Member-only story
Beyond Prompting: Using Agent Skills in Data Science
How I turned my eight-year weekly visualization habit into a reusable AI workflow
In my last article, I shared how to use MCP to integrate LLMs into your full data science workflow. I also briefly mentioned another helpful tool: skills.
A skill is a reusable package of instructions and optional supporting files. It helps AI handle a recurring workflow more reliably and consistently. At a minimum, it needs a SKILL.mdfile containing metadata (name and description) and detailed instructions for how the skill should work. People often bundle it with scripts, templates, and examples for standardization and accuracy.
At this point, you might be wondering why we use skills instead of just writing the whole thing directly into the Claude Code or Codex context. One advantage is that skills help keep the main context shorter. AI only needs to load the lightweight metadata at first—it can read the remaining instructions and bundled resources when it decides that the skill is relevant. You can find a great public collection of skills at skills.sh.
Let me make the idea more concrete with a simple example.
