redshift-explore

Redshift Guided Explore

Three-step wizard from zero context to a concrete column. Each step lists candidates comment-first so users read and pick.

When to use / NOT

• Use when user is new to a cluster / schema and needs orientation.
• NOT when the answer is already known (jump direct); NOT non-interactive.

Inputs

Form	Behavior
(none)	start at Step 1
<schema>	skip to Step 2
<schema>.<table>	skip to Step 3
--keyword <kw>	pre-filter Step 1 via search_schemas
--max-list N	candidates per page (default 10)

Flow

Step 1 — pick a schema

list_schemas(include_comments=true) → returns {schemas: [{name, comment}]}. Render numbered list, comment-first:

Pick a schema:
  1. dbt_marts     — Final marts layer
  2. dbt_staging   — Staging models
  3. raw_orders    — Raw event stream
  ...
Reply: number / name / keyword.

Ranking: pin schemas with non-empty comments to top, alphabetical; empty-comment schemas labeled (no comment) below. If --keyword passed, prefer search_schemas(keywords). If user replies a keyword, re-render via search_schemas. Auto-pick if cluster has only one schema.

Step 2 — pick a table

list_tables(schema_name, include_comments=true) → {tables: [{name, type, comment}]}. Same render shape. Keyword fallback uses search_tables(keywords, schema_name). Page through --max-list at a time if > 50 tables; show "1-10 of 47, reply 'more'".

Step 3 — pick a column

list_columns(schema_name, table_name, include_comments=true) → {columns: [{name, type, nullable, comment}]}. Render with PK-shaped columns first (heuristic: name = id or <table>_id), then commented columns, then rest:

You picked dbt_marts.fct_orders. Now pick a column:
  1. order_id      bigint       — Unique order identifier (PK?)
  2. status        varchar(32)  — Order lifecycle state
  ...

Recognized replies: number/name → Step 4; keyword → search_columns(keywords, schema_name, table_name) (Step 3 always passes the picked table_name; /redshift-grep-columns is the schema- wide variant); back → Step 2.