This post will go through the setup of a basic Jekyll website, with configuration and adapters for Yeoman / Grunt combo. We’ll enable and bridge the LiveReload server of Yeoman with the auto-regeneration feature of Jekyll. We’ll also setup the bits of logics in both our project configuration and custom tasks to adjust the build script to run on a Jekyll directory structure.

Today, I’ve decided to play with Jekyll a bit. This was once I stumbled upon the new Github Page generation thing, with all these new shiny themes. I thought “Would be cool to set this up as yeoman generators, probably as remote templates?”

I also wanted the generator to generate from a minimalist and efficient starting point for Jekyll. So, I started building the thing on @necolasjekyll-boilerplate project.

Here is what the result look like:

  yeoman init jekyll LOCATION [options]

  -h, --help       # Print generator's options and usage
                   # Default: false
      --theme      # Choose a theme from the following GitHub page templates:
                     - orderedlist/modernist
                     - orderedlist/minimal
                     - mattgraham/Midnight
                     - jsncostello/slate
                     - mattgraham/Leap-Day
                     - jonrohan/time-machine-theme
                     - cameronmcefee/merlot
                     - broccolini/dinky
                     - tactile
                   # Default: tactile
      --gruntfile  # Gruntfile to be invoked
                   # Default:

    Yeoman generator based on @necolas' Jekyll Boilerplate.

    This generator will simply fetch the latest state of jekyll-boilerplate's
    master branch (unless the content is already cached by yeoman at
    `~/.yeoman/cache/necolas/jekyll-boilerplate)` in the `location` argument


    - location: Defaults to current working directory. Base directory to copy
                the content of the remote template.


    This generator will also generate:

      - a basic Gruntfile if none were found
      - a tasks/ directory with:
        - server: Override the built-in server command to run Jekyll instead.


    yeoman init jekyll

    This will create:
        |-- .gitignore
        |-- README
        |-- _config.yml
        |-- _layouts
        |   |-- default.html
        |   |-- post.html
        |-- _includes
        |-- _posts
        |   |--
        |-- index.html

Pretty sweet right?

The actual console output:

   invoke  jekyll
   create    .gitignore
   create    _config.yml
   create    _includes/.gitignore
   create    _layouts/default.html
   create    _layouts/post.html
   create    _posts/
   create    index.html
    force    _layouts/default.html
   create    stylesheets/print.css
   create    stylesheets/pygment_trac.css
   create    stylesheets/stylesheet.css
   create    tactile.html
   invoke      gruntfile:jekyll

Scanning the directory structure...OK

Please answer the following:
 Directory - app (app)
 Directory - temp (temp)
 Directory - dist (dist)
 Directory - scripts (scripts)
 Directory - styles (stylesheets)
 Directory - vendor (scripts/vendor)
 Directory - images (images)
 Do you need to make any changes to the above before continuing? (y/N)

Writing local .yeomanrc file with following values: {
  "app": "app",
  "temp": "temp",
  "dist": "dist",
  "scripts": "scripts",
  "styles": "stylesheets",
  "vendor": "scripts/vendor",
  "images": "images"

   create        .yeomanrc
   create        Gruntfile.js


One of the most requested feature request in Yeoman is the ability to define alternate directory structure. The team made some decisions you might disagree on. Some like it called js where others like it as scripts. Others will want it to be assets/javascript.

We know we have to be more flexible

Anyway, we weren’t not really that far from being able to support this feature, the build script and the tasks involved all work on configuration values read from the application Gruntfile.

It’s just that it wasn’t easy at all to decide to change one of these key-directories, since they were used throughout the entire Gruntfile, and sometimes even hard-coded in some of our tasks.

I’m glad to be able to say that better configuration and support for build / server and generators are on the way, and so far, it has proven very efficient. I was able to setup some very fancy directory structure.

The key paths identified so far:

  • app: path to the app directory, defaults to app/.
  • temp: path to the temporary directory, defaults to temp/.
  • dist: path to the final output of the build command, defaults to dist/.
  • test: path to the tests directory, where the unit tests and the test framework runner can be found.
  • components: path to the components directory, where stuff installed via package manager should be.

