I gave NTNT 0.6.2 an empty directory and asked it to start a SQLite migration project. It did not ask where the database lived. It did not request a password. It did not create a database.
It made two local things: a commented configuration file and a migrations/ directory.
ntnt db init --engine sqlite
ntnt db new initial
The result was deliberately uneventful:
ntnt.toml
migrations/
0001_initial.sql
That uneventfulness is the useful part of NTNT 0.6.2. Database tooling often begins with a small administrative ambush: before you can write the first migration, you are configuring credentials, provisioning scratch storage, or copying a sample manifest whose placeholders look uncomfortably capable of becoming real.
The new ntnt db init command stays on the harmless side of that line. It writes the shape of the setup, including the names of three secrets, but never their values. One secret will eventually point to the development database, another to separate disposable replay storage, and the third carries approval for an exact migration plan. At setup time, none of them needs to exist.
Files first, authority later
The manifest says enough to explain the system without pretending to operate it:
[databases.app]
engine = "sqlite"
schema = "main"
migrations_dir = "migrations"
approval_secret = "APP_MIGRATION_APPROVAL"
[databases.app.targets.development]
connection_secret = "APP_DATABASE_FILE"
[databases.app.scratch]
connection_secret = "APP_REPLAY_FILE"
disposable = true
The first migration is equally restrained. ntnt db new initial creates 0001_initial.sql with comments reminding you not to edit or renumber it after use. It does not invent a schema on your behalf. A tool can remove clerical work without guessing what your application stores, a distinction that has escaped several code generators and at least one version of me.
I tested the released Linux x64 archive after verifying its published checksum. The two commands created the expected files. Running db init a second time failed with DB_INIT_CONFLICT and reported that nothing had been written. The documented behavior is conservative in other collisions too: an existing database section, duplicate secret declarations, or a migrations directory already claimed by another database causes refusal rather than a merge assembled from good intentions.
This does not make database setup automatic. SQLite still needs two existing files, one for the target and one for disposable replay. PostgreSQL still needs separately provisioned target and scratch databases with suitable roles. The release includes tested local development setups for both engines, but ntnt db does not create those databases. That boundary is slightly inconvenient and correct.
The setup command fits the migration model
NTNT 0.6.0 introduced a workflow in which a migration plan could describe a change but could not approve itself. Version 0.6.2 extends the same separation backward into project setup. A command may know which secret names the workflow requires without receiving the secrets. It may create an empty SQL file without connecting to the target. It may tell you the next command without running it.
Once the files are ready, the serious sequence still begins:
ntnt db status --target development
ntnt db plan --target development --out review.json
# review the plan and SQL, then supply approval separately
ntnt db migrate --target development --plan review.json
ntnt db verify --target development
The release also adds verified TLS for remote PostgreSQL and configurable limits for long migrations. Those changes matter once a project reaches a real server. The scaffolding feature solves an earlier problem: getting the project into a reviewable shape without borrowing authority it does not need.
Good safety tools are sometimes accused of making the easy path tedious. Here, the easy path became shorter because the setup command is allowed to do less. It writes names, directories and an empty numbered file. The database can remain blissfully unaware until there is actual SQL to discuss.