Windsurf, Codex, Gemini). No AWS CLI or kubectl required for offline workflow design. Live deployment uses helm install\u002Fupgrade (Cluster Autoscaler, Karpenter, KEDA, descheduler), kubectl apply (HPA, VPA, PDB, priority classes, Karpenter NodePool\u002FNodeClaim), aws eks update-nodegroup-config (managed node group scaling) — AWS CLI v2, kubectl, helm v3.","main","eeace51bd5dd1db2030729d83eeab988861553aa","skills\u002Feks-autoscaling-automator",0,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills","https:\u002F\u002Fgithub.com\u002Fjhkchan","CTO & Co-Founder @Beever-AI  & @votee. CAISP. AWS Community Builder. Organizer@GDGCloudHK. MongoDB Champion. Technopreneur + Infornographer.","2026-09-24 08:54:10",{"source":8,"skillName":5,"commands":52,"prompt":62,"zipAvailable":63},[53,56,59],{"tool":54,"command":55},"npx","npx skills add jhkchan\u002Faws-agent-skills --skill eks-autoscaling-automator",{"tool":57,"command":58},"bunx","bunx skills add jhkchan\u002Faws-agent-skills --skill eks-autoscaling-automator",{"tool":60,"command":61},"pnpm","pnpm dlx skills add jhkchan\u002Faws-agent-skills --skill eks-autoscaling-automator","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Feks-autoscaling-automator\u002FSKILL.md",true,[65,71,75,79,83,87,91,95,99,103,107,111,115,119,123,127,131,135,139,143],{"path":66,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":69,"sourceHtmlUrl":70},"eval\u002Ftest-cases.yaml","REFERENCE","参考文档",6663,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Feks-autoscaling-automator\u002Feval\u002Ftest-cases.yaml",{"path":72,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":73,"sourceHtmlUrl":74},"evals\u002Fbaselines\u002Fca-to-karpenter-migration.md",892,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Feks-autoscaling-automator\u002Fevals\u002Fbaselines\u002Fca-to-karpenter-migration.md",{"path":76,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":77,"sourceHtmlUrl":78},"evals\u002Fbaselines\u002Fkarpenter-spot-full-deploy.md",1316,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Feks-autoscaling-automator\u002Fevals\u002Fbaselines\u002Fkarpenter-spot-full-deploy.md",{"path":80,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":81,"sourceHtmlUrl":82},"evals\u002Fbaselines\u002Fkeda-sqs-scale-to-zero.md",1025,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Feks-autoscaling-automator\u002Fevals\u002Fbaselines\u002Fkeda-sqs-scale-to-zero.md",{"path":84,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":85,"sourceHtmlUrl":86},"evals\u002Fbaselines\u002Fmissing-metrics-server.md",750,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Feks-autoscaling-automator\u002Fevals\u002Fbaselines\u002Fmissing-metrics-server.md",{"path":88,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":89,"sourceHtmlUrl":90},"evals\u002Fbaselines\u002Fvpa-hpa-conflict.md",929,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Feks-autoscaling-automator\u002Fevals\u002Fbaselines\u002Fvpa-hpa-conflict.md",{"path":92,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":93,"sourceHtmlUrl":94},"evals\u002Feval_config.yaml",1720,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Feks-autoscaling-automator\u002Fevals\u002Feval_config.yaml",{"path":96,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":97,"sourceHtmlUrl":98},"evals\u002Fevals.json",4209,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Feks-autoscaling-automator\u002Fevals\u002Fevals.json",{"path":100,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":101,"sourceHtmlUrl":102},"evals\u002Fprompts\u002Fca-to-karpenter-migration.md",760,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Feks-autoscaling-automator\u002Fevals\u002Fprompts\u002Fca-to-karpenter-migration.md",{"path":104,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":105,"sourceHtmlUrl":106},"evals\u002Fprompts\u002Fkarpenter-spot-full-deploy.md",959,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Feks-autoscaling-automator\u002Fevals\u002Fprompts\u002Fkarpenter-spot-full-deploy.md",{"path":108,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":109,"sourceHtmlUrl":110},"evals\u002Fprompts\u002Fkeda-sqs-scale-to-zero.md",672,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Feks-autoscaling-automator\u002Fevals\u002Fprompts\u002Fkeda-sqs-scale-to-zero.md",{"path":112,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":113,"sourceHtmlUrl":114},"evals\u002Fprompts\u002Fmissing-metrics-server.md",483,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Feks-autoscaling-automator\u002Fevals\u002Fprompts\u002Fmissing-metrics-server.md",{"path":116,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":117,"sourceHtmlUrl":118},"evals