Query and analyze Tower-managed Iceberg lakehouse data using read-only SQL through DuckDB.
npx skills add https://github.com/tower/agent-skills --skill tower-data-tower
Please help me install this Agent Skill. Skill: tower-data Source: https://github.com/tower/agent-skills/tree/main/skills/tower-data Command: npx skills add https://github.com/tower/agent-skills --skill tower-data-tower
tower catalogs query vends short-lived read-only credentials and runs SQL through a sandboxed DuckDB in a single command, with JSON output for programmatic use. - Schema discovery without credentials: tower catalogs show --full returns every namespace, table, and column type in one call, so the agent never guesses table names. - Persistent data knowledge: tower catalogs knowledge records team definitions (metrics, grain, join keys, column meanings) on the catalog itself, so semantics carry across sessions and agents. - Use Case: A user asks "what was our weekly active user count last month?" The agent reads saved knowledge for the team's definition of "active", discovers the gold-layer tables, runs a single aggregation query with proper timestamp handling, and answers in plain language with auditable numbers. ## Quick Start Ask the agent to answer a business question from your Tower lakehouse, such as "how many orders did we have per day last week?", and it will discover the schema, run the SQL, and report the results.High-intent search queries and answers about installing and using this skill.
Run `tower catalogs query <catalog> -j --sql 'SELECT ...'` to vend short-lived read-only credentials and execute SQL through DuckDB in one command. Reference tables as `"catalog".namespace.table` and quote the catalog name since `default` is a SQL keyword.
Run `tower catalogs show <catalog> --full -j` to get every namespace, table, and column with types in a single call, without vending credentials. This replaces per-table DESCRIBE round trips during exploration.
The query sandbox disables the local filesystem, so DuckDB cannot autoload the icu extension that timezone-aware date functions need. Cast Iceberg TIMESTAMP WITH TIME ZONE columns to plain TIMESTAMP first, then apply date_trunc, year, or CAST to DATE.
The `-j` JSON output silently truncates results at 1000 rows with no warning signal. Aggregate in SQL instead of client-side, or pass `--max-rows 0` for no limit or an explicit `--max-rows N` when you need raw rows.
No. The command accepts exactly one statement per invocation and rejects DDL/DML unless you explicitly pass `--write`. Use CTEs (`WITH ... SELECT`) for multi-step logic, and use the tower-integration skill for ingestion pipelines instead of ad-hoc writes.
It stores team-specific data definitions on the catalog itself, such as metric SQL, grain, join keys, and column meanings, shared across people and agents. Read it before inventing filters, and write back confirmed definitions so future sessions skip re-deriving them.
mathisdrnLoad data into MotherDuck from files, object storage, dataframes, and external databases.
EquiStampConvert validated EquiStamp evaluations into automated production monitoring systems.
exasol-labsDerives, authors, validates, and publishes Exasol Semantic Views models from physical schemas.
leary-pokenConfigure Databricks Iceberg tables with UniForm and Iceberg REST Catalog access.