Monorepo workspace tooling (turbo / nx / lage / wireit)

When to use this: the repo uses pnpm/yarn workspaces with an internal package graph — apps/server depends on packages/db and packages/shared, all built from source. The naive pnpm run dev inside apps/server/ runs only that package's dev script and skips the dep graph; you'll get stale builds or missing-export errors at runtime.

The fix is to run the workspace task runner (turbo / nx / lage / wireit) from the repo root with a filter, so it knows to build dependencies first.

owned:
  server:
    # WRONG: builds only apps/server's TS, not the @packages/* it depends on.
    # cmd: pnpm run dev
    # cwd: apps/server

    # RIGHT: turbo orchestrates the build graph across workspace deps,
    # then runs the `dev` script in apps/server with everything pre-built.
    cmd: pnpm exec turbo run dev --filter=server --env-mode=loose
    cwd: .
    port: { published_env: PORT }
    ready_when:
      http_get: /health

The pnpm exec prefix is also load-bearing: cmd: invocations don't necessarily have node_modules/.bin on PATH, so a bare turbo run dev may not resolve to the workspace-local turbo binary. Use pnpm exec (or yarn run / npm exec) to route through the package manager, which sets up PATH correctly. (LEV-498 is in flight to auto-prepend node_modules/.bin for cmd: invocations; until that ships, the pnpm exec workaround is the recipe.)

Equivalent shapes for the other workspace runners:

• nx: cmd: pnpm exec nx run server:dev
• lage: cmd: pnpm exec lage dev --scope server
• wireit: cmd: pnpm exec wireit (in the app's cwd:, with wireit deps wired in package.json)

Common mistake: setting cwd: apps/server + cmd: pnpm run dev because that's what the README says — locally a developer runs it from the app dir after a manual pnpm -w build to seed dist/. Lich doesn't know about that prior build; cold starts hit "Cannot find module '@packages/db'" because nothing built it.