Intro
Anthropic came and published their skills documentation and what they are roughly 2 months ago, but I suspect they and other AI companies were using them internally even before they published. The idea of people seeing them as a new thing and being astonished is abysmal..because we were already using it..but without the naming convention and the context limiting. The concept is simple: you create a folder named the skill, then inside it, the main file is called skill.md. This takes less context; you tell the AI to scan all skill folders for skills that it can use ad hoc. The problem it solves with MCP is that this uses way less context than others and only triggers when necessary.
Example: PlantUML Generator
So the idea is simple; as you can infer from what I was saying, let’s go with an example. I wanted to create an architecture diagram in the office and have to update them, ..in office we are using the PlantUML to create sequence diagrams.
The problem is you have to ask the AI to generate it, and then generate the image and maintain the image somewhere. The problem with this is it takes time and is a manual process. How can we use skills to automate this? We can write instructions to ask the AI chat / agent to be like, if a user asks x, then do the process and update the diagram. We can add these instructions in the skill.md file, keep the name and descriptions to work with the architecture diagram.
You can write scripts and then use them to generate things; the AI will execute them based on the instructions. The next is dependencies, if any. Since PlantUML is an open-source project, so we can download the jar file and put that in the skills folder. It’s byte code and it’s platform independent.
So everything can be packaged into a single skill.
you can see the skills folder here: https://github.com/aghontpi/dot-configs-ai/tree/main/skills/
Embedding Code in Images
On top of the skill, I was thinking about for a optimization. In college, I conducted a Capture the Flag; the event was like Google CTF and other CTF events from that time. I was inspired by them, Anyway. one of the competitions was to find the hidden details from the image. Inside an image, you can add a data to it, it can be any data. So I was thinking, why can’t we use the same thing here? So I asked ai to write a script to embed code into the images. So the PlantUML diagram is stored inside the images, the images are the output of the code. It’s like a chicken and egg problem once you delete the code file from the repo.
We want crud options on this; the update image, edit image, and create image commands are just a set of instructions that the AI can read and execute. Python scripts, when wanted to edit / asked to edit, it will extract the code from the image and create a temporary file we can work on, and when we ask to store it, it saves it again inside the image, deleting the code file again.
Other Skills
There are other skills that I use here, one skill that I use to review code. It has custom code review rules.. it keeps changing overtime.
The main one I wanted to share is the PlantUML generator; it has all the features that I need to be adding to demonstrate the features of the skill.
Sub-agents & Lifecycle Hooks
At the end of Feb 2026, the Claude team updated the skillset to have custom features, like sub-agents. What it is, is that the sub-agent is being called by the main agent when doing a certain task.. so it will fix the context problem. it’s one of the major updates to skills; it’s the major one that makes a significant difference, because it won’t clutter the main context; the main agent is not interfered with.
It also has access to the lifecycle events, Lifecycle hooks (on_start, on_success, on_error, PreToolUse, etc.).. but i haven’t adopted them yet; but this will help me with the cryo vault project: https://github.com/aghontpi/cryo-vault
Personal update
I’m having to juggle a lot of things in my life currently; finding focus time to think on what is essential or next is hard with this attention-seeking world; socializing; maintaining health and ad hoc new plans; argh! I started to have a new habit; hopefully, when I am consistent on this.. I’ll post on this later!.. future me..! what is happening? hope you are good!