Bikeshed Documentation

Living Standard,

This version:
Issue Tracking:
Inline In Spec
Tab Atkins-Bittner


Bikeshed is a spec-generating tool that takes in lightly-decorated Markdown and spits out a full spec, with cross-spec autolinking, automatic generation of indexes/ToC/etc, and many other features.

1. Installing

1.1. Using Bikeshed Without A Local Install

1.1.1. Via the Web

If you use Bikeshed infrequently, and are okay with requiring a network roundtrip every time you invoke Bikeshed, you probably want to use the Bikeshed API instead. In return, the API version is always up-to-date, so you don’t have to remember to update things yourself. See § 2 Invoking Bikeshed Without Installing for options.

Note: The remote API has several limitations; most notably, it can’t take multiple files. If you are using custom boilerplates, or include files, you’ll have to run Bikeshed yourself.

1.1.2. Via pipx

If you use Bikeshed infrequently, but have the prereqs for a local install (a recent Python 3, and pipx; see below), you can run an always-up-to-date Bikeshed locally as:

pipx run bikeshed update && pipx run bikeshed

# on subsequent invocations, just call:
pipx run bikeshed

This does not have the limitations of the web version; you can use all the features of Bikeshed, including custom boilerplates and include files, or alternate run modes like bikeshed watch.

Note: This actually creates a temporary local install, which automatically disappears after a few days. It does not add the bikeshed command to your PATH. However, it generally won’t require manual upgrading; so long as it’s been long enough for pipx to delete the cached version, it’ll always grab the newest version of Bikeshed.

1.2. Prereqs

Bikeshed is a Python program, and getting Python working well in modern systems has a few (pretty easy) steps. (These might already all work for you!)

1.2.1. python3

Bikeshed requires Python 3.9 or later. (It’s currently tested up to 3.11.)

To tell what version you have, run:

python3 --version

If it reports 3.9 or later, you’re fine.

If it’s earlier, you’ll need to update your local Python version. There are several ways to do so, but I recommend pyenv as a safe and easy method. gives instructions for installing pyenv on various environments. It also links to the auto-installer, which can make things even easier. Windows users have slightly different instructions, but it links to a windows fork as well.

If you know your way around pyenv already, feel free to do what you’re used to here. Otherwise, it’s usually easiest to set up 3.10 as your "default" Python:

pyenv versions

will list what versions exist, and then

pyenv global 3.10.4

will set 3.10.4 as your "global" version, used by default. (Substitute in whatever the name of the version you saw in versions was.)

1.2.2. pipx

Installing packages globally is usually a bad (and sometimes insecure!) idea. They should instead be isolated into "virtual environments". This used to be a little tricky to set up, but now pipx makes it very simple.

To install pipx, your system software distribution might already have it; for example, distributions using apt can run

sudo apt-get install pipx

If this is not the case for you, see pipx’s install instructions and do whatever it says.

1.3. Installing Bikeshed Itself

1.3.1. Installing For Normal Use

Assuming the prereqs (above) are satisfied, installation is trivial:

pipx install bikeshed

When this is completed, Bikeshed should be installed, and the bikeshed command should work in your shell.

Note: If this is your first time running pipx, you might get a message complaining about your PATH variable, and Bikeshed’s attempt to update its data files will fail. If so, follow pipx’s instructions, then run bikeshed update again.

After this, invoking Bikeshed is just:

cd ~/my-spec-folder

Remember to update Bikeshed regularly by running:

pipx upgrade bikeshed

See § 1.4 Updating Bikeshed for more details.

You can also manually refresh Bikeshed’s datafiles by running:

bikeshed update

But this will run automatically whenever you use Bikeshed if your data files are more than a few days old, so it usually shouldn’t be necessary.

Note: If you’re on a Mac, bikeshed update might fail with some errors about certificates. This is a known issue with the version of Python 3.7 shipped with XCode, which might be the default on your system. To fix this, install Python 3.7 with Brew instead.

1.3.2. Installing Bikeshed for Development

If you’re installing Bikeshed so you can work on it, or want to ensure you have the bleeding-edge tip-of-tree version, the instructions are just a tiny bit more complex.

First, clone the Bikeshed repository:

git clone

Then navigate to that folder (by default, it’ll be a folder called "bikeshed" in the current folder you’re in) and run:

pip3 install -e .

This will spam your console with a bunch of install progress. When it successfully completes, the bikeshed module should be globally available for import in Python, and a bikeshed command should now work in your shell. Installing With Pipenv

You probably don’t want to install all of Bikeshed’s dependencies globally, so I recommend using pipenv to install it into a virtual environment (similar to what pipx does for non-dev installs).

Follow the same instructions as above, but instead of running pip3 install, run:

pipenv install --dev -e .
pipenv run bikeshed update

This will not install a bikeshed command by default; instead, you run Bikeshed with pipenv run bikeshed. If you’d like bikeshed to work by itself, either start a pipenv shell with pipenv shell, or add an alias to your machine like:

bikeshed=pipenv run bikeshed

1.4. Updating Bikeshed

To update bikeshed to its latest version at any time, just run:

pipx upgrade bikeshed && bikeshed update

This’ll pull the latest version of Bikeshed, and ensure that you’re looking at the latest version of the data files, rather than whatever stale version is currently sitting in the repo.

Note: If you did an "editable" install up above, to update it you just run the install command again and invoke bikeshed update afterwards.

1.5. Building a Bikeshed docker image

Note: I don’t actually understand Docker or know how to use it, so this guide is based on one community member’s efforts. If something is wrong or outdated, uh, figure out how to fix it and tell me so I can fix the docs.

1.5.1. Invoking Bikeshed using a docker image

Note: As stated up in § 1.5 Building a Bikeshed docker image, I don’t know how Docker works and this guide is based on a community member’s experience, so use at your own risk.

Typically, this requires login to the docker registry and pulling a Bikeshed image.