\u002Fprompts\u002Fvpa-hpa-conflict.md",518,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Feks-autoscaling-automator\u002Fevals\u002Fprompts\u002Fvpa-hpa-conflict.md",{"path":120,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":121,"sourceHtmlUrl":122},"examples\u002Fend-to-end.md",9740,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Feks-autoscaling-automator\u002Fexamples\u002Fend-to-end.md",{"path":124,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":125,"sourceHtmlUrl":126},"references\u002Fadvanced-patterns.md",5459,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Feks-autoscaling-automator\u002Freferences\u002Fadvanced-patterns.md",{"path":128,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":129,"sourceHtmlUrl":130},"references\u002Fdiagnostic-commands.md",643,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Feks-autoscaling-automator\u002Freferences\u002Fdiagnostic-commands.md",{"path":132,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":133,"sourceHtmlUrl":134},"references\u002Fhpa-keda-descheduler.md",11234,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Feks-autoscaling-automator\u002Freferences\u002Fhpa-keda-descheduler.md",{"path":136,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":137,"sourceHtmlUrl":138},"references\u002Fkarpenter-configuration.md",8261,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Feks-autoscaling-automator\u002Freferences\u002Fkarpenter-configuration.md",{"path":140,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":141,"sourceHtmlUrl":142},"references\u002Fworked-examples.md",648,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Feks-autoscaling-automator\u002Freferences\u002Fworked-examples.md",{"path":144,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":145,"sourceHtmlUrl":146},"skill.usf.yaml",4632,"https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Feks-autoscaling-automator\u002Fskill.usf.yaml",[148,155,162,169,176,183],{"skillNo":149,"name":150,"nameZh":6,"description":151,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":152,"categories":153,"keywords":154,"publishedAt":19},"01M7ADVD60X4PCAPKSE58P9NDC","xray-tracing-deployer","Deploys AWS X-Ray distributed tracing with production-grade configuration: X-Ray daemon deployment (EC2 systemd, ECS sidecar, EKS DaemonSet, Lambda built-in), SDK instrumentation (Java, Python, Node.js, Go, .NET) with annotations and metadata, sampling rules (default reservoir + rate vs custom rules with URL \u002F method \u002F service-name predicates), service map generation, trace query and retrieval, groups and insights, CloudWatch ServiceLens integration, IAM permissions (PutTraceSegments, GetSamplingRules), and latest features (Lambda Powertools tracing, OpenTelemetry \u002F ADOT Collector support, W3C trace context). Emits a READY_TO_DEPLOY checklist with every configuration item verified. Use when enabling X-Ray on a new application, adding tracing to ECS \u002F EKS \u002F Lambda, validating a sampling configuration, or generating daemon deployment CLI and IaC templates. Triggers: X-Ray, distributed tracing, X-Ray daemon, sampling rules, service map, X-Ray SDK, OpenTelemetry, ADOT, Lambda Powertools, CloudWatch ServiceLens.","https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fxray-tracing-deployer\u002FSKILL.md",[],[],{"skillNo":156,"name":157,"nameZh":6,"description":158,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":159,"categories":160,"keywords":161,"publishedAt":19},"01M7ADTVJ7DCBM9CB1FY51GH7J","wellarchitected-workload-auditor","Audits AWS Well-Architected Tool workloads for review staleness, per-pillar high-risk issue counts, milestone tracking gaps, and remediation plan completeness. Emits a deterministic verdict (STALE_REVIEW | HIGH_RISK | CONFIG_GAP | OK) per workload with per-pillar risk breakdown, milestone trend analysis, and specific remediation. Use when reviewing Well-Architected workloads, checking WA review freshness, auditing high-risk issues per pillar, validating milestone coverage, or tracking remediation plans.","https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fwellarchitected-workload-auditor\u002FSKILL.md",[],[],{"skillNo":163,"name":164,"nameZh":6,"description":165,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":166,"categories":167,"keywords":168,"publishedAt":19},"01M7ADT59G6FTTNZYHT2QRPNNP","wellarchitected-review-operator","Operates AWS Well-Architected Tool reviews end-to-end — creates workloads, runs pillar reviews (operational excellence, security, reliability, performance, cost optimization, sustainability, Prosperity), answers questions with risk-tier rationale, generates reports and improvement plans, creates