Cursor Cloud agent environment

If you point a Cursor Cloud agent at your SyntaxKit repo, it now lands on a fully working app instead of a bare VM. The kit ships a .cursor/ environment that builds the machine, boots the database, and gets the product running before the agent starts working.

What the agent gets

The environment definition builds an Ubuntu image with Node 22, pnpm, and PostgreSQL, with dedicated development and test databases. The install script starts Postgres, writes machine-local env files, installs dependencies, and runs migrations plus the demo and test seeds. The start script re-bootstraps on every run, so even a just-in-time boot without a build snapshot comes up seeded, with the dev server on port 3000 and a working demo login (admin@demo.syntaxkit.com / password123).

Env files are generated only if they are missing and are never overwritten; they contain nothing beyond machine-local defaults, like the local Postgres URL and a random auth secret. Turborepo passes dashboard-supplied secrets through to next dev, so anything you configure in Cursor Cloud reaches the app without ending up in a file.

Safe by construction

Automatic migrate-and-seed is destructive by nature, so it is gated hard. The safety check resolves the environment exactly the way Prisma does, and it exits successfully only if every write target is a loopback address (localhost, 127.0.0.1, or ::1). It refuses outright if vault mode is detected, covers the test database path too, and understands env overload precedence. The env writer additionally refuses to run outside Cursor Cloud at all. An agent can break its own sandbox database; it cannot touch yours.

Small quality-of-life fixes along the way

Making the environment self-bootstrapping surfaced a few rough edges that are now fixed for everyone: git hooks install cleanly even when a hooks path is managed, the root test scripts load TypeScript package exports correctly, and pnpm setup:doctor resolves workspace packages under pnpm's isolated linker.