For some <organization> and some <date>:

docker login
docker pull <organization>/bikeshed:<date>

Regardless of host environment’s operating system, running Bikeshed from a docker image requires two things:

Example for a Unix host:

docker run --rm -v $(pwd):/data <organization>/bikeshed:<date> bikeshed spec /data/<some *.bs file> [/data/<some output file>]

Example for a Windows host:

docker run --rm --volume C:\mystuff\project1\:/data <organization>/bikeshed:<date> spec /data/ /data/out/Index.html

Note that the § 3.1 Global Options apply to running Bikeshed from a docker image. Since the Bikeshed docker image is read-only, it does not make sense to execute the Bikeshed update command from a docker image.

2. Invoking Bikeshed Without Installing

While a locally-installed Bikeshed does provide additional functionality, if all you need to do is process a spec, there is an API server you can talk to, maintained by Peter Linss.

2.1. Using Curl

These instructions assume use of the curl command, but you can use any equivalent "talk HTTP at a server" command you might have access to.

curl -F -F force=1 > index.html

Simplest possible usage:

Additional possible arguments:

If your source file is online, such as in a git repository, you can pass the url directly instead:
curl -F url= -F force=1 > Overview.html

If you are using additional files beyond your plain source file—​local boilerplate files, sub-document includes, custom definition files, etc—​this is the only way to make those still work via the curl API; Bikeshed will look for the additional files relative to the source document’s url.

(Any additional files that get included implicitly, such as local boilerplates, must be listed in the Local Boilerplate or External Infotrees metadatas to get picked up. Files explicitly listed in the source file, such as in a <pre class=include>, will be picked up automatically.)

The separate bikeshed issues-list command (for generating issues lists for a W3C Disposition of Comments) can also be invoked via curl:
curl -F file=@Issues.txt -F input=issues > Issues.html

2.2. Using the Web Form

Alternately, you can just visit directly, and upload a file, point to a URL, or directly paste in your spec source.

It defaults to outputting both the HTML and the errors, but you can switch it to one or the other only if you want.

2.3. Using CI Tools

2.3.1. Travis CI

To use bikeshed on Travis CI's github integration, you’ll need the following .travis.yml commands:

language: python
  - "3.8"
  - pip install bikeshed
  - bikeshed update
  # Invoke bikeshed here, at your own leisure. E.g.:
  - bikeshed spec

Raymond Toy has written a very thoro guide on the full setup you need to auto-build and -publish the generated files to GitHub via Travis, so you only need to check in the source files themselves.

2.3.2. GitHub Actions

The W3C maintains a spec-prod GitHub Action that allows automatically processing commits or PRs as either Bikeshed or ReSpec.

To use, add a file to .github/workflows/ folder in your GitHub project, containing:

name: CI
  workflow_dispatch: {}
  pull_request: {}
    branches: [main]
    name: Build, Validate and Deploy
    runs-on: ubuntu-20.04
      contents: write
      - uses: actions/checkout@v3
      - uses: w3c/spec-prod@v2
          TOOLCHAIN: bikeshed

          # Modify as appropriate
          GH_PAGES_BRANCH: gh-pages

          # if your doc isn't in the root folder,
          # or Bikeshed otherwise can't find it:
          SOURCE: src/

          # output filename defaults to your input
          # with .html extension instead,
          # but if you want to customize it:
          DESTINATION: src/index.html

The linked document has a number of usage examples to achieve various outcomes—​simple validation, auto-publishing on, building multiple specs from a single repository, and some others. It also explains the entire process from start to finish, if you need more help setting things up.

3. Invoking Bikeshed Locally

Locally-installed Bikeshed is invoked via the command-line. There’s a bunch of options, but you’ll only use a few in day-to-day use.

If you’ve installed Bikeshed via one of the standard local installs above, you can typically just type bikeshed in the folder where your .bs file lives and it’ll do its thing, automatically invoking the spec subcommand.

However, Bikeshed has a number of different subcommands, and both global and subcommand-specific flags that can be invoked.

3.1. Global Options

There are a few options that apply to many different options, so they’re specified before the individual commands:

-h or --help

Shows the help message for the current command. Can be run without any command, like bikeshed -h to show the list of valid commands and the global flags, or after a command, like bikeshed spec -h to show the help text for that command in particular.

--die-on = [ nothing | fatal | link-error | warning | everything ]

Bikeshed categorizes errors into several categories, depending on their severity: clear mistakes are "fatal", errors in resolving autolinks are "link-error", and things that look a little questionable but are totally okay to generate a document with are "warning".

The --die-on flag controls which sorts of errors cause Bikeshed to fail to produce output and exit with a non-0 status (indicating the command failed).

Possible values are "nothing", "fatal", "link-error", "warning", and "everything", with each level including all previous levels. Defaults to "fatal".

It can sometimes be useful to set this to "nothing" if you just need to push thru and generate a spec that happens to throw some unimportant errors you don’t have time to fix. The -f flag is a shorthand for this.

Alternately, it can be useful to set it to a level stricter than "fatal" if you want to ensure that, for example, a CI build fails when link errors creep into your document, or if you just generally want to ensure that your document builds "cleanly", without any warnings at all. "everything" is equivalent to "warning", but is guaranteed to always die on all error messages, even if more levels are added in the future.

Note that this can also be controlled from within your document (or your group’s metadata) via the Die On metadata.

--die-when = [ early | late ]

When Bikeshed encounters a disallowed error (per the --die-on value), it can either die immediately (early) so you can see the error and fix it without the rest of the document potentially being incorrect and throwing nonsensical errors; or it can wait until the end of processing (late) to die, so you can see all the errors without having to continually rerun the command.

Either way, Bikeshed will exit with a non-zero status, and refuse to actually produce an output document.

If unspecified, this defaults to late.

(If using late and processing a document with many output messages, it might be useful to temporarily run it with -q or -qq, so you can focus on the more important messages first.)

Note that this can also be controlled from within your document (or your group’s metadata) via the Die When metadata.

-f or --force

A shorthand for --die-on=nothing.

-q or --quiet

The -q flag suppresses one level of error messages. It can be passed multiple times to suppress additional levels, in the order "warnings", then "link errors", then "fatal errors".

-s or --silent

The -s flag suppresses all console output from Bikeshed, regardless of source. (It’s more powerful than -q, as it’ll also suppresses things like the success/failure message.)

-d or --dry-run

The -d flag prevents Bikeshed from actually writing anything to disk. It does everything else, just skips that final "save" operation.

