feat(core): add cluster health check tool

[ropatil010]

Hi Team,

Can you PTAL on this.

PR about:
Implements comprehensive cluster health check tool that examines:

• Cluster operators (OpenShift)
• Nodes (readiness, schedulability, resource pressure)
• Pods (failures, crash loops, image pull errors, high restarts)
• Workload controllers (Deployments, StatefulSets, DaemonSets)
• Storage (PVC status)
• Recent warning events

Features:

• Text and JSON output formats
• Verbose mode for detailed diagnostics
• Configurable event checking
• Clear severity indicators (critical/warning/healthy)

[ropatil010]

Output format:

===============================================
Cluster Health Check Report

Cluster Type: OpenShift
Cluster Version: 4.20.0
Check Time: 2025-11-03T17:33:34Z

Checking Cluster Operators...
✅ All cluster operators healthy (34/34)
Checking Node Health...
✅ All nodes healthy (6)
Checking Pod Health...
❌ CRITICAL: 1 pod(s) in failed/pending state

• openshift-xxxx/nodename [Failed]
  Checking Workload Controllers...
  ✅ All deployments healthy
  ✅ All statefulsets healthy
  ✅ All daemonsets healthy
  Checking Storage...
  ✅ All PVCs bound
  Checking Recent Events...
  ✅ No recent warning events

===============================================
Summary

Critical Issues: 1
Warnings: 0

❌ Cluster has CRITICAL issues requiring immediate attention

[matzew]

so this only is OCP?

[manusa]

I'm not really sure about this one, it seems to be quite opinionated.
Given the raw tools, the LLMs should be able to perform the operations themselves.

In https://github.com/Flux159/mcp-server-kubernetes they're implementing similar functionality via a prompt. Which doesn't seem like a bad idea.

Similarly, in https://github.com/GoogleCloudPlatform/kubectl-ai (AFAIU) they're leveraging the system prompt for this purpose: https://github.com/GoogleCloudPlatform/kubectl-ai/blob/main/pkg/agent/systemprompt_template_default.txt

IMO if we want to proceed with this feature we should either:

1. Reimplement it as a prompt for the core toolset.
2. If we really want this as a tool, add it to a specific opt-in toolset diagnostics or something similar.

This is also a good case to test evals and see if they can be used to make a better decision on how to implement this feature.