check-services

name	check-services
description	Health-check lies.exposed services on the production Kubernetes cluster. Detects CrashLoopBackOff, high restart counts, ERROR log patterns, and queue backlogs. Trigger keywords: check services, service health, pod health, is api running, queue backlog, worker stuck, agent down.

check-services Skill

Diagnose lies.exposed services on the production Kubernetes cluster.

Variables

NAMESPACE=liexp
SERVICES=api,agent,worker,ai-bot,web,admin
LOG_TAIL=300

Step 1 — Pod status overview

kubectl -n liexp get pods -o custom-columns="NAME:.metadata.name,STATUS:.status.phase,READY:.status.containerStatuses[0].ready,RESTARTS:.status.containerStatuses[0].restartCount,AGE:.metadata.creationTimestamp"

Flag if:

• STATUS != Running
• READY = false
• RESTARTS > 3

Describe flagged pods:

kubectl -n liexp describe pod <name> | grep -A5 -E "State:|Last State:|Events:|Reason:|Exit Code:"

Step 2 — Error logs for flagged services

kubectl -n liexp logs -l app=<service> --tail=300 --since=15m 2>&1

Scan for: ERROR, error, exception, FATAL, unhandled, stack traces.

Step 3 — Queue backlog check (api service)

kubectl -n liexp port-forward svc/api 3010:3010 &
curl -s 'http://localhost:3010/v1/queues?status=pending' | jq '.total'

Flag if pending queue count > 50 (possible ai-bot jam).

Step 4 — Synthesize & report

╔══════════════════════════════════════╗
║    lies.exposed service health       ║
╚══════════════════════════════════════╝

Service     Status    Ready  Restarts
──────────────────────────────────────
api         Running   true   0        [OK]
agent       Running   true   0        [OK]
worker      ...
ai-bot      ...
web         ...
admin       ...

── Errors in last 15m ────────────────
<none | list with timestamps>

── Queue backlog ──────────────────────
Pending jobs: N  [OK if <50 | ⚠ if ≥50]

══════════════════════════════════════
Overall: [HEALTHY | DEGRADED | CRASHED]

Step 5 — Suggest action (if not HEALTHY)

Overall	Suggested action
DEGRADED	Show error lines. Ask user if they want full logs.
CRASHED	Show last state + events. Ask before restarting.

Do NOT restart automatically. Always confirm with user first.

Kubeconfig note

Use default kubeconfig. Production namespace is liexp.