--print= [ plain | console | markup ]

Specifies how Bikeshed should output its messages. Default is "console", which outputs colored text using console color codes. "plain" outputs the same text as "console", but without any color codes (which look like gibberish if you’re not outputting to a console). "markup" outputs the text in a light markup structure (suitable for parsing as HTML), for easier parsing of the output by tools.


By default, Bikeshed checks how old its data files are every time it runs, and automatically triggers an update (as if you ran bikeshed update) if they’re more than a few days old.

If this is causing problems (for example, if something has gone wrong with the update service, or you are offline and don’t want to wait for the network to timeout), you can prevent it from doing the check with --no-update.


Bikeshed has the ability to include arbitrary files into your spec, and in a few cases the ability to execute shell commands or Python on your behalf (to allow for some extensions).

By default, these are restricted: you can only include files from the spec’s own folder or subfolders, and arbitrary code execution is disallowed.

If you trust the specs you are processing, and need these abilities, however, the restrictions can be relaxed by passing these flags.

3.2. bikeshed spec

The spec command is the most common one you’ll use. It turns a Bikeshed source file into an output HTML file. The rest of this document mostly describes how to format a source file for use with this command.

Most of the time you can just run bikeshed spec and things will "just work" as long as your source file has the .bs extension.

Relevant flags:


If you’re using Inline GitHub Issues heavily, you might run into the GitHub rate limit. You can generate an OAuth token at and then pass it to Bikeshed via this flag to raise your limit considerably.

-l or --line-numbers

If you’re having trouble locating the source of an error in your code, run Bikeshed with this flag and almost all errors will report the source line number they appear at (or somewhere very close - it’s per-element).

This code is unfortunately forced to be fairly hacky, and has the potential to lightly corrupt your file (inserting extra debugging info into plain text that just happens to look like markup), so it automatically triggers a dry-run, as if you’d specified --dry-run. When you’re done debugging, just run again without this flag to actually get some output.

After any flags, you can optionally specify the input file path and output file path. Both of these can usually be omitted; if you omit the output file, it’ll just output to a file with the same name as the input file, but with the extension changed to .html; if you omit the input file, it’ll search thru the current directory and assume you want the first file with a Bikeshed extension (.bs).

If you want to feed Bikeshed from stdin or have it output to stdout, just use - as the appropriate filename.

3.3. bikeshed watch

The watch command is identical to the spec command, except it sets up a watcher and auto-rebuilds the spec every time it changes. With this turned on, you can edit with a simple save->refresh cycle, rather than having to do save->build->refresh.

It accepts all the same arguments as spec. (Tho using stdin/stdout might get a little weird.)

Use Ctrl-C to stop the watcher (or whatever key combo kills the currently running process in your console).

3.4. bikeshed template

The template command outputs a minimal "skeleton" document to stdout. It’s useful to get started with Bikeshed, without having to remember what all the required metadata is.

On a Linux-like command line, it should be used like:

bikeshed template >

3.5. bikeshed echidna

The echidna command hooks into the W3C’s Echidna auto-publishing system, letting you publish W3C specs from the command-line, rather than having to go thru a staff contact.

Note: Currently, only Working Drafts (not including FPWD) and Candidate Recommendation Drafts can be published via Echidna.

It’s invoked similarly to bikeshed spec, with three required flags:


Your W3C username


Your W3C password

--decision DECISION_URL

A link to a publicly-visible message approving the publication of this draft.

There are some optional flags:


Marks the publication as an editorial update.


A comma-separated list of email addresses which the W3C will ping when the publication request is completed.

--additional-directories [DIRECTORY ...]

By default, Bikeshed will look for directories named images, diagrams, and examples, and include them and their contents in the publication. This flag will override that list, letting you specify any number of files and/or folders that should be included with the publication.


Conversely, this flag signals that the document doesn’t use anything external, so it shouldn’t include anything even if one of the default directories does exist.


The Echidna service works by accepting an uncompressed TAR file of all the files that will be published. Passing this flag just constructs that TAR file and prints it to stdout, without actually contacting Echidna.

After the flags, you can optionally specify the input file path, just like for bikeshed spec; if omitted, it selects a file automatically in the same way.

Bikeshed then builds the spec normally, performs a little bit of fixup for common issues that PubRules complains about, and submits the spec to Echidna. It will print out a URL for you to check on the progress of your publication; it should complete in a few seconds.

Most likely you’ll have some PubRules errors when you check your progress; fix those (ask in for help if needed) then just run bikeshed echidna again.

Note: If auto-publishing via Travis CI, this Travis blog post about encrypted environment variables will be helpful for safely using your W3C password.

3.5.1. Testing Publication Locally

Publishing via Echidna automatically does some "fixup" on your document, to fix some common errors that would make the doc fail the W3C’s PubRules check.

These fixes are controlled by the Prepare For TR metadata, so if you want to see precisely what your spec will look like when it’s published, add that to your metadata block manually.

If you want to go further, and see the actual TAR file that will be sent to the Echidna service, you can pass --just-tar to the command (and omit the authentication flags that are normally required). This will cause Bikeshed to create a file named test.tar in your current directory.

3.5.2. Echidna Hooks

Bikeshed’s default publication behavior is usually sufficient, but if you need to customize it in some way, there are a few hooks you can use (assuming you can write Python). In an include file named bs-extensions.include, provide an implementation of the following methods (check the default version of this file for the default implementations, for the methods you don’t want to change):


This method is called after processing is done, when the document is ready to be packed up and sent to Echidna. Here, you can perform whatever final post-processing is required.

For example, the CSSWG EDs link to a stylesheet in the parent directory of the server they’re stored on. In their BSPrepTR(), they search for the link to that stylesheet, and update it to point to the current directory instead.

This method allows you to specify what files should be included in the publishing bundle. It must return an array of additional files/folders; each entry must either be a string, referring to a file/folder in the spec’s directory or subdirectories, or a tuple of 2 strings, the first of which points to a file outside the spec’s directory, and the second of which provides the path the file should have within the spec’s directory.

