DevOps AMS 2026 Demo

This demo shows how to use kagent to diagnose and fix a problem with a k8s deployment.

Prerequisites

• A Kubernetes cluster (kind, minikube, or remote)
• kagent installed and configured (kagent.dev)
• GitHub MCP server configured in kagent
• A GitHub repo for the demo (e.g., peterj/devopsams-2026)

1. Create the Github MCP server

source .env

kubectl create secret generic githubsecret -n kagent --from-literal=token=$GITHUB_TOKEN

# Create mcp server
kubectl apply -f github-mcp.yaml

2. Create the agent

kubectl apply -f agent-with-skills.yaml

3. Deploy the broken app

kubectl apply -f manifests/namespace.yaml
kubectl apply -f manifests/checkout-broken.yaml

Verify it's crashing:

kubectl get pods -n demo-store
# Should show CrashLoopBackOff or OOMKilled

Demo

Part 1 — Incident Triage

Prompt the agent:

"The checkout-service in the demo-store namespace is down. Can you figure out what's wrong and fix it?"

What the agent will do:

1. kubectl get pods -n demo-store → sees CrashLoopBackOff
2. kubectl describe pod <name> -n demo-store → sees OOMKilled in events
3. kubectl logs <name> -n demo-store → sees stress output / killed
4. Identifies memory limit (64Mi) is too low for the workload
5. Suggests increasing memory limit to 512Mi
6. Applies the fix

Part 2 — Skill: Post-Incident Checklist

Prompt the agent:

"Can you run the post-incident checklist?"

What the agent will do:

1. Verify checkout-service is now Running with 0 recent restarts
2. Check other deployments in demo-store for similar resource risks
3. Generate an incident summary with root cause & resolution

Part 3 — GitHub MCP Server

Prompt the agent:

"Create a GitHub issue documenting this incident"

What the agent will do:

1. Use the GitHub MCP server to create an issue
2. Title: "Incident: checkout-service OOMKilled in demo-store"
3. Body: the incident summary from Part 2

Reset Between Runs

kubectl delete -f manifests/checkout-broken.yaml
kubectl apply -f manifests/checkout-broken.yaml