Declarative Schema Renames

Atlas now detects renames declaratively. Add renamed_from to a table, column, or index in HCL, or an -- atlas:renamed_from directive in SQL, and Atlas plans ALTER ... RENAME statements instead of DROP + CREATE. The interactive rename prompt is a human-only contract: agents cannot answer it. Making the intent part of the schema removes that blocker.

Atlas has historically detected renames at plan time by prompting interactively to confirm whether a removed object matched a newly added one. That contract is human-only: an autonomous agent running plan/apply non-interactively has no way to answer the prompt, so a renamed table or column falls back to DROP + CREATE and the data is lost.

The new renamed_from attribute makes the intent declarative. It lives next to the renamed resource, is reviewable in pull requests, and is resolved without prompts. The planner emits driver-native ALTER ... RENAME statements that preserve data.

HCL: renamed_from on Tables, Columns, and Indexes

Add renamed_from to the new resource and set it to the previous name. The same attribute works on tables, columns, and indexes:

table "accounts" {
  renamed_from = "users"
  column "email_address" {
    type         = varchar(200)
    renamed_from = "email"
  }
  index "idx_email_new" {
    columns      = [column.email_address]
    renamed_from = "idx_email"
  }
}

SQL: the -- atlas:renamed_from Directive

In SQL schemas, add an -- atlas:renamed_from <prev> comment above the object. The directive can be statement-level (above a CREATE TABLE or CREATE INDEX) or inline (above a column or named-index line inside a CREATE TABLE body):

-- atlas:renamed_from users
CREATE TABLE accounts (
  id bigint NOT NULL,
  -- atlas:renamed_from email
  email_address varchar(200) NOT NULL,
  PRIMARY KEY (id)
);

-- atlas:renamed_from idx_email
CREATE INDEX idx_email_new ON accounts (email_address);

What Atlas Emits

The planner produces driver-native rename statements rather than DROP + CREATE. Below are examples for the same set of renames (rename a column with a type change, rename a table, rename a column, rename an index) on MySQL and PostgreSQL:

-- Modify "posts" table
ALTER TABLE `posts` CHANGE COLUMN `title` `headline` text NOT NULL;
-- Rename a table from "users" to "accounts"
RENAME TABLE `users` TO `accounts`;
-- Modify "accounts" table
ALTER TABLE `accounts` RENAME COLUMN `name` TO `first_name`, CHANGE COLUMN `email` `email_address` varchar(200) NOT NULL, RENAME INDEX `idx_email` TO `idx_email_new`;

On MySQL, when a single column has both a rename and a type change Atlas coalesces them into one CHANGE COLUMN <old> <new> <new-type> clause atomically (see the accounts.email column above). Multiple alterations to the same table are merged into a single ALTER TABLE statement.

-- Rename a column from "title" to "headline"
ALTER TABLE "posts" RENAME COLUMN "title" TO "headline";
-- Modify "posts" table
ALTER TABLE "posts" ALTER COLUMN "headline" TYPE text;
-- Rename a table from "users" to "accounts"
ALTER TABLE "users" RENAME TO "accounts";
-- Rename a column from "email" to "email_address"
ALTER TABLE "accounts" RENAME COLUMN "email" TO "email_address";
-- Rename a constraint from "users_pkey" to "accounts_pkey"
ALTER TABLE "accounts" RENAME CONSTRAINT "users_pkey" TO "accounts_pkey";
-- Rename an index from "idx_email" to "idx_email_new"
ALTER INDEX "idx_email" RENAME TO "idx_email_new";

On PostgreSQL, every rename is its own statement (RENAME COLUMN, RENAME TO, RENAME CONSTRAINT, ALTER INDEX ... RENAME TO) and primary-key constraint renames are issued automatically alongside their table rename (see users_pkey → accounts_pkey above).

Why It Matters for Agents

Schema changes proposed by an agent now flow through the same review path as any other code change: the agent edits the schema (HCL or SQL), commits the diff (including renamed_from), and Atlas plans the migration without asking a human to disambiguate at the terminal. The contract between agent and planner is now entirely textual, which is what every other part of an agent-driven CI pipeline already expects.