The files value provides the default values, which you probably want to just extend, rather than fully replace. It defaults to ["images", "diagrams", "examples"], indicating that those folders, if present, will be included.

For example, as stated in the example for BSPrepTR(), the CSSWG needs to include its stylesheet in with its specs. To do so, it just needs to add the entry ["../default.css", "default.css"] to the files array, indicating that Bikeshed should grab the default.css file from the parent directory, and put it in the spec’s directory (in the bundle) with the same name.

3.6. bikeshed update

The update command updates Bikeshed’s datafiles, which it uses for autolinking and similar things. By default it’ll update all of its datafiles, but if you want to update only particular ones, you can pass any or all of the following flags:

By default, Bikeshed’s update system relies on the Bikeshed-Data project, which preprocesses the various data sources directly into Bikeshed’s data formats, and prepares a manifest of changed files, letting Bikeshed quickly download only what it needs. However, this can be up to 10 minutes out of date (or longer, if the update process has fallen over); if you need absolutely up-to-date information right now, pass the --skip-manifest flag to force Bikeshed to do its full manual update process. (You can combine it with any of the flags above to only get the exact files you need.)

3.7. bikeshed refs

The refs command lets you search Bikeshed’s anchor database, letting you see what sort of things it’s looking at when doing autolinking. This can be very useful for debugging, or just for a quick check of where something is defined.

It’s flags are similar to the attributes used on autolinks:

Any or all of these flags can be specified, and Bikeshed will display all the refs it can find matching the criteria.

3.8. bikeshed source

The source command applies various transformations to the source document itself, rather than producing a separate output document. Its options are described in § 14 Source-File Processing: bikeshed source.

3.9. bikeshed issues-list

The issues-list command processes a plain-text document in an "issues-list" format pioneered by the CSSWG into an equivalent HTML version.

Relevant flags:

After the command you can pass the input and output filenames. As usual, one or both can be omitted: if you omit the output, it’ll write to a file with the same name as the input, but a .html extension; it you omit the input, it will look in the current folder for any files starting with "issues" and ending in ".txt", and then extract the digits from those filenames and select the one with the largest number. If you name your file as suggested above, with an ISO date, it’ll correctly always choose the latest issues list.

Define the issues-list format. The -t output is already more than enough to actually work with, but it would still be good to describe it more fully.

4. Metadata

Crucial to the processor’s ability to automatically generate most of a spec’s boilerplate is a small metadata section, typically located at the top of a file.

A metadata block is just a <pre class='metadata'> element, with contents like:

Status: UD
Shortname: css-variables
Level: 1
Editor: Tab Atkins Jr., Google,
Editor: Daniel Glazman, Disruptive Innovations,
Abstract: This module introduces cascading variables as a new primitive value type that is accepted by all CSS properties,
  and custom properties for defining them.

The syntax of a metadata block is very simple - it’s a line-based format, with each line consisting of a key and a value, separated by a colon. Or if you’re adding multiple lines with the same key, you can just start the subsequent lines with whitespace to have it reuse the last-seen key.

Several keys are required information, and will cause the processor to flag an error if you omit them:

There are several additional optional keys:

A boolish value is a string representing a boolean value (true or false). The string "yes", "on", "true", and "y" all represent true values, while the strings "no", "off", "false", and "n" all represent false values.

A soft boolish value is either a boolish value, or one of the strings "maybe", "if possible", or "if needed", indicating an intermediate intention between something being turned on and off.

You can also provide custom keys with whatever values you want, by prefixing the key with !, like !Issue Tracking: in spec. Any custom keys are collected together and formatted as entries in the "spec metadata" boilerplate dl. Specifying a custom key multiple times will put all the values as dds under a single dt for the key. A custom key name equal to one of the auto-generated keys will add your custom value as an additional dd under that auto-generated key.

Some of the metadata keys are deprecated; you shouldn’t use them, but just in case you run into them in the wild, they’re documented here for understanding. Each one recommends what you should be using instead.

4.1. Default Metadata

To specify default metadata for all specs generated for a given group and/or spec status, add an appropriate defaults.include file to the bikeshed/boilerplate/ folder. This file must be a JSON file, with the keys and values all strings matching the above descriptions.

Here’s an example file:

  "Mailing List": "",
  "Mailing List Archives": ""

4.2. Computing Metadata From Other Metadata

If your Group produces several specs, and they all share some formulaic metadata that could be auto-generated with a bit of string-replacement, you can achieve this automatically with a "computed-metadata.include" file in the bikeshed/boilerplate/ folder. It has the exact same syntax as the defaults.include file, except that you can use text macros in the file, based on all the previous metadata (defaults, <pre class=metadata>, and command-line overrides).

Note: Don’t forget about custom text macros with the Text Macro metadata, if none of the pre-defined ones give you the substitution strings you need.

Note: By necessity from the definition, these metadata override all previous metadata. Make sure that everything you set in this file is really something that every single spec in your Group should have.

Note: Bikeshed assumes that any text macros you use will be used inside of JSON strings, and escapes them accordingly. (In other words, you don’t have to worry about JSON-escaping when writing your macros, but you do need to ensure that all macros occur inside of double-quotes in your computed-metadata.include file.)

4.3. Overriding Metadata From The Command Line

If you want to generate multiple versions of a spec from the same source (such as a primary spec, plus some snapshots), you can override the metadata from the command line to generate the alternate forms.

For any metadata key defined above, just pass it as a --md-foo=bar command-line argument. For example, to override the Status metadata, run bikeshed spec --md-status=ED.

(If the metadata name has spaces in it, use dashes to separate the words instead.)

4.3.1. Known Issues

  1. You can’t override the Use <i> Autolinks status, because you can’t input the <> characters. I don’t intend to fix this, as you shouldn’t be specifying this in the first place.

  2. You can’t supply custom metadata keys (ones with a ! prefix). If you want to do this, let me know, and I’ll work on it.

  3. Passing values with spaces in them is tricky. This is an issue with the argparse library. The only way around it is to specify both of the positional arguments (the input and output filenames), then put the offending argument after them.

4.4. Ordering Between Metadata Sources

When you specify a particular metadata key multiple times in your document, the exact behavior depends on which metadata key it is, but all of the behaviors depend on the order things appear in. For example, if you specify Title twice, the second one is used.

