dbt Agent Skills

Summary

An open-source collection of Agent Skills from dbt Labs that teach AI agents how to execute dbt workflows. Skills are organized into analytics engineering (model building, testing, semantic layer, dbt Mesh, job diagnosis, MCP configuration) and one-off migration tasks (Core to Fusion, cross-platform). The key insight: skills are not slash commands but natural-language-invoked capabilities that agents load automatically when prompts match use cases. All follow the Agent Skills specification for cross-agent portability (Claude Code, Cursor, Cline, Copilot).

Why This Was Bookmarked

“dbt agent skills. Good for my MG work. May adapt or compare to our existing agentic tooling.”

Direct relevance to 01-projects/phdata/index consulting work. The skill architecture here — folders containing scripts, assets, and data rather than just markdown — mirrors how we think about 06-reference/concepts/skills-as-building-blocks. Worth evaluating whether to adopt these directly for client dbt work or use the patterns to strengthen our own skill library.

Key Ideas

• Skills are folders, not files — they contain scripts, reference snippets, data, and configuration alongside the markdown instruction
• Nine skill categories: model building, unit testing, semantic layer, natural language querying, dbt Mesh governance, job failure diagnosis, MCP server config, doc lookup, CLI execution
• Installation via marketplace: Claude Code marketplace, Vercel Skills CLI, or Tessl package manager
• Apache 2.0 licensed — free to fork, adapt, and redistribute
• Agent Skills spec: a portable format ensuring skills work across different agent products

Connections

The folder-based skill architecture validates our approach in 04-tooling/2026-03-29-infrastructure-decisions. Our skills already follow this pattern — markdown plus scripts — but dbt Labs’ 9-category taxonomy is a useful framework for auditing completeness.

The MCP server configuration skill is particularly interesting given our existing MCP integrations. Compare to 06-reference/2026-03-31-block-hierarchy-to-intelligence for how skills compose into higher-order capabilities.

Open Questions

• Which of these skills overlap with what we already have for MG dbt work?
• Should we fork and customize, or treat as reference and build our own?
• Is the Agent Skills specification worth adopting as our standard format for portability?