Talking with Carlos and Shiv, maybe we should explore the option to migrate to [vocs](https://github.com/wevm/vocs) for a more lightweight and out of the box llm files generation