This also happens when metadata keys are duplicated between multiple sources, like from your defaults.include file and your <pre class=metadata> block. It’s important, then, to know what order the different sources are resolved in, so you know which key "wins", or in what order a list of things is constructed in.

The ordering is, from first to last:

  1. "baseline" metadata (the default values that Bikeshed uses)

  2. defaults.include metadata

  3. computed-metadata.include metadata

  4. <pre class=metadata> blocks in your document

  5. command-line metadata

In general, this follows the ordering of least-specific to most-specific.

4.5. HTML "Metadata" (lang="", <meta>, etc)

Generally, any HTML metadata (attributes on <html>, <meta> or <link> elements) should be handled in your header.include boilerplate file. However, sometimes you do need to add additional metadata to a specific document.

For attributes that belong on <html>, such as dir="" or lang="", simply add an <html> element with those attributes into your source file. A quirk of the HTML parser is that multiple <html> elements, wherever they appear, coalesce their attributes onto the one "true" <html> element that wraps the entire document, rather than actually showing up in the DOM. Bikeshed’s serializer will then output the one true <html> element with all the coalesced attributes.

Similarly, if you need to add <meta>, <link>, or <style> elements to your document, simply put them in your source. Bikeshed will automatically transfer them to the <head> of the output document. (<script> elements are left where they are, as their location can matter.)

5. Markup Shortcuts

Bikeshed’s source format is roughly HTML, but it allows you to omit or shorten several verbose/annoying parts of the language, reducing the amount of format noise in the spec source, making it easier to read and write.

5.1. Markdown

Bikeshed uses a Markdown variant called Bikeshed-flavored Markdown (BSMD). By default, it recognizes all of the "block-level" Markdown constructs defined by CommonMark, except for indented code blocks:

Because it doesn’t have to handle indented code blocks, BSMD’s HTML handling is generally superior to other Markdown implementations. You can freely switch back and forth between Markdown and HTML as needed, and indent properly. (In fact, you should properly indent everything; if you don’t, you can sometimes get confusing errors!)

If the Markup Shorthands: markdown yes metadata is specified, it also recognizes several of the Markdown "inline elements":

It does not recognize "autolinks" (surrounded by <> characters), images, or "hard line breaks"; use HTML for these. It also does not recognize "link references" (defining the link location elsewhere, and referring to it by reference instead).

In addition to standard (CommonMark) Markdown, Bikeshed also recognizes definition lists, with the following format:

Here's the dl syntax:

: key
:: val
: key 1
: key 2
:: more vals

For all three list formats, on the rare occasions you need to add a class or attribute to the list, you can wrap it in the appropriate list container, like:

<ol class=foo>
  1. first
  2. second

Bikeshed will use the container you provided, rather than generating a fresh one like it does by default.

Bikeshed also supports adding IDs to headings, via the Markdown Extra syntax:

Header 1 {#header1}

### Header 2 ### {#header2}

More of Markdown will be supported in the future, as I get closer to adhering to the CommonMark specification.

5.2. Notes, Issues, Examples, Advisements

The Markdown processor specially recognizes paragraphs starting with "Issue: ", "Advisement: ", "Assertion: ", "Note: ", or "Note, ", and will add a matching class to the paragraph automatically. These classes default to .issue, .advisement, .assertion, and .note, but can be customized with the metadatas Issue Class, Advisement Class, Assertion Class, and Note Class. (However, the default classes receive styling from the default stylesheet, so make sure you provide your own styling if you change them.)

The default styling of these blocks includes a generated-content "header" containing the word "NOTE:", etc. Elements with the .example class will also be given a generated-content "header" in the format EXAMPLE n with an auto-incrementing number, and have self-links created to allow linking directly to the example.

If you’d like to provide your own custom header, write out the container element yourself (rather than using the Markdown shorthand, just create any element with the appropriate class), and add a heading="YOUR TEXT HERE" attribute to the container. It will automatically have "NOTE: "/etc prepended to it. (This is used by § 5.10 Remote Issues to put the GitHub issue title on the issue containers.)

Alternately, if you want full control over the contents of the heading, write out the container element and create a <div class=marker> element containing the text you want as its first child. (This is what the heading="" attribute does; it just automatically prepends "Note:"/etc to the text to keep it visually in line with other notes/etc by default.)

5.3. Typography Fixes

Bikeshed will automatically handle a few typographic niceties for you, ones that it can reliably detect:

There are several autolink shortcuts that make autolinks much shorter to write, such as 'foo' to link to a CSS property or descriptor, or {{foo}} to link to an IDL construct. These are documented in § 7 Autolinking.

Other types of linking shorthand also exist, such as [[#foo]] to link to sections of a spec (documented in § 7.9 Section Links), or [[FOO]] to link to generate and link to bibliography entries (documented in § 8 Bibliography).

5.4.1. Autolink Shortcuts That Work Anywhere: the l element

The autolink shorthands don’t work in elements like pre, because it’s too easy for code constructs to look like autolinks. You can still manually write out your autolink as an a element, but that can be much more verbose than the terse syntax that the shorthands provide. (For example, {{Foo/bar()}} vs <a method for=Foo>bar()</a>.)

You can avoid this by using the l element, and putting an autolink shorthand inside of it, like <l>{{Foo/bar()}}</l>. This will autolink exactly the same as normal, but works anywhere in your document. Extra bonus: it will recognize all the autolinking syntaxes, even if you have some turned off with the Markup Shorthands metadata.

Note: Currently the CSS value autolinking syntax (''foo'') does not work.

The l element itself is removed from the document and replaced by the autolink, but if you specify any attributes on the l, they’ll be transferred to the generated a. In particular, this is useful for specifying linking attributes that can’t be specified by the shorthand, like the spec to search for, that would otherwise require converting all the way to a manual a autolink.

5.5. var and Algorithms

The var element (or its shorthand equivalent, |foo|) is often used to mark up "arguments" to a prose algorithm. Bikeshed explicitly recognizes this, and has several features related to this.

Algorithms can be explicitly indicated in your markup by putting the algorithm attribute on a container element or a heading. All vars within an algorithm are "scoped" to that algorithm.

