The Problem
Working with dbt isn’t exactly a walk in the park for AI agents—or developers, for that matter. Anyone who's tried automating dbt workflows knows the pain: agents misinterpret commands, fail to account for dbt's endless quirks, or just outright choke on anything beyond the basics. Add migrations, semantic layers, or troubleshooting into the mix, and you've got yourself a recipe for frustration. dbt-agent-skills exists to fix that.
What This Does
This repo is a collection of pre-built "Agent Skills" specifically tailored to help AI agents execute dbt workflows without blowing up or making a mess. Think of it as a cheat sheet, but for agents. The skills/ directory contains folders like dbt and dbt-migration, each packed with SKILL.md files, references, and scripts designed to guide agents through tasks like adding unit tests (skills/dbt/skills/adding-dbt-unit-test) or migrating dbt projects (skills/dbt-migration/skills/migrating-dbt-core-to-fusion).
On top of that, the evals/ folder includes testing tools—like the evals/src/skilleval/cli.py script—to validate these skills in various scenarios. Want to make sure your agent can troubleshoot job errors? Run the tests in evals/scenarios/dbt-job-failure and see if it actually gets the job done instead of throwing cryptic error messages.
And for those who care about CI/CD (which should be everyone), there’s a codeowners-check.yml workflow in .github/workflows/ to keep your contributors in line. Because who doesn’t love a good pull request template?
Real-World Use
Let’s say you want your AI assistant to add a unit test for a new dbt model. Without these skills, you’re probably looking at a lot of back-and-forth with your bot as it struggles to understand the difference between an ephemeral model and a regular one. With dbt-agent-skills, you just install the adding-dbt-unit-test skill, and the agent loads the SKILL.md instructions, uses the examples in references/, and generates the right test. Need to verify? Run evals/src/skilleval/cli.py to test its output against predefined scenarios.
The Bottom Line
If you're working with dbt and have AI agents involved, this repo is worth a look. The skills are well-documented, the structure is logical, and the testing tools are solid. That said, the project is clearly geared towards teams who want to go deep with dbt automation—probably overkill for smaller projects or casual setups. Still, if you’re tired of babysitting your AI, this is a step up.