Relative to app

  • scripts: path to the scripts directory, defaults to scripts/
  • styles: path to the scripts directory, defaults to styles/
  • vendor: path to the vendor directory, defaults to sciprts/vendor

Back to the generator

Scanning the directory structure...OK

Maybe you noticed the message in the previous console outputs. The gruntfile:jekyll generator is an hook of jekyll. Its role is to setup the necessary Grunt configuration, and do a best-guess on the key paths that Yeoman expects.

The idea of scanning the directory structure consist in trying to locate filename patterns we expect from a certain kind of file (like .js or .coffee files for JavaScript sources). This is done thanks to the globbing ability of grunt.file API, we then reduce the results into a list of unique directory names. If the set of detected directories is composed of just a single entry, then we assume it should be a good candidate for that directory.

This goes along the prompt of each directory value. Typing ? on any questions asked will output more informations about this specific directory. Ex:

 Directory - app (app) ?
error:  Invalid input for Directory - app
error:  The "app" directory is the most important directory, and is where your
source files are located.

`scripts`, `styles` and `vendor` are all resolved relative to this directory.

A good place for that is simply `app/` (default value), but if you want the
root of your repository to be the root of your application, you can choose to
set this to `./`

Defaults to `app/`.

 Directory - temp (temp) ?
error:  Invalid input for Directory - temp
error:  The "temp" directory is used mostly with the build and server commands.

It acts as a mirror of your "app" directory, compile-ish steps like coffee,
compass or even rjs should be configured to output in this directory.

The server task when creating the HTTP server will use both directories to
serve static files, from both locations. Order matters, and the "app" directory
is usually configured prior to the "temp" one, meaning that the server will
respond with the file in "app" when the same file appears in both direcory.

Additionally, this directory is used by the build task, to copy the files in
the "app" dir and optimize files in that temporary directory.

Defaults to `temp/`.

 Directory - dist (dist) ?
error:  Invalid input for Directory - dist
error:  The "dist" directory is only used at the end of the build task, simply copying
the "temp" directory to the "dist" one.

Defaults to `dist/`.

 Directory - scripts (scripts) ?
error:  Invalid input for Directory - scripts
error:  The "scripts" directory is resolved relative to the "app" one.

This is where the JavaScript files of your application are located.

This path is used namely with the lint, coffee, and rjs task (every task that
is looking up for JS sources)

Defaults to `scripts/`.

 Directory - styles (stylesheets) ?
error:  Invalid input for Directory - styles
error:  The "styles" directory is resolved relative to the "app" one.

This is where the raw CSS of pre-processor files of your application are

Namely used with the compass.

Defaults to `styles/`.

We have detected the following directory as a good candidate: stylesheets
 - stylesheets/print.css
 - stylesheets/pygment_trac.css
 - stylesheets/stylesheet.css

 Directory - vendor (scripts/vendor) ?
error:  Invalid input for Directory - vendor
error:  The "vendor" directory is resolved relative to the "app" one.

This is where the third-party JavaScript files of your application are located.

Usually called scripts/vendor/ (defaults) or js/libs

You should map the value of "scripts" directory with the value of "vendor", ex.
scripts/ and scripts/vendor or js/ and js/libs.

Defaults to `scripts/vendor`.


To enable below commands, the project Gruntfile should load the plugin tasks, with the standard procedure of calling grunt.loadNpmTask()

I’ve bundled in this package tasks that aims at integrating Jekyll with Yeoman’s workflow. Or Yeoman with Jekyll’s workflow, this works just as well in this direction.


Run the build script

yeoman build

This will compile the Jekyll website in the intermediate directory _temp/, go through yeoman build script, and finally copy the result over _site/

Preview server

yeoman server

This will run Jekyll with the default configuration setup in _config.yml. Both auto and server options should be turned on. Jekyll will watch for modifications and rebuilt automatically in _site.

Yeoman / Grunt are configured to watch any files with _site directory, and will start a LiveReload compatible server on port 35729.

Install the LiveReload extensions, visit http://localhost:4000 (default hostname / port, you change it in the _config.yml), click the LR icon, connection should establish and you can then enjoy the live reloading of your Jekyll website.

