Sometimes I have to use Kubernetes.

Everyone hates Kubernetes because you have to write so much YAML, and using Kubernetes is also very complicated.

The good news: You don't have to write YAML anymore. https://github.com/xchangeee/claudernetes/ will put Claude Code right where it belongs: An ArgoCD config management plugin.

Fed up with helm charts? Don't feel like writing YAMLs anymore? Just put a markdown file in your git repo and you're good to go!

#kubernetes #cloudnative #argocd #AI #SRE

@xchange So this is interesting, I like the fact it's clearly documenting the intent of the developers for posterity but the lack of determinism is a red flag for me... I would not be comfortable deploying this in production and having yamls regenerated every time a change occurs.

Is it possible to have it generate the manifests once and only regenerate when the claudernetes.md changes by examining recent git commits?

Actually, now that I think about it... Are the manifests generated out sideof the PR/review process? Cos generating them after a change hits ArgoCD, when it's too late for human review, would stress me out 😬

Maybe all of this anxiety is because I've never worked with ArgoCD CMPs before 🤣

@nicr9 I'm torn. I hope no one actually uses this. But I would also laugh really hard if someone does.

The whole AI frenzy thats going on at the moment is exhausting af, and I came up with this idea on the toilet.

GenAI makes code generation really cheap, so instead of doing a random shitpost on mastodon, I can burn a few tokens to turn my shitpost idea right into a working prototype.

During testing, I couldn't even get it to one-shot a working yaml that doesnt CrashLoopBackOff.

@nicr9 Going further with this, the next obvious step now would be a claude operator, where you put your markdown definition into a CRD, and then the operator can feedback pod status and events back into claude and maybe after a few iterations and burning even more tokens, we get working YAML.

@xchange ya know, if there was a mechanism for the agent starting from a template that's shared across teams like a detailed "policy doc" with yaml snippets that could actually help with the determinism aspect...

I also think that while sticking it at the tail end of an ArgoCD deployment is crazy... There is a lot of teams that actually hate dealing with YAML and if you had a way to shore up the determinism part of it, some of them would actually take this seriously

Oh, I love the CRD idea like a hole in the head 🙃

@nicr9 I think everyone would be better off replacing their YAML with https://cuelang.org/ and a pre-commit-hook that implements the rendered manifest patern.
CUE

Configure Unify Execute Validate, define, and use dynamic and text‑based data Learn more Get started with CUE CUE makes it easy to validate data, write schemas, and ensure configurations align with policies. Get started learning about CUE with these links ..

CUE

@xchange I really need to take a closer look at this...

Years ago I got really interested in Jsonnet and especially how it could be used for building things like Grafana dashboards but after experimenting with the tooling for a while I gave up. The support from Grafana was limited and it was ultimately demanding more in terms of tooling setup, syntax and knowledge of the underlying tech to fix things when they went wrong. Almost like the abstractions WERE the objective instead of making things simpler for the end user to work with and reason about.

I'm not working on dashboards as much these days but if I find a similar problem at least I'll know where to start looking for a solution 😄

@nicr9 Yeah, I've used a lot of jsonnet and monitoring-mixins in my last job. Jsonnet solves boilerplate and config generation, but with jsonnet you still can generate invalid schemas. Cuelang solves all of this in a proper way, but its still a long way for adoption.

I like the idea of dashboards as code, but imho most attempts at grafana over the years have failed.

I havent tried https://grafana.github.io/grafonnet/ myself and in the meantime I just dump large json files in git, if at all.

Home - Grafonnet

@xchange I honestly don’t know if this is sarcasm or not but I can’t imagine wanting to write markdown in hopes that it would produce yaml I wanted.

Seems like a solution looking for a problem to me.

@xchange @lasombra_br I dunno about you - but that sounds like a security nightmare.
@talios @lasombra_br A security nightmare would be to give molt full admin access to the company's AWS root account and and then let the magic happen.
@xchange @talios I don’t believe you even need to give it permission if you have it laying around your environment variables . Haven’t checked though.
@lasombra_br @talios I would like to see kubebook, where autonomous agents battle against each other in the fight over cluster resources. Like a boring version of robot wars.
@lasombra_br @talios If CEOs can be automated away by a markdown file, we can also automate away red teaming.
@xchange @talios Let the battle commence
Claude Code for Infrastructure | Hacker News

@xchange I like your approach to making the world a better place.