Skip to content
Home CLI Core workflow Metadata

Metadata

The /metadata namespace is the universal editing surface for descriptions that already exist (or that you want to author by hand). It is engine-agnostic — the same commands work whether the underlying backend is PostgreSQL, BigQuery, Databricks, Snowflake, or any other supported adapter.

Use this surface when you want to:

  • inspect what AMX currently knows about a schema, table, or column,
  • check coverage across a schema before running /analyze, or
  • hand-write a description without going through the full agent pipeline.

/manual is an alias for /metadata.

Prerequisites

  • An active DB profile (/use-db <name> or scope via --db-profile).
  • For /edit writes to land on the database, the connection user needs comment privileges on the target object.

Commands

/inspect [schema] [table]

Inspect the current metadata state for an asset.

> /inspect sales orders
Schema: sales
Table:  orders                       (12 columns, table-level comment present)
Columns:
  order_id        bigint    PK       "Surrogate key for the order line"
  customer_id     bigint             "FK → customers.customer_id"
  placed_at       timestamp          (no comment)
  ...

If schema and table are omitted, AMX inspects whatever is pinned by /use-db + current_schema + current_table. With only schema, every table in that schema is listed with a coverage summary.

/monitor [schema]

Show metadata coverage across the active scope.

> /monitor sales
Schema: sales
  orders         12/12 columns commented  (100 %)
  order_items     8/14 columns commented  ( 57 %)
  customers       9/9  columns commented  (100 %)
  ...
Total: 78 / 102 commented (76 %)

Useful as a "where do we stand?" check before scheduling a top-up run or deciding whether the documentation effort is mature enough to apply.

/edit [db][.schema[.table[.column]]]

Two ways to invoke:

Path form. The fully-qualified asset address goes straight to the editor:

> /edit warehouse.sales.orders.order_id
Editing column comment: warehouse.sales.orders.order_id
Current: "Surrogate key for the order line"
New:     ▮

The path can be partial — /edit warehouse.sales.orders opens the table-level editor, /edit warehouse.sales opens the schema-level editor.

Wizard form. Bare /edit walks you through a picker:

> /edit
[1/4] Pick a database / catalog: ...
[2/4] Pick a schema:              ...
[3/4] Pick a table:               ...
[4/4] Pick a column (or skip to edit the table comment): ...

Writes are staged into the same pending queue /run populates, so a hand-edit goes through the same human-in-the-loop apply path. Confirm with /apply (or the Studio Pending page).

Verify

> /inspect sales orders
> /monitor sales
> /edit sales.orders                # opens the table-level editor

After saving an edit, the next /inspect shows the new text immediately (it lives in the pending queue) and the underlying database is updated once you run /apply.