Article written by Malo Lecoursonnais, July 6, 2026
AI agents, in whatever form they take, are reshaping the world of software development more than any other field. Their value is no longer in question when it comes to modern code — this blog series aims to give you every key you need to bring that same revolution to the legacy world, and to IBM i in particular.
So what exactly are skills?
AI agents respond to your questions, your prompts. In a software development context, that means sending your lines of code to coding assistants: IBM Bob, Claude Code, GitHub Copilot, Mistral Vibe, ChatGPT Codex…
But an application estate is also made up of company-specific practices and CI/CD pipelines designed by teams. Passing on those processes to AI is exactly what skills are for.
How skills work
Adding a field to a physical file, refactoring monolithic code, running regression tests… the DevOps tasks that are increasingly getting AI assistance tend to follow the same pattern, built on established best practices.
A skill is a scenario the agent needs to follow, giving it structure. In practice, it lays out the steps to take and the tools to use so nothing gets missed on the task at hand.
What makes up a skill
A skill takes the shape of a folder containing markdown files (similar to .txt):
my-skill/
├── SKILL.md # Main instructions (required)
├── checklist.md # Checklist for the agent to follow
├── examples/
│ └── sample.md # Example output showing the expected format
└── scripts/
└── validate.sh # Script the agent can run
The first file, and the most important one, is SKILL.md. It contains the main instructions for the scenario, points to the other markdown files, and includes a short description in its file header. Here’s an example of description:
--- name: arcad-add-field description: Interactive workflow to add a field to an IBM i file. Will ask for field details, analyze ARCAD dependencies, present impact report, and implement changes only after user approval. ---
The other files support the main one, so the end result matches the technical or business need as closely as possible.
Getting the best skills you can
While the principles behind skills are well established, it’s hard to say that any one skill is definitively “the best.” A good skill is simply one that best captures your own development processes.
In the IBM i world, skills matter even more, since LLMs are known to be less trained on RPG and COBOL code.
With that in mind, and drawing on 34 years of expertise, ARCAD Software’s MCP Server offers a library of skills written and tested by its own IBM i experts.
This article is part of the AI on IBM i series:
- AI on IBM i: demystifying MCP, the toolkit for agents
- AI on IBM i: turn your expertise into skills
- AI on IBM i: agentic AI for modernization, ARCAD for reliability (coming soon)
- AI on IBM i: a real-world use case of modernization made faster and safer (coming soon)
About the author
Malo Lecoursonnais
Solutions Architect
Malo holds a degree from IMT Atlantique and specializes in the integration of AI solutions for IBM i and modern environments. As Solutions Architect at ARCAD, he supports our clients in modernizing their application portfolio, from design through to implementation, with a relentless focus on user experience.
His expertise with AI tools and his product vision feed directly into the evolution of Gianni and ARCAD MCP Server, helping IBM i teams regain mastery of their code. Drawing on international experience and a passion for teaching, he helps spread best practices in AI development internationally.
REQUEST A DEMO
Let’s talk about your project!
Speak with an expert