Suppose that you begin with the following skeleton application:

(index.html - uncompiled)

<div ng-app="myApp">
  <div ng-controller="MainCtrl">
    <my-directive>
      <p>HTML template</p>
      <p>Scope from {{origin}}</p>
      <p>Overwritten? {{overwrite}}</p>
    </my-directive>
  </div>
</div>

(app.js)

angular.module('myApp', [])
.controller('MainCtrl', function ($scope) {
  $scope.overwrite = false;
  $scope.origin = 'parent controller';
});

The most basic setup is to have the directive scope inherit from the parent scope that will be used by the directive within the link function. This allows the directive to manipulate the parent scope. This can be done as follows:

(app.js)

.directive('myDirective', function () {
  return {
    restrict: 'E',
    link: function (scope) {
      scope.overwrite = !!scope.origin;
      scope.origin = 'link function';
    }
  };
});

This will compile into the following:

(index.html – compiled)

<my-directive>
  <p>HTML template</p>
  <p>Scope from link function</p>
  <p>Overwritten? true</p>
</my-directive>

There's nothing tricky going on here. The directive has no template, and the HTML inside it is subject to the modifications that the link function makes to the scope. As this does not use isolate scope and there is no transclusion, the parent scope is provided as the scope parameter, and the link function writes to the parent scope's models. The HTML output tells us that the template was rendered from our index.html markup, the link function was the last to modify the scope, and the link function overwrote the original values set up in the parent controller.