How these commands are implemented

I feel like it would be handy, or at least appropriate, to try to explain how and where you can extend, and more specifically override part of Yeoman’s feature.

In our Jekyll use case, the first thing that comes in mind is enabling yeoman server to run a local Jekyll instance as if we were running jekyll --server directly.

Then, we’ll want to enable the watch system, making Jekyll regenerate when necessary, and finally it would be shiny if we can support and bridge into the LiveReload feature of Yeoman.

Turns out that it wasn’t not that difficult.

Server - Spawn, oh sweet Spawn

The goal of this task is to override yeoman’s server task to run jekyll instead.

With the default config in _config.yml, both server & auto should be turned on. Jekyll will generate the website & start up a local server on port 4000 (also configurable in _config.yml) auto is here to watch for changes and regenerate if necessary.

Here we alias the server task to run our own set of tasks.

grunt.registerTask('server', 'jekyll:config jekyll:server watch:jekyll');

The actual server task.

grunt.registerTask('jekyll:server', 'Jekyll server task', function() {
  var jekyll = grunt.util.spawn({
    cmd: 'jekyll'
  }, function(err, out, code) {
    if(err && code !== 0) {



As you can see, it’s as simpler as it can get. The configuration is managed by _config.yml, the task simply delegate the server implementation over Jekyll.

We also ensure (as the first task invoked in jekyll:config) proper config in Gruntfile. In case there are no config for jekyll, this sets up the default watch configuration. We watch for any changes within _site, and notify connected clients via the reload task and LiveReload mechanism.

var watch = grunt.config('watch.jekyll');
// already configured, noop.
if(watch) return;

grunt.config('watch.jekyll', {
  files: ['_site/**/*'],
  tasks: 'reload'

For LiveReload support, we also configure and initialize the “reactor”, with a very basic HTTP server. Its main role is to enable the connection via LiveReload browser extensions, we don’t want to use that server for browsing only to allow the extensions to establish connection.

var livereload = '';
var server = http.createServer(function(req, res) {
  var url = parse(req.url);
  if(url.pathname === '/livereload.js') {
    return request(livereload).pipe(res);

  res.writeHead(200, {'Content-Type': 'text/plain'});

var port = grunt.config('server.port') || 35729;
// ensure proper instantiation of the reactor object
var reactor = grunt.helper('reload:reactor', {
  server: server,
  apiVersion: '1.7',
  host: grunt.config('server.hostname') || grunt.config('') || 'localhost',
  port: grunt.config('server.port') || 35729

server.listen(port, function() {
  grunt.log.ok('LiveReload server listening on ' + port);

Here is for the server part. jekyll:config is responsible of setting up the necessary config, and proper initialization of the LiveReload mechanism. jekyll:server replaces the usual server task by delegating the HTTP server part completely to Jekyll.

Build - Matter of configuration

Since Jekyll is all about generating static files, and that Yeoman is all about working on static websites (or application, read pure frontend app / no dynamic server-side thingy), it wasn’t that difficult either.

Well, maybe a little, we can certainly improve a lof of stuff in this area.

The idea of yeoman build in the typical Jekyll setup is:

  1. Build & compile the Jekyll website as usual: jekyll --no-auto --no-server, with output destination -> temp.
  2. Change the working directory to temporary directory (where the Jekyll website has been compiled)
  4. Cd back to original directory
  5. Do the actual copy from “temp” to “dist”

This was done by overwriting the mkdirs task to run, once again, Jekyll instead, and use it as the cp helper of choice (with the jekyll <path to source> <path to write generated site> pattern)

The final result in _site/ should be the Jekyll result of ., with the full set of optimizations delivered by Yeoman (css/js concat & minified, filenames revved, usemin blocks & references updates, if images, then they are processed through optipng etc.)

You then need to take of the very last step: deployment.


The exercise is interesting because it demonstrates how one can completely rewrite and reimplement some of the built-in tasks involved in Yeoman’s commands, such as build and server. Most of this flexibility is not coming from Yeoman, but from Grunt which is really the foundation of Yeoman.

There is a lot of development environment, a lot of different setup, a lot of different HTTP server or backend framework. Yeoman won’t be able to support them all, but he might try…

Mickael Daniel

mklabs mklabs