Note: Algorithms need to have a name. By default, it’ll infer one if there’s only a single dfn inside the algorithm container, but if there’s zero or multiple dfns, or you just want to give a clearer name, you can supply it as the value of the algorithm attribute, like algorithm="to foo a bar".

Generally, vars are used at least twice in an algorithm: once to define them, and at least once to actually use them for something. If you use a var only once, there’s a good chance it’s actually a typo. Bikeshed will emit a warning if it finds any vars used only once in an algorithm. If this singular usage is correct, you can instruct Bikeshed to ignore the error by adding an ignore attribute to the var itself. (There’s no way to do this with the |foo| syntax; you have to convert it back to an actual var element.)

5.6. Code Blocks (pre, etc)

Bikeshed has several nice features related to "code blocks"—​pre, xmp, and sometimes code.

5.6.1. pre whitespace stripping

Using a pre element in HTML is unsatisfying, because it forces you to break your indentation strategy, pulling the content back to the margin edge (or else employing silly hacks with comments and explicit newlines). The preprocessor fixes this.

Whenever a pre element is encountered, the processor records how much whitespace precedes the first line, and then strips that much whitespace from it and all following lines.

Additionally, if the closing </pre> is on its own line, the processor automatically pulls it up onto the end of the previous line, so there’s no final blank line in the content.

In other words, you can now write:

<div class='example'>
    An example:


The preprocessor will automatically convert it into:

<div class='example'>
    An example:


5.6.2. xmp To Avoid Escaping Markup

The xmp element is an old HTML element that’s now deprecated (but still required to be supported). It was intended for markup examples (thus the name), and has magical parsing properties, where its contents (everything until the </xmp> closing tag) are treated as literal text, and not interpreted as HTML. In particular, this means that within an xmp you don’t need to escape your < or & characters.

Bikeshed supports using xmp anywhere a pre can be used; it converts it into a properly-escaped pre in the output to avoid validation errors. For example, the two markup examples in the previous section are using xmp in this document’s source.

Use of xmp is particularly useful for IDL blocks, as IDL uses <foo> syntax for higher-order types (like sequence<DOMString>). If you’re using a pre and don’t remember to escape this, you’ll end up with confusingly-broken IDL, as the <DOMString> part is interpreted as an opening HTML tag. If you use xmp instead, there’s no need to remember to escape anything; you can write or copy raw WebIDL into the block and it’ll be interpreted correctly.

5.6.3. Syntax Highlighting

You can syntax-highlight code blocks. Just add either a highlight="foo" attribute or a lang-foo class to the element, or set a Default Highlight metadata, and the element will automatically be syntax-highlighted according to the "foo" language rules. (Setting no-highlight closer will turn if off for an element; it looks for the closest ancestor with one of the attributes.)

The syntax highlighter uses Pygments, which supports a large set of languages. See for the full list. (Use one of the "short names" of the language for the "foo" value.)

Bikeshed comes with a default color scheme based loosely on that of Prism.js. If you would like to use your own color scheme, turn off the automatic styles with a Boilerplate: style-syntax-highlighting off metadata, then supply your own.

Note: If you use highlight=html, script and style elements are automatically highlighted with JS and CSS rules. Normative WebIDL blocks (class=idl) are automatically highlighted, but you can use highlight=idl to highlight non-normative ones.

Note: If your code block already has markup in it, this feature will safely "merge" the highlighting into your existing markup.

5.6.4. Line Numbers

You can automatically generate line numbers for code blocks by adding a line-numbers attribute to the element or an ancestor, or setting the Line Numbers metadata to affect all elements. (Setting no-line-numbers closer will turn it off for an element; it looks for the closest ancestor with one of the attributes). The numbers are added to each line via CSS generated content, so they’ll always stay in sync and won’t get copied when you highlight a chunk of text.

Haikus are easy:five-seven-five syllables,spread over three lines.

The lines default to starting at 1; to change that, set the line-start attribute on the element to your desired starting value.

If you would like to produce your own line-numbering CSS, turn off the automatic styles with a Boilerplate: style-line-numbers off metadata, then supply your own.

Note: Similar to syntax highlighting, this feature will safely "merge" the line-numbers markup into your existing markup. The only complication is if the original markup has elements that span more than one line; in this case, the "line" element will expand to cover all the lines spanned by your original markup. It will still put a line marker on the first and last line of this span, so the effect isn’t too noticeable if the original markup only spans two lines.

5.6.5. Highlighting Particular Lines

If you want to bring attention to particular lines in your code snippet, but need to show more surrounding code for context, you can highlight the important ones with a line-highlight attribute.

Another haiku.This is the important line.Ignore what’s down here.

The syntax of line-highlight is a comma-separated sequence of either integers (like 5), or integer ranges (like 2-4). For example, line-highlight="2-4, 6" will highlight lines 2, 3, 4, and 6.

Like line-numbers (above), this defaults to considering the first line of the snippet as line 1, but the numbering can be changed with the line-start attribute.

If you would like to produce your own line-highlighting CSS, turn off the automatic styles with a Boilerplate: style-line-highlighting off metadata, then supply your own.

Note: This feature has the same restriction on line-spanning elements as the line-numbers feature does. In particular, if any line being wrapped in a multi-line "line" element is highlighted, the entire "line" element gets highlighted.

5.7. Automatic ID Generation

If any heading, issue, or dfn element doesn’t have an id attribute, one will be automatically generated by the processor, to ensure it’s usable as a link target.

Heading IDs are generated directly from the text contents of the element, cleaning up the characters to be a valid id. This often isn’t the best for complex heading texts, so it’s not recommended to rely on this. (Bikeshed will warn you that it’s generating IDs, and suggest you supply one manually.)

If a heading changed significantly, so that you want to change the ID, but you want links to the old heading ID to still work, put the old ID in an oldids="" attribute on the heading element. If there are multiple, comma-separate them.

Issues (elements with class="issue") will generate IDs of the form "issue-###", where "###" is substring of a hash of the issue’s contents. This means that an issue’s ID will be stable against changes elsewhere in the document, including adding or removing issues above it in the source, but will change if you change the contents of the issue.

