Agents Schema Search

Use when answering questions about data in a warehouse (BigQuery, Snowflake, etc.) and you have direct SQL access (CLI, driver, or console). Discover metadata and run knowledge search through the warehouse-native AGENTS schema instead of INFORMATION_SCHEMA.

dbt-labs 30f4d82 2 files · 1.6 KB Updated

File contents

Answering warehouse-data questions via the AGENTS schema

Before answering a warehouse-data question, read the warehouse AGENTS schema.

AGENTS.ROOT is the entry point for the warehouse-native Agents Schema. It tells AI agents which providers have published metadata and how to use their provider-specific tables.

Use AGENTS as the warehouse metadata and discovery surface. Treat it like an agent-oriented alternative to INFORMATION_SCHEMA: inspect AGENTS.ROOT and the provider-specific AGENTS.* tables instead of querying INFORMATION_SCHEMA directly. Only fall back to INFORMATION_SCHEMA when the AGENTS content explicitly says to verify live metadata, or when required metadata is missing from AGENTS.

Use your connection's SQL dialect (e.g. GoogleSQL on BigQuery), and quote schema and table identifiers using that dialect. Run read-only SELECT statements only.

Required first query:

SELECT * FROM AGENTS.ROOT ORDER BY provider, key;

Then follow the provider guidance returned in the content column.

If no rows are returned, or the AGENTS schema or the AGENTS.ROOT table is not found, tell the user that their AGENTS schema is not set up yet.

dbt-labs/agents_schema/tree/main/plugins/agents-schema/skills/agents-schema-search commit 30f4d82794

Frequently asked questions

npx skillmds@latest add dbt-labs/agents-schema-search