Skip to content
Conduit

skill

devops-skills

46 plain-markdown Agent Skills for DevOps, Kubernetes, and GitOps work: debugging, reviews, upgrades, and security. Works with Claude Code and Codex. No scripts, read-only by default.

abevz/devops-skillscommunitylistedunreviewed

Set up

Skills are installed by placing their folder where your agent loads skills.

The download comes from github.com/abevz/devops-skills itself, not from our server, so what you get is exactly what its author published.

Install instructions come from the README

Skills do not use an MCP config. Follow the steps in the README below.

README

synced Oct 22, 2026

DevOps Skills

A small, personal collection of Agent Skills for DevOps / Platform / Kubernetes / GitOps / Go engineering work — Git hygiene, code and PR review, incident and root cause investigation, Kubernetes and Helm review, ArgoCD/GitOps workflows, Terraform/OpenTofu review, Go bugfixing and testing, and clear technical English.

Every skill is a plain-markdown SKILL.md (plus, for a few, a small references/ file). There are no install scripts, no postinstall hooks, and nothing here executes commands on your system on its own — every skill's Safety rules require read-only investigation first and explicit confirmation before anything mutating.

Who this is for

Built for a single engineer's daily stack (Kubernetes, ArgoCD, GitOps, Terraform/OpenTofu, Go), but every skill is self-contained and generic enough to be useful to any DevOps/platform engineer. Compatible with any agent that supports the Agent Skills format: Claude Code, Codex-style agents, CodeWhale, OpenCode, and others.

Installing and updating

Claude Code plugin

In Claude Code:

/plugin marketplace add abevz/devops-skills
/plugin install devops-skills@devops-skills

Or from a terminal:

claude plugin marketplace add abevz/devops-skills
claude plugin install devops-skills@devops-skills
claude plugin list

Restart Claude Code after installation. Skills use the devops-skills: namespace, for example /devops-skills:kubernetes-debug. claude plugin list shows the installed version; compare it with the latest release and CHANGELOG.md.

To update, refresh the marketplace first, then the plugin, and restart Claude Code:

claude plugin marketplace update devops-skills
claude plugin update devops-skills@devops-skills
claude plugin list

Other Agent Skills clients

Install from this public repository with the skills CLI:

npx skills add abevz/devops-skills

The CLI lets you choose skills and target agents. To refresh skills installed through it:

npx skills update

GitHub installs work directly for this public repository. A private fork needs working Git credentials; alternatively, clone it first and install a local path such as npx skills add ./skills/kubernetes-debug.

Maintainer's symlink setup

The maintainer keeps the main checkout at ~/github/devops-skills/main as the source of truth:

# ~/.agents/skills/<name> points to ~/github/devops-skills/main/skills/<name>
# ~/.claude/skills/<name> points to ~/.agents/skills/<name>
readlink ~/.agents/skills/kubernetes-debug
readlink ~/.claude/skills/kubernetes-debug

# pulling updates the linked skill files
cd ~/github/devops-skills/main
git pull --ff-only

Keep this setup when maintaining the repository. Installing the plugin on the same machine would duplicate every skill (for example, kubernetes-debug and devops-skills:kubernetes-debug). Use one installation route per agent.

To link a newly added skill from the main checkout:

cd ~/github/devops-skills/main
ln -s "$PWD/skills/<name>" "$HOME/.agents/skills/<name>"
ln -s "../../.agents/skills/<name>" "$HOME/.claude/skills/<name>"

Check each agent's skill list after setup or a new session: a link on disk alone does not prove that the skill is available to the agent. Symlinks follow the checkout; CLI-managed copies and plugin caches need the update commands above.

Safety-first installation advice

  • Read a skill's SKILL.md before installing it — these are short, plain markdown.
  • Start with selected skills through npx skills add abevz/devops-skills; the Claude Code plugin installs the whole collection.
  • None of these skills bundle install scripts or postinstall hooks. Treat third-party skills as untrusted input first — see SECURITY.md.
  • A few review skills mention optional local CLI tools (e.g. kubeconform, trivy, checkov) in references/tools.md. Nothing installs them for you; skills explicitly instruct the agent not to install or auto-run them.

Skills

For the top-down view — entry-point flows, umbrella↔debug pairs, the DevSecOps chain, and the full reference inventory — see docs/skill-map.md.

