How it works

The CLI is a templated pipeline: your answers become a TemplateContext, Handlebars renders Markdown into .ai/, then IDE adapters copy or merge that into each tool’s native format.

End-to-end pipeline

1. Prompts or preset — Same data shape: framework, ORM, DB, validation, auth, tests, CI, IDEs, skills.
2. buildContext — Derives flags and strings the templates need (hasPrisma, ormServiceName, testCommand, …).
3. Handlebars — Rules, skills, AGENT.md, context, and tracking Markdown are rendered from .hbs files.
4. .ai/ — The canonical copy; treat it as source of truth when in doubt.
5. Adapters — Transform .ai/ into Cursor .mdc rules, merged CLAUDE.md, Copilot instructions, etc.

Deeper dive (for contributors)

Templates live under templates/ in the repo (agent.hbs, rules/, skills/, partials/). The engine registers helpers (eq, neq, includes, or, and) and partials like migrationCommands and gitRules so ORM- and framework-specific text stays DRY.

One template, many stacks

data-layer.hbs is a good mental model: the same template branches on orm to emit Prisma vs TypeORM vs Drizzle commands and paths.

That is how the whole generator stays maintainable — one rule file per concern, filled with your choices.

---

Next: get the CLI on your machine — Installation.