I finally made my childhood dream come true
Before I start with describing what I ( Claude ) created, first short intro. This post is not about "New, awesome kubernetes troubleshoot app". This post is a short story, how I finally created something on which I stopped working on long time ago ( of course from zero, because Claude did a way better job then me ).
Tool is open-source and it's on Artifacthub, and for testing I was using kind and groq.
What it does?
Detects pod failures in real-time. An agent runs as a DaemonSet and watches the Kubernetes API for pods entering CrashLoopBackOff, ImagePullBackOff, Pending, OOMKilled, FailedMount, FailedScheduling, and other failure states. When it finds one, it immediately reports it to the backend — no polling, no manual scanning.
Generates AI-powered troubleshooting steps. The backend collects the pod's events, logs, container statuses, and manifest, then sends everything to an LLM. You get a "What's Wrong" explanation and a step-by-step "How to Fix It" guide with specific kubectl commands.
Scans your cluster for security misconfigurations. A separate scanner runs 50+ checks — privileged containers, missing seccomp profiles, containers running as root, dangerous capabilities, untrusted registries, missing resource limits, and more.
These three features are the most important features, but there is also a feature to integrate slack/teams for notifications, user management, monitoring and logging, but not like Grafana + Loki. Monitoring is only for Nodes, and logging to see live logs, and history. But those are all "nice to have features. Maybe latter they will also become "important features".
What I learned?
The thing on which I worked the most and what I learned the most, is not the react, or python, or anything else regarding the "tech part". It was how to work and "configure" Claude, to get better answers. I started spending more and more time writing .md files ( without Claude's help ), which started giving me better and better solutions. I would suggest this to anyone, because every project is unique. Don't spend time copy/pasting others people configuration ( of course paste some parts ), but spend some time refactoring them. Make them to work for you and you'r project/team better.
What next?
I will continue with this project. It's fun I will try to improve this tool as much as I can. For once I want to contribute to open-source community.
Top comments (0)