// the find
dbt-labs/dbt-agent-skills
A curated collection of Agent Skills for working with dbt, to help AI agents understand and execute dbt workflows more effectively.
A collection of context documents and scripts that teach AI agents how to use dbt correctly — writing models, running CLI commands, troubleshooting job failures, working with the semantic layer, and migrating to dbt Fusion. This is an official dbt Labs repo, not a community project, so the content reflects actual dbt internals rather than guesses. Useful if you're already using Claude Code or Cursor on a dbt project and want the agent to stop hallucinating CLI flags.
First-party authorship matters here: the reference docs for unit test edge cases (ephemeral deps, incremental models, versioned models) and warehouse-specific data type caveats are the kind of thing community-built prompts typically get wrong. The eval harness is a real differentiator — there are actual A/B test scenarios with graders to measure whether a skill variant improves agent behavior, not just vibes-based prompt tuning. Multi-client support via the Agent Skills spec means the same skill files work in Claude Code, Cursor, Cline, and others without manual porting. The split between always-on skills (analytics engineering) and one-off migration skills is sensible packaging.
The Agent Skills ecosystem it depends on (agentskills.io) is nascent enough that 'compatible agents' is a short list and the spec could change under you. The skills are read-only documentation injection — the agent still has to execute dbt commands itself, so if your agent can't shell out or doesn't have dbt in PATH, these skills add context but fix nothing. There's no version pinning between skill content and dbt versions; a skill written against dbt 1.8 syntax may silently mislead agents running dbt 1.6 projects. The Fusion migration skills are essentially undocumented externally — if Fusion's behavior diverges from what's in the reference files, you won't know until the agent does something wrong.