I swear this post is not made because I basically have zero thing to do at 4AM. I swear that it's also not because I accidentally drank coffee at 1AM that resulted me being in a state of "tired and need to sleep" "wondering my life decisions so far" and "in constant worrying if I'm going to be alright at office tomorrow, I mean today". Yeah, definitely.

So, as a way for me to cope, I'll talk a bit about a little project of mine, Tsukuyo. Now I know what you're thinking, "Ugh, another weeb-ish sounding name", but honestly that's one of the only way that I can motivate myself into making something. And that's definitely because of the growing guilt after having an entire 1-2 months of Github Copilot subscription wasted because not only I didn't do much coding, I realized that I basically had about 500 premium requests per month which I didn't use at all. So yeah, with that fact in mind, I tried to make something "useful", at least for myself, using the ~~wasted money~~ privilege of having a Copilot subscription. And it turned out pretty cool!

Tsukuyo CLI - Lite Config Inventory Manager

Let's not go into the detail on why the name "tsukuyo" was chosen, because it's one of the lamest reason ever. So let's go with the motivation.

So yeah... uhh... motivation... Yeah...

Am I someone that would have "motivation" to begin with? I don't even know tbh. The motivation really came because I had too many moments of asking myself whether a certain config value is correct, whether I have an information of a certain thing, and some random things along the line. It's mostly caused by these weird events or cycles.

Wanting to do SSH tunneling but forgot which port I am targeting on.
Wanting to do SSH to a node but I forgot the details of the node.
Wanting to simply login into a database instance, but let alone knowing the credentials, I even forgot if it were exist in the first place.
And so on...

The problem was mostly caused by too many different setups for these resources. For instance, my VM has databases running in Docker containers and also running natively. I really had some bad moments because I once tried to create do database migration, but I targeted database inside a Docker container instead of the one that's running natively. And the fucked up part is that I set the port forwarding of the Docker container running Postgres with the same port in the host 5432:5432 . No wonder I didn't able to even find my database definition inside Postgres running natively. For unknown reason there weren't colliding port error when starting the service, which, yeah, kinda weird. I solved that problem by changing the port of the native Postgres service to something different, like 5433 because yeah...

It was mostly a fault on my memory, but to think again, "why cant I have a better way to properly document my inventories?", which the grew into "why don't I create an easier way to not only document about my resources, but also use it along the line?", which resulted in me initiating this project.

So, with the problem statements in mind, and some questions that follow along, I wanted to create a CLI tool that:

Able to save arbitrary item definition and do manipulation into it (the usual CRUD)
Able to use the data to perform other task, like SSH-ing, tunneling, or even turning it into environment variables used for running something.
Able to be queried using "easy" notations (which later I used jq as my main reference for the query style)

So, how did I make this? It sounds pretty complex, although it looks "easy".

Implementations

I'll going straight into mentioning that I'm using Golang as my language of choice. The reason? I have seen a lot of CLI tooling that were written in Golang, which I kinda want to jump into the bandwagon as well. But really, some CLI packages that I used, like cobra and promptui were really helpful that it actually made my CLI looks cool! But I actually only know cobra by the time I started the project, and promptui was entirely out of my radar. It was thanks to Copilot that I managed to know about the library and used it within my tool.

Talking about Copilot, I was someone that kinda against the rising of AI because as someone that also do stuffs in the creative industry, I see AI as a threat. But again, that's only a mindset issue. After looking, thinking, and trying, I managed to flip my mindset into thinking that AI is more into a "convenient tool" to really speed up my workflow. And with the data used for training being sourced mostly from public repositories, I have made peace with the "ethicality" problem it has.

Though, that doesn't mean that AI usage, specifically generative AI, in creative industry should be justified. They, people who claimed to be an "AI artist", don't even aware about the concept of copyright, as how it works between coding and creative contents are entirely different. No attribution, no respect, and the result is just bland. Not only it made some mediocre tasteless and lifeless images, it also spit upon the original creator that actually need the support. Those "AI artists" can just rot in hell.

Ahem... that someone went somewhere different.

But yeah, using Copilot to bootstrap this was a great step because things that need some tedious preparation and setups can be done by Copilot, which gave me something to work at. From that on, I continue iterating the code while keeping it in check with my original intention. Some parts might be generated differently, wrongly, or poorly, in which I think the human intervention is there and needed. Much less of a "vibe coding", but more into "controlled-vibe coding". I mean, hey, at least I know what I'm doing :D

The easiest way for me, at least, into crafting a good prompt, is by writing down what you want into a text file or a markdown document. I explained about my idea, my intention, my plan, and features that I want to implement just like how I would write a README.md file. Then by using the Agentic mode of Copilot in VSCode, then using Claude Sonnet 4 for the model, I simply said

Follow the instructions inside DIRECTIVES.md file

And somehow, after so many things happened inside the Copilot Chat, it managed to create some good stuffs. Though, it's worth noting that the model can still hallucinate and I'll need to reorient it into the correct way. In the end, it costed me several prompts in order to properly have a result that's just good enough.

I won't really talk much about how is the implementation be done. Definitely because I don't know about the code being generated, yeah, definitely. :D

Alright, to be brutally honest, I don't know about the implementation details because sometimes it's just something that's either so obvious or so simple or already being implemented so many times that it's not original anymore. But at least I know about the flow of the data, how the processing is being done, and some other things that worth mentioning like performance optimization, achieving "arbitrary data structure", and other things that I suddenly forgot while writing this. I do think that as you step up your game in the programming world, the contents of the implementation is growing less relevant, and the overall view, like the design pattern, data flow, and "decisions" that matters the most. After all, some ultra-specific details like how do you implement recursion inside Golang, would probably be sourced from StackOverflow when it was pre-GPT era. It's just how we changed our cartels.

Result

When talking about the result, I can only point into the repository which the project is being hosted. It's far from done, as I still have some "laundries" that I want to take care of. But it's already usable to the point that I'm using it myself. Will this be useful for other people? If they have a very ultra specific workflow that resembles mine, yes. Other than that, it's mostly a toy project for self-satisfaction.

https://github.com/arung-agamani/tsukuyo

I am particularly happy on how it able to do jq -ish query. It ended up like something that resembles Redis, but with support for more complex objects. I mean, hey, having Redis but JSON is cool, right?

I'll still need to improve it so that it's more usable towards the general audience, and potentially developing this into something that's even more awesome. Like I am thinking of making a centralized configuration server which can be used for syncing the data between machines. And maybe, just maybe, I'll have something like Doppler CLI. I think that would be awesome.

So, what about the name?

So, Tsukuyo is the name of a character from Blue Archive, Oono Tsukuyo. She's a part of Ninjutsu Research Department, alongside Izuna and Michiru. Why did I pick her as the name? Because I already have ongoing projects for Izuna and Michiru :D

It is that lame, honestly.