dbt-labs/dbt-agent-skills

dbt‑Agent‑Skills – AI‑powered helpers for dbt

What it is – A library of Agent Skills that let large‑language‑model agents (Claude, Cursor, OpenAI, Gemini, etc.) perform dbt‑related tasks automatically. Each skill is a small bundle of prompts, scripts and metadata that the agent can load when a user’s natural‑language request matches its purpose.

Why it matters – dbt (data build tool) is the de‑facto way teams build, test and document analytics pipelines. By exposing dbt actions as AI‑agent skills, developers can ask an LLM to "add a unit test to my model" or "configure the dbt MCP server" and the agent will run the appropriate dbt commands, generate SQL, or edit files without the user writing the low‑level CLI syntax.

Key parts

• Skill bundles – dbt (day‑to‑day analytics engineering), dbt‑migration (one‑off upgrades), and dbt‑extras (e.g., Mermaid DAG diagrams).
• Installation – via Claude Code, the Vercel Skills CLI, Tessl, or as a regular dbt package (dbt deps). The repo publishes three independent dbt packages that can be added to a project's packages.yml.
• Supported agents – any AI product that implements the Agent Skills spec, including Claude, Cursor, GitHub Copilot, Gemini, OpenAI, and the dbt Wizard CLI.
• Skill catalogue – examples include:
  • using-dbt-for-analytics-engineering – build/modify models, debug, explore sources.
  • adding-dbt-unit-test – generate TDD‑style tests.
  • building-dbt-semantic-layer – create MetricFlow metrics/dimensions.
  • working-with-dbt-mesh – manage contracts, versions, cross‑project refs.
  • troubleshooting-dbt-job-errors – diagnose platform failures.
  • Migration skills such as migrating-dbt-core-to-v2.

How it works – Skills are not manual slash commands. Once installed, the agent watches the user’s prompt; when it detects a match, it automatically loads the relevant skill and executes the defined workflow (prompt → script → dbt CLI call → response).

Getting started – Install the marketplace entry (/plugin marketplace add dbt-labs/dbt-agent-skills) and then the desired skill set (/plugin install dbt@dbt-agent-marketplace). For non‑Claude agents, use the Vercel CLI (npx skills add …) or Tessl (tessl install …). After installation, set ai_provider in dbt_project.yml to tell dbt which agents should load the skills.

Who should use it – Data analysts, analytics engineers, and platform engineers who already work with dbt and want to speed up routine tasks by speaking to an LLM instead of typing CLI commands.

License & community – Open‑source under the repo’s LICENSE. Contributions are welcomed via pull requests; the project maintains issue trackers, discussion forums, and an evaluation harness for A/B testing skill variations.