SQLite foreign-key enforcement
KeyTillage is a Windows AI coding client for inspecting SQLite foreign-key setup. Ask a compatible model to review connection state and orphan reporting in your project, while retaining control of database changes.
import sqlite3
def enable_foreign_keys(db):
if not isinstance(db, sqlite3.Connection) or db.in_transaction:
raise ValueError("foreign_keys.connection")
db.execute("PRAGMA foreign_keys = ON")
if db.execute("PRAGMA foreign_keys").fetchone() != (1,):
raise RuntimeError("foreign_keys.unavailable")
return db.execute("PRAGMA foreign_key_check").fetchall()Connection
Base URL, credentials and a reachable server.
Project
Check a rejected invalid insertion and an old orphan, returning audit rows without repairing data.
Review before applying
Compare actual behavior with the expected result. If a check fails, investigate it before applying more changes.
import sqlite3
def enable_foreign_keys(db):
if not isinstance(db, sqlite3.Connection) or db.in_transaction:
raise ValueError("foreign_keys.connection")
db.execute("PRAGMA foreign_keys = ON")
if db.execute("PRAGMA foreign_keys").fetchone() != (1,):
raise RuntimeError("foreign_keys.unavailable")
return db.execute("PRAGMA foreign_key_check").fetchall()Review database setup before changing stored data
KeyTillage brings database setup code into a Windows AI coding conversation grounded in your project files. Start by identifying where connections are created and what the application assumes about them. Ask the connected assistant to inspect that lifecycle before proposing a setting change, so the review accounts for when the code runs as well as what it says.
Enabling SQLite foreign-key enforcement is one useful project task. Discuss the connection's transaction state and how the code verifies the setting, then distinguish that setup from reporting violations already present in the database. A proposal should explain both behaviors without assuming that enabling a rule also repairs historical records. Keep any decision about data correction separate from the configuration edit.
Review the patch and test it against a disposable fixture containing an existing orphan and an attempted invalid insertion. Check the reported audit rows and rejection behavior using your own database environment. KeyTillage supports OpenAI-compatible or Anthropic endpoints, with credentials when required. You can choose an unrestricted model where available, subject to the provider's rules and the server's tools and context limits. The result to aim for is understandable connection code whose behavior you have observed before applying it to a real application.
Questions before accepting a connection patch
Does enabling enforcement repair existing orphan rows?
Do not assume that it does. The illustrated task enables a connection setting and reports existing violations. Review any proposed data repair as a distinct operation with explicit expected changes, rather than letting it slip into a setup patch.
Why does transaction state matter to this task?
The existing contract requires enabling the setting outside an active transaction. Ask the assistant to inspect when the connection enters that state and verify the resulting setting instead of assuming that issuing the statement establishes the intended behavior.
What should the test database contain?
Use a small disposable database with valid relationships, a known orphan and a write that should be rejected under enforcement. Check both audit output and insertion behavior, keeping the fixture independent of production data.
Can I send a production database to the model?
Prefer schema and sanitized fixtures containing only what the task needs. Review the configured endpoint and tool destinations before sharing material. A local model requires its own running server and does not make every connected tool local.
KeyTillage · Updated 2026-10-25