SkillCategoryPurpose
git-messageGitWrite Conventional Commits-style commit messages from a diff
pr-reviewReviewLead-engineer-style pull request / diff review
root-cause-analysisInvestigateEvidence-driven investigation of bugs, regressions, and unclear behavior
incident-analysisInvestigateStructured writeup of a production-like incident
production-readinessInfraCheck whether a service/workload/module is ready for production
observability-reviewInfraReview logs, metrics, traces, dashboards, alerts, SLOs
observability-stackInfraOperate the stack: Prometheus HA/sharding, Thanos/Mimir, OTel Collector, Loki/Tempo
architecture-reviewArchitectureReview repo/system boundaries, coupling, and layering
migration-planArchitectureBuild a phased, reversible migration plan
kubernetes-debugKubernetesDiagnose a failing Kubernetes workload, read-only first
kubernetes-yaml-reviewKubernetesReview manifests before they're applied
kubernetes-securityKubernetesSecurity audit of workloads, RBAC, and cluster config
helm-reviewKubernetesReview Helm chart templates, values, and upgrade safety
kubernetes-autoscalingKubernetesDesign/review autoscaling: HPA/VPA/KEDA at the workload layer, cluster-autoscaler/Karpenter at the node layer
storage-debugKubernetesDiagnose PVC/attach/mount/expansion failures and Longhorn/Rook-Ceph/NFS backends, data-safe
cluster-backupKubernetesDesign/review backup & DR: the GitOps gap inventory, Velero, etcd snapshots, restore testing
upgrade-readinessKubernetesPre-upgrade review: skew rules, removed-API scans, addon matrix, drain readiness, staged per-minor plan
argocdGitOpsInstall/operate Argo CD: HA, sharding, SSO/RBAC, AppProject, repos, sync design, backup/DR
argocd-debugGitOpsDiagnose an out-of-sync or degraded ArgoCD Application
argocd-applicationsetGitOpsDesign/review ArgoCD ApplicationSet generators and blast radius
gitops-reviewGitOpsReview a GitOps repo's layout, secrets strategy, and promotion flow
terraform-reviewTerraformDiagnose-first review of Terraform/OpenTofu modules and plans
dockerfile-reviewContainersReview Dockerfiles for size, security, and reproducibility
cicd-reviewCI/CDSecurity and reliability review of pipeline definitions (GitHub Actions, GitLab CI)
ciliumNetworkingDesign/deploy/review Cilium: install, kube-proxy replacement, L3/L4/L7 policy, Hubble, encryption, Cluster Mesh, Gateway/Egress/BGP
cilium-debugNetworkingDiagnose Cilium CNI drops and CiliumNetworkPolicy issues via Hubble evidence
istioNetworkingOperate Istio: install/revisions, sidecar vs ambient, traffic management, mTLS/authz, telemetry
istio-debugNetworkingDiagnose Istio mesh issues: sidecars, mTLS, routing, 503 classification
ingressNetworkingDesign/review north-south traffic: ingress-nginx (EOL), Gateway API, Ingress→Gateway migration
cert-manager-debugNetworkingDiagnose stuck certificates: issuance chain walk, ACME HTTP-01/DNS-01, rate limits, renewal
grafana-dashboardsObservabilityDesign and review Grafana dashboards as code
alert-rule-reviewObservabilityWrite and review Prometheus alert rules (PromQL, noise, coverage)
runbook-writerOperationsWrite on-call runbooks with read-only diagnosis and marked mutations
go-bugfixGoFix a Go bug with a minimal, tested change
go-testing-reviewGoReview Go test quality and coverage gaps
go-code-reviewGoReview Go code for idiomatic style and correctness
english-technical-messageCommunicationPolish short technical English (PR comments, Slack, recruiter replies)
supply-chain-securitySecuritySBOM, image signing, provenance, deploy-by-digest, registry hygiene
admission-policy-reviewSecurityKyverno/Gatekeeper/PSA policy design, audit→enforce rollout, exceptions
vulnerability-triageSecurityTurn scanner findings into ranked actions; suppression discipline
runtime-security-reviewSecurityFalco/Tracee coverage and tuning, runtime alert triage
dast-reviewSecurityZAP DAST scan placement, authenticated/API scans, findings gate
secrets-managementSecurityExternal Secrets Operator / Sealed Secrets / SOPS: deploy, provider auth, rotation
vaultSecurityOperate HashiCorp Vault/OpenBao: HA, auto-unseal, auth methods, policies, secret engines, DR
interview-system-designCareerStructure and practice system design interview answers
homelab-change-planHomelabPlan homelab infra changes with no staging: tiers, snapshots, bail-out

Where to start

  • You know the symptom ("PVC Pending", "certificate not Ready", "Application OutOfSync"): open the quick task index and follow the link.
  • You want to see how skills fit together — entry points, domain boundaries, and the review/investigation loops: sections 1–3 of docs/skill-map.md.
  • Installing only a few: kubernetes-debug, argocd-debug, storage-debug, cert-manager-debug, upgrade-readiness, terraform-review.

Repository layout

.
├── README.md
├── SECURITY.md
├── CONTRIBUTING.md
├── LICENSE
├── CHANGELOG.md
├── .claude-plugin/
│   ├── plugin.json              # plugin identity and version
│   └── marketplace.json         # installable as devops-skills@devops-skills
├── .github/workflows/validate.yml   # CI: Agent Skills, plugin/version, structure, safety checks
├── skills/
│   └── <skill-name>/
│       ├── SKILL.md
│       └── references/        # only where genuinely useful: tool lists, bad→good examples
└── docs/
    └── design-notes.md

See docs/design-notes.md for why these skills (and not others) were chosen and which third-party collections inspired them.

↑↓ navigate openesc close