Definition IDs are also generated directly from the text contents of the element. Most definitions additionally get a prefix, such as "propdef-", to avoid clashes with other definitions.

If an automatically-generated ID would collide with any other ID, it’s automatically de-duped by appending a number to the end. This isn’t very pretty, so if you want to avoid it, supply an ID yourself.

5.8. Assertions

Bikeshed provides the ability to mark "assertions" that tests can refer to. The assertion can either be added with the fake element <assert> or by adding an assert attribute to a div. (e.g. <div assert>).

For the <assert> tag, it will be replaced by a <span> tag in the generated source. The <div> tag will pass through but the assert attribute is removed.

In both cases a unique ID is generated from the tags contents of the form "assert-###", where "###" is a substring of a hash of the assertions contents. This ensures that you have a unique ID that won’t change arbitrarily, but will change when the contents of the assertion change, making it easier to tell when a test might no longer be testing the assertion it points to (because it’s no longer pointing to a valid target at all!).

Giving IDs to important things in your document, like headings and definitions, is great, but of little use if people don’t know they can link to them. Bikeshed will automatically generate a "self-link" in the margin next to certain linkable elements which just links to the element, so people can click on the link and then just copy the URL from their address bar to get a link straight to what they care about.

Self-links are currently auto-generated for headings, definitions, and issues, and notes, examples, <li>s, and <dt>s that have been given IDs.

5.10. Remote Issues

As defined earlier, you can start a paragraph with Issue: to cause Bikeshed to automatically format it as an inline issue paragraph. You can also refer to remote issues, which are tracked in some other issue tracker. To do so, instead start your paragraph with Issue(###): , where the ### is some identifying value for the issue.

If the identifying value is of the form user/repo#number, Bikeshed assumes you are referring to GitHub repository, and points the issue at the corresponding issue.

If you have Repository set up to point to a GitHub repository (or it was auto-detected as such, because you’re working on the spec from within one), then a numeric identifying value is assumed to be an issue number for your repository.

Otherwise, you need to tell Bikeshed how to convert the identifying value into a remote issue link. Specify a format string in the Issue Tracker Template metadata, with a {0} in the place where the identifying value should go. Bikeshed will then point the issue at the generated url.

5.11. Including Other Files

Sometimes a spec is too large to easily work with in one file. Sometimes there’s lots of repetitive markup that only changes in a few standard ways. For whatever reason, Bikeshed has the ability to include additional files directly into your spec with a <pre class=include> block:

<pre class=include>
path: relative/to/spec/location

The included document is parsed just like if it were written in locally, except that metadata blocks aren’t processed. (For various reasons, they have to be parsed before any other processing occurs). This means that the include file can use markdown, data blocks of various kinds (<pre class=anchors>, <pre class=railroad>, etc), and both provide definitions for the outer document and refer to ones defined by the outer document.

Note: All include paths are processed relative to the outermost spec document’s location. If you’re including a document in another folder, and it includes a document itself, take care that your paths are correctly relative to the outermost spec, not to the included file.

If you’re including a block of repetitive markup multiple times, and want to vary how it’s displayed, you can pass additional "local" text macros in the block, which are valid only inside the included file:

<pre class=include>
  foo: bar
  baz: qux qux qux

With the above code, you can use [FOO] and [BAZ] macros inside the include file, and they’ll be substituted with "bar" and "qux qux qux", respectively. (Remember that you can mark text macros as optional by appending a ?, like [FOO?], in which case they’ll be replaced with the empty string if Bikeshed can’t find a definition.)

5.11.1. Including Code Files

While it’s easy to include short code snippets in your document inline with an ordinary <pre> block (particularly with a highlight attribute so it’s formatted nicely), you sometimes want to instead include snippets of a larger external file. (For example, so you can keep that file up-to-date easily, without having to go in and fiddle with all the chunks you’ve inlined into your document.)

Normal includes don’t work for this, as they include the whole file and assume that it’s more Bikeshed code; instead, use <pre class=include-code> to include part (or all) of a file and have it rendered as highlightable source code.

<pre class=include-code>
path: bigFile.cpp
highlight: c++
show: 100-120

This snippet will create a nicely highlighted code block containing lines 100-120 of the file bigFile.cpp.

You can additionally include any of the more advanced highlighting options (§ 5.6.4 Line Numbers, § 5.6.5 Highlighting Particular Lines), using the same names and value syntax as you would use when specifying them as attributes:

<pre class=include-code>
path: someFile.cpp
highlight: c++
line-highlight: 2-5, 10, 12

If any of your options cause line-numbers to show up (using line-numbers or line-highlight), and you’re showing a snippet of the larger file with show, the numbering will default to using the numbering of the original file (so if you said show: 6-10, the first visible line will be numbered "6"). You can override this with line-start, as usual.

Note that, unlike <pre class=include>, there is no macros option.

5.11.2. Including Non-Bikeshed HTML/SVG/etc

<pre class=include> lets you insert chunks of Bikeshed source code into a larger outer document. But what if you want to include some markup that isn’t formatted as Bikeshed code? For example, SVG outputted by a design tool might not conform to Bikeshed’s syntax—​it might not indent properly, or it might use markup that will get messed up if parsed as Markdown, etc.

To avoid this, you can use <pre class=include-raw>. Just like <pre class=include>, it takes a single path line:

<pre class=include-raw>
path: myDiagram.svg

…and replaces itself with that markup directly, without doing any additional Bikeshed syntax processing.

Note: Unlike <pre class=include>, there are no local macros, as macro substitution is part of Bikeshed’s syntax processing, which is skipped here.

Note: Most parts of later Bikeshed processing do work, such as linking to definitions, etc., because inclusions are processed very early in Bikeshed’s transformation pipeline. However, if you’re writing something that’s intended to use Bikeshed’s features, you should really write Bikeshed source code and use <pre class=include>, in case some things move to being part of syntax processing. For example, inline autolinking shorthands currently aren’t, and so will work in include-raw substitutions, but that will change in the future.

5.12. Tracking Vectors

A tracking vector can be annotated with the tracking-vector attribute on the enclosing element. This will result in a link containing an image being prepended to the element.

