Lets start with the blog

August 19, 2026

If it's worth doing, it's worth overdoing

Before we even start with the blog, we have to go back a few steps.

I have worked with Claude Code for a quite a long time but never really touched the cli interface. Since I'm now more than unemployed and have plenty of spare time, I might as well dive into it.

Security, security, security

I really don't have much to say about security but there was something I knew for sure. If i'm running some LLM straight through the cli (and probably giving it way too many permissions) I have to do it in a virtual machine.

The fact that I have no real experience, if Claude cli can break out of the vm, I still decided to follow my intuition. If it somehow could, my little personal blog project would be the smallest problem to worry. There would be bigger ones in the world for sure.

So first we setup a VMWare Fusion, which is by the way free for personal use nowadays. I have never touched it in my life and have always used Oracle's Virtual Box, which i've always hated. It's like you had been drinking only cheap lager for your whole life and somebody passes you a pint of Guinnes for the first time. Everything just works.

So what about the blog?

We setup the vm with Ubuntu (yes I know, but I'm just a software dev), install Claude and start working (prompting).

I wanted everything to be as lightweight as possible so we endup with a static site generation (f*ck you javascript) using Eleventy to write blog posts with markup. Simple and effective.

Since I'm using the vm to build the project (pls Claude stay there), I needed a way to write the markdown files on my host machine. I stumbled upon a few different markdown apps and they were all sh*t.

Sidenote: Actually I noticed something I've been wondering for a while since the LLM's are becoming more and more common. When I asked Claude what markdown programs are available, it gave me 5 options from which two first were purchasable. I've been wondering how the SEO stuff will translate to the LLM visibility optimization. Maybe I'll deep dive into that some time.

I ended up using Visual Studio just manually editing files and rendering the markdown straight on the site. The setup looks like this:

To get the workflow smooth, I created a read-only shared folder from the host to the vm, symlinked it to the blog repository and setup a file polling. Since git can only see the symlink, we had to create a prehook for the git commit and sync physical copies of the blog-posts files. Properly over engineered!

  eleventyConfig.setChokidarConfig({
    usePolling: true,
    interval: 500,
  });

To get the images flowing we (me and Claude ofc) setup a passthrough for the images.

  eleventyConfig.addPassthroughCopy("src/images");
  eleventyConfig.addPassthroughCopy("src/posts/**/images/**");

Everything seems to works now but we still had to make a github repository, assign a deployment ssh key for the repository (not for the whole account you dumb), so that Claude can push our code to it (and only to it).

Last step was to wire up the github repository to Vercel (yeah I know, I should self host this sh*t) and we are good to go!

So I thought... After realizing how the symlink actually works and how I can't really use it with the actual git directory, we decided to take a slightly different approach. We got rid of the symlink completely and created a custom rsync script that runs every second with nodemon, syncing the read-only host folder to the git working directory. Even more properly over engineered!

At this point everything was ready, untill I realized that the mardown doesn't really render properly... After some digging I found out that it actually does everything it's supposed to do. Eleventy just doesn't come up with any specific js or css for the markup conversion. Just plain old html. Purrfect. We will keep it like that, for now at least.

Till next time.