milestones, and integrates with Trusted Advisor and Well-Architected Labs. Runs deterministic pre-checks (workload existence, lens availability, pillar coverage), emits the exact update-answer \u002F get-consolidated-report CLI behind a CONFIRM gate, and verifies risk-tier transitions. Emits READY | BLOCKED | COMPLETED. Use when creating a workload, answering pillar questions, generating an improvement plan, creating a milestone, or integrating with Trusted Advisor. Triggers: well-architected tool, well architected review, workload review, pillar, operational excellence, security, reliability, performance, cost optimization, sustainability, prosperity, improvement plan, milestone, trusted advisor.","https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fwellarchitected-review-operator\u002FSKILL.md",[],[],{"skillNo":170,"name":171,"nameZh":6,"description":172,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":173,"categories":174,"keywords":175,"publishedAt":19},"01M7ADSH5JA9QHNPPHQCD7EVA7","wafv2-web-acl-deployer","Provisions AWS WAFv2 Web ACLs with correct production defaults: scope selection (CLOUDFRONT vs REGIONAL), managed rule groups (AWSManagedRulesCommonRuleSet, SQLiRuleSet, LinuxRuleSet, WindowsRuleSet, IPReputationList, BotControlRuleSet, AmazonIpReputationList, ATP), custom rules (byte match, IP set, regex pattern, geo match, size constraint, rate-based), rule priority ordering, logging destinations (Kinesis Firehose, CloudWatch Logs, S3), resource association (ALB, API Gateway, CloudFront), rate-based rules with aggregate keys (IP, forwarded IP), Challenge and CAPTCHA actions, and account takeover prevention. Emits a READY_TO_DEPLOY checklist. Use when creating a Web ACL, attaching managed rules, configuring rate limiting, enabling WAF logging, associating a Web ACL with an ALB\u002FAPI Gateway\u002FCloudFront, or configuring CAPTCHA\u002FChallenge actions. Triggers: create WAF Web ACL, WAFv2, managed rule groups, WAF rate limiting, WAF logging, WAF CAPTCHA, WAF Challenge action, bot control, ATP managed rules.","https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fwafv2-web-acl-deployer\u002FSKILL.md",[],[],{"skillNo":177,"name":178,"nameZh":6,"description":179,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":180,"categories":181,"keywords":182,"publishedAt":19},"01M7ADRYFAT8R73FBNCHRC573X","wafv2-web-acl-auditor","Audits AWS WAFv2 Web ACL configurations to determine whether the ACL provides effective protection: default-action posture (Allow vs Block), managed-rule-group coverage gaps, rule effectiveness (BLOCK vs COUNT, shadow\u002Fbypass rules, stale exclusions), rate-based rule correctness, logging and visibility configuration, and text-transformation bypass vectors. Emits a deterministic verdict per Web ACL. Use when reviewing a WAFv2 Web ACL for security posture, checking managed-rule coverage, validating rate-limit configuration, auditing WAF logging, or hardening a WAF before production deployment. Triggers: WAF, WAFv2, Web ACL, managed rule group, AWSManagedRules, rate-based rule, CAPTCHA, Challenge, OverrideAction, Count mode, forwarded IP, Bot Control, ATP, text transformation, WAF logging, WAF visibility, OWASP, SQLi, XSS, RFI.","https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fwafv2-web-acl-auditor\u002FSKILL.md",[],[],{"skillNo":184,"name":185,"nameZh":6,"description":186,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":187,"categories":188,"keywords":189,"publishedAt":19},"01M7ADR8W7G72JX4KFGN03QK8Q","waf-rule-deployer","Provisions AWS WAFv2 rule sets with production defaults: Web ACL creation (CloudFront scope in us-east-1 vs Regional scope), rule groups (managed vs custom) with capacity planning in WCU (Web ACL Capacity Units, 1500 default per ACL), managed rule groups (AWSManagedRulesCommonRuleSet, AWSManagedRulesSQLiRuleSet, AWSManagedRulesAmazonIpReputationList, AWSManagedRulesLinuxRuleSet, AWSManagedRulesWindowsRuleSet, AWSManagedRulesUnixRuleSet), custom rules (byte-match, regex pattern set, geo-match, rate-based), IP set and regex pattern set resources, rule priority ordering (lowest number evaluated first), action types (allow, block, count, CAPTCHA, Challenge), bot control managed rule group, account takeover. Triggers: create WAF rule, WAF managed rule group, WAF custom rule, WAF byte match, WAF geo match, WAF rate based rule, WAF IP set, WAF regex pattern set, WAF WCU budget, WAF CloudFront ACL, WAF logging Kinesis Firehose, WAF bot control, WAF ATP, WAF label matching, WAF Challenge action, WAF CAPTCHA.","https:\u002F\u002Fgithub.com\u002Fjhkchan\u002Faws-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fwaf-rule-deployer\u002FSKILL.md",[],[]]