After hearing about Agent skills, I thought I would try it out and where better to start than Hello World!
Agent skills is a way to define a skill that can be given to an agent and it should work with whatever AI tool you are using as it is an open standard. The SKILL.md details what the agent should do when triggered.
Resources used in this work can be found at
With the source code for what I did at https://github.com/Daveu1983/skills-hello-world
The Agent.md
# Agent Instructions
This repository is a collection of Agent Skills — reusable instructions that AI coding agents can load on demand to perform specific tasks.
## Repository structure
skills/<skill-name>/SKILL.md # Skills (Agent Skills format)
AGENTS.md # This file
## Skills format
Skills follow the [Agent Skills](https://agentskills.io) open standard. Each skill is a directory containing a `SKILL.md` file with YAML frontmatter and Markdown instructions.
## Adding a new skill
1. Create a directory under `skills/` named after your skill (lowercase, hyphens only).
2. Add a `SKILL.md` file with valid frontmatter (`name`, `description`) and clear instructions.
3. Keep instructions agent-agnostic — avoid referencing tools specific to a single agent product.
4. Optionally add `scripts/`, `references/`, or `assets/` subdirectories for supporting files.
The hello-world SKILL.md
---
name: hello-world
description: "Generates a working Hello World application in a user-chosen programming language. Asks the user to pick a language, then returns correct source code that prints \"Hello, World!\" to the console. Handles both compiled and interpreted languages correctly."
---
Ask the user which programming language they would like a Hello World application in. Offer a selection of common languages (e.g. Python, JavaScript, TypeScript, Rust, Go, C, C++, Java, C#, Ruby, Swift, Kotlin, Zig) and allow them to specify any other language.
Once the user has chosen a language:
1. Determine whether the language is **compiled** or **interpreted**.
2. Produce a complete, minimal source file that prints exactly `Hello, World!` to standard output (console).
3. The code must be correct and complete — it should compile (for compiled languages) and run without any errors on a standard toolchain with no extra dependencies.
4. Determine the conventional filename for a Hello World program in that language (e.g. `main.go`, `hello.py`, `Main.java`).
5. Write the source code to that file in the root of the current project using the Write tool (or equivalent file-writing tool).
6. Confirm the file was created and show the filename to the user.
7. After creating the file, provide a short "How to run" section that shows the exact shell commands needed to:
- **Compiled languages**: compile the source file and then run the resulting binary.
- **Interpreted languages**: run the source file directly with the appropriate interpreter/runtime.
Keep the code idiomatic and minimal — no unnecessary boilerplate beyond what the language requires for a valid runnable program.
Testing it out
When running on Gemini with the instruction run hello-world skill, it initially attempted to run a skill I assume you can set up for Gemini to run globally on your system, when it could not find the skill it asked if I wanted to run the local one it found. After confirmation of this it found the skill.
Through the thought process it decided that I should just be given four options of what language to create a Hello World Application for with the fifth option being another language, this was not prompted by myself and not part of Agent.md or SKILL.md.
I chose Go and it gave me a source code file and instructions to run that were successful.
When running on GitHub CoPilot it found the skill straight away and gave me a free format question with what language I would like it to create the Hello World application for. I typed in Java and it produced a working source code file and instructions to run that were successful. Interestingly the instructions and source code that was given were out of date with new versions of Java, since Java 11 you can run in one line with 'java Main.java', and since Java 25 (preview from 21) you can use the code
void main() {
System.out.println("Hello World!");
}
CoPilot gave me
The below shows the chat with CoPilot to generate this
Conclusion
Was this the best and most sustainable use of AI probably not! But the results across the two agents I tested were consistent, and working code was produced. Skills are being used widely see https://agents.md/#examples for more details and only likely to become more and more sophisticated in the future.
Top comments (0)