OpsKit
Setup

Install OpsKit in one command

No download, no zip, no global install. OpsKit writes everything into your project's ./.claude directory and wires itself into Claude Code.

Requirements

  • Node.js 18+
  • python3
  • kubectl, already pointed at the cluster context you want to read
  • Claude Code, since OpsKit installs as agents, skills, and commands for it

Run the installer

zsh
$ npx github:shahneil76/opskit init

It asks which kubeconfig context to read, then installs 34 skills, 42 agents, 47 commands, builds the read-only Kubernetes MCP connector, and installs the opskit-guard safety hooks — all scoped to the current project.

What a clean install looks like

expected output
Checking prerequisites
  ✓ Node.js v20.11.1
  ✓ python3
  ✓ kubectl

Connecting
  ✓ context: production-eks
  ✓ API server reachable

Installing
  ✓ 34 skills, 42 agents, 47 commands
  ✓ safety layer installed (opskit-guard, MIT — auditable)

Building the cluster connector
  ✓ 15 read-only tools ready
  ✓ connector and hooks wired into .claude/settings.json
  ✓ CLAUDE.md created (fill this in — biggest single quality win)

Fill in CLAUDE.md before your first investigation

The installer scaffolds a CLAUDE.md with sections for clusters, key services, runbooks, escalation paths, and your observability stack. OpsKit reads it before every investigation — accurate context here is the single biggest improvement to how good the output is, more than any prompt.

Verify it worked

zsh
$ opskit doctor

Then set your cluster context and try it against a real pod: export OPSKIT_CONTEXT=your-cluster-context-name, open Claude Code in the project directory, and run /k8s-triage <pod name> is failing.

Limitations

The install is per-project, not global — each repo you want investigated needs its ownnpx github:shahneil76/opskit init. And OPSKIT_CONTEXT must be set correctly before every session; OpsKit will read whatever context it's pointed at, including the wrong one.

Ready to point it at your cluster?

Get instant access — Free