How a skill works

Before we look at using the dagster-expert skill, it helps to know what a skill is and how the agent uses it.

What skill files look like

A skill is a structured document (often a SKILL.md or similar file) that your coding agent can load when it sees a relevant trigger, such as typing /dagster-expert or mentioning "Dagster" in the prompt. The skill file typically includes:

• When to use the skill: triggers, slash commands, or intent phrases (e.g. "create a Dagster project", "add a schedule").
• Instructions: what to do first (e.g. use dg for scaffolding, run dg check defs after changes).
• References: pointers to CLI commands, project layout, and patterns (e.g. use uvx create-dagster, put assets under defs/assets/, use DuckDBResource from dagster-duckdb).

The agent doesn't "run" the skill like a script; it reads the skill as context and uses that to decide which commands to run and how to structure code.

How skills are used

When you invoke the skill (e.g. /dagster-expert create a new Dagster project called university):

1. The interface or agent matches your prompt to the skill (e.g. "dagster-expert" or "Dagster project").
2. The skill's content is added to the context sent to the model.
3. The model follows the skill's instructions: use dg and the recommended layout, run dg check defs after scaffolding, add dependencies with uv, etc.

So the same high-level request ("create a project", "add three assets") leads to different, more opinionated behavior when the skill is active: the agent prefers dg over hand-written structure and validates with dg check defs. In Lesson 4, we'll see that in practice by creating the project with the skill.