From ccds-infra
Kubernetes platform specialist. Owns cluster design, namespaces and RBAC, workload patterns (Deployments, StatefulSets, Jobs), autoscaling, ingress, and GitOps delivery. Auto-invoked for K8s manifest, cluster, controller, or deployment work.
How this skill is triggered — by the user, by Claude, or both
Slash command
/ccds-infra:infra-k8sThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Kubernetes rewards discipline and punishes drift. Requests/limits, PDBs,
Kubernetes rewards discipline and punishes drift. Requests/limits, PDBs, graceful shutdown, and GitOps are the difference between a platform and an incident factory.
minAvailable: 1 minimum; never a PDB that permits zero
disruptions on a 1-replica workload (it just blocks the drain instead).preStop sleep
(5–10 s) so endpoint/load-balancer deregistration propagates before the
process exits; size terminationGracePeriodSeconds (default 30 s) to the
real drain time. In-flight 502s on every deploy trace back here.kubectl apply to production. Drift is then detectable and
revertible, and "what changed" has one answer.spec:
replicas: 3 # ≥2 + PDB, spread across zones
template:
spec:
terminationGracePeriodSeconds: 45
topologySpreadConstraints: # zone spread for multi-AZ resilience
- maxSkew: 1
topologyKey: topology.kubernetes.io/zone
whenUnsatisfiable: ScheduleAnyway
containers:
- resources:
requests: { cpu: 250m, memory: 256Mi } # from observed p95
limits: { memory: 256Mi } # mem limit = request
readinessProbe: { httpGet: { path: /ready, port: 8080 } }
livenessProbe: { httpGet: { path: /healthz, port: 8080 } }
lifecycle:
preStop: { exec: { command: ["sleep", "8"] } }
---
apiVersion: policy/v1
kind: PodDisruptionBudget
spec: { minAvailable: 2, selector: { matchLabels: { app: svc } } }
latest image tags or mutable tags in GitOps repos — reconciliation becomes
non-deterministic; pin digests or immutable tagsconcurrencyPolicy: Forbid and idempotent payloads —
overlapping runs double-processRelated: infra-networking (CNI, ingress, mesh beyond the cluster),
infra-observability (cluster + workload metrics), infra-sre (SLOs the
rollout strategy must protect), infra-finops (bin-packing and node sizing) ·
domain agent: infra-architect · output/ADR format: playbook-conventions
npx claudepluginhub ggrace519/claude-code-dev-studio --plugin ccds-infraProvides behavioral guidelines to reduce common LLM coding mistakes, focusing on simplicity, surgical changes, assumption surfacing, and verifiable success criteria.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.