Delay side effects in a project generator

Repeated Next.js setup created the same manual choices around language, styling, API, state, linting, and runtime. A fixed starter was easy to copy, but optional choices would either leave stale files or multiply full-template variants.

A generator writes into a developer’s filesystem. It must not guess how to merge a non-empty destination, and cancellation should not leave a half-configured project. Required choices therefore need to be known before template mutation begins.

• Write files after each prompt answer and clean up if a later prompt is cancelled.
• Overwrite or automatically merge a generated starter into a non-empty folder.
• Maintain one complete starter for every possible option combination.

Resolve the selection set first, stop when the target is unsafe, copy one base template plus only the selected additions, and make optional dependency installation the final operation. This favors predictable failure and centralized shared files over convenience in existing folders.

The result is a published TypeScript CLI, a public repository, and an inspectable npm package. The evidence supports delivery and the decision sequence; it does not establish download adoption or time saved. The learning is that cancellation and destination safety are product behavior, not implementation footnotes.