Agent skill
A skill is a folder with a SKILL.md file that teaches an AI agent a procedure: when to use it (its description) and how to do it. The agent reads only the description up front and loads the rest when a task calls for it.
Select any words here to ask your Maestro about them.
You have 20 skills installed. When a conversation starts, how much of them does the agent read?
What it's for: getting an agent to run a procedure your way without explaining it every time: how you vet a new file, how you build your report, the steps of a reconciliation. You write down your expertise once, and the agent applies it when it's needed.
How it works: each skill is a folder with a SKILL.md file. At the top go a name and a description; below, the instructions. When it starts, the agent reads only the descriptions of all its skills, which takes up very little of its context window. When a task matches a description, it loads that whole skill, along with its scripts and examples if it has any.
The description is the part that matters most: if it doesn't say when to use the skill, the agent won't find it. Skills are an open format used by Claude Code, Claude Cowork and other agents; you can install them in the Claude chat app too.
In Agentic AI for data, you package your own procedure so any agent can follow it.
How it's used
A skill for checking CSVs
.claude/skills/check-csv/
- $ cat SKILL.md
- ---
- name: check-csv
- marked: description: Use when asked to analyze a new CSV, before any calculation.The only part the agent reads up front. This line is how it decides whether to load the skill.
- ---
- 1. Say what a row is and how many there are.
- 2. Run python check.py <file> and report what it prints.A skill can bring its own scripts for the steps that never change.
- 3. Never drop rows without saying which ones and why.
When to look closer
Skill or instructions file?
Rules for a data project
- Amounts are in dollars, including sales tax.
- How to check a new CSV before using it
- Never delete raw data in the data/ folder.
- How to build the monthly report for leadership
Which of these rules always apply, and which only apply to one task?
Ask your Maestro about Agent skill
It answers from the free course page closest to this word, the way it answers while you study.
You: What does "Agent skill" mean, in plain words?
Your Maestro: I'll explain it with a case of yours: it sticks better than a definition. Which procedure at work would you like to write down just once?
Asking about: Answers from the page "What's a folder, and what's a file path?"
Only this browser keeps what you ask here. How it works
Learn more
Related terms
- Agent instructions file
A text file in your project that the agent's harness loads at the start of every session, so the agent knows from the start what it always needs to know: rules, facts about the data and how to check its work.
- Claude Code
Claude Code is Anthropic's AI agent that works on your computer: it reads your project's files, runs commands and makes changes over several steps, asking your permission for what matters. You use it from the terminal, the desktop app or editors like VS Code.
- Context window
Everything a language model has in front of it at a given moment: your instruction, the files it has read and the results of its tools. What isn't in the context, the model can't use.
- Script
A script is a short program, saved in a text file, that does one task from start to finish every time you run it, the same way each time. It doesn't think or guess: it reads what you give it, does its steps and hands back a result or an error.