<p tracking-vector>This is an example of an element annotated with that attribute.

There is a tracking vector here. This is an example of an element annotated with that attribute.

The various "Tracking Vector *" metadata keys can be used to customize the appearance of this annotation, though it’s encouraged to leave those in their default state for a consistent appearance across standards.

5.13. Image Size Detection

If an img element has a neither a width nor a height attribute, Bikeshed will automatically detect these dimensions from the image file and insert both attributes into the resulting markup.

Given this markup:
<img alt="…" src="img.jpg">

if the img.jpg file is 300 pixels wide and 100 pixels tall, bikeshed will generate the following output:

<img alt="…" src="img.jpg" width="300" height="100">

Bikeshed will also detect and insert the dimensions of the image if the element has no src attribute but does have a srcset attribute. In that case, it accepts a narrow subset of the srcset syntax: a single image source URL followed by a single pixel density descriptor, limited to integers. Since the src attribute is mandatory in HTML, bikeshed will also add it to the output.

In order to specify high resolution images, use the following syntax:
<img alt="…" srcset="highDPI.jpg 2x">

If the highDPI.jpg file is 600 pixels wide and 200 pixels tall, bikeshed will generate the following output:

<img alt="…" src="highDPI.jpg" srcset="highDPI.jpg 2x" width="300" height="100">

Note: Bikeshed will only successfully detect the size of local images, and will not attempt to fetch resources over the network, as there is no way for Bikeshed to know whether the size it would get is stable. Using local images is generally preferred, but if you need to use a remote one, consider setting the width and height manually.

If either of the width and height attributes is already set (or both are), or if both the src and srcset attributes are set, Bikeshed will assume you know what you’re doing, skip detection, and leave the element unchanged.

To opt out of this automated size detection for a single image, set the boolean no-autosize attribute on it. To opt out for the entire document, specify the Image Auto Size metadata with the value false.

6. Definitions

Defining a term is as easy as wrapping a dfn element around it. Most of the time, this is all you’ll need to do - the definition automatically gains an id, and is usually automatically exposed as an autolink target for local and cross-spec autolinks.

Autolinking is a special mechanism in the processor to let you link terms to their definitions without having to explicitly provide a url. Instead, the text of the link is matched against the text of the definitions, and if a match is found, the link’s href is set up to connect the two.

6.1. Conjugating/Pluralizing/etc the Linking Text

Bikeshed can automatically handle a wide range of English conjugations and pluralizations. For example, if you define the term "snap", you can link to it with "snapping" or "snapped" without having to manually add those variations to your dfn manually.

As such, it’s best to define your term in the "base" form, singular and present tense. Use lt='...' if necessary to set up the correct "base" linking text, if your visible text needs to be in a conjugated form due to the surrounding text.

These variations only work for the first or the last word in a phrase, so it can conjugate "activate the widget" into "activating the widget", "activate the widgets", or "activating the widgets". If you have a longer phrase where it’s a middle word that conjugates differently, you do still have to manually handle that, either by defining multiple linking texts on the dfn, or by manually specifying the linking text on the a.

In addition to English variations, these "conjugations" handle some IDL variations as well. For example, IDL terms that match an IDL keyword must be defined in IDL with a leading underscore, to avoid grammatical ambiguities, but actually define the term without the underscore. You can link to the term with either syntax.

Similarly, because methods and attributes technically live in the same namespace, it’s safe to link to a method without the parens after its name. (This is useful when linking to a method in the middle of prose, when you might be providing sample arguments that would interfere with linking, or linking to things in the arguments, which would interact badly with the arguments themselves being part of the method’s link text.)

6.2. Changing the Linking Text

Sometimes, the text of the definition isn’t exactly what you want it to be linked by, or you may want it to be linkable with more than one phrase. For example, an algorithm named "Check if three characters would start an identifier" may also want to be linkable from the phrase "starts with an identifier". To alter the linking text, simply add an lt attribute (for "Linking Text") to the definition; the linking text is used instead of the text content. You can separate multiple linking phrases by separating them with the pipe "|" character.

6.3. Defining Extra-Short "Local" Linking Texts

Sometimes you want to use an extra-short version of a term for within a spec, but don’t want to confuse things by exporting it globally. To achieve this, add a local-lt attribute with the terms you want to be only usable within the spec; the syntax is identical to that of the lt attribute, described above.

Using local linking text does not disturb the normal linking-text process; that still takes from either the element text or the lt attribute, as normal.

6.4. Definition Types

All definitions have a definition type. This allows for "namespacing" of the linking text, so you can define, for example, both a property and a term with the same linking text, such as "direction" or "color".

There are several types for CSS values:

There are additional types for WebIDL definitions:

And for HTML/SVG/etc element definitions:

A special type for URL schemes, like "http" or "blob":

A special type for HTTP headers:

A special type just for definitions of operators used in grammar definitions, like || and similar:

A special type for browser permissions:

And finally, some categories for "English" terms:

The processor will attempt to infer your definition type from the context and text content of the definition:

(This auto-detection is obviously skewed towards CSS types; Bikeshed started as a CSS spec preprocessor, and the CSS types are easier to auto-detect syntactically than anything else.)

Note that this auto-detection is a last-resort operation. There are methods (defined below) to explicitly indicate what type a definition is, and those win over the auto-detection.

If your value doesn’t fit one of these categories, you’ll have to tag it manually. Just add the type as a boolean attribute to the definition, like

attribute DOMString <dfn attribute for=Foo>name</dfn>;

Alternately, if you’ve got several definitions of the same type that share some container element (such as a <pre> or <dl>), just add a dfn-type="type-goes-here" attribute to the container. Anything which isn’t explicitly tagged otherwise will take that type by default.

(There are more methods to determine definition type, but they’re only meant for legacy content, and so are not documented here.)

6.5. Namespacing a Definition

Some types of definitions are defined relative to a higher construct, such as values for a particular property, or attributes of a particular IDL interface. This is useful, as it means these names don’t have to be globally unique, but that means your autolinks may have a hard time telling which name you intend to link to.

To fix this, the processor enforces that some types of definitions must define what they are for. This is specified with a for attribute on the definition.


Just like with the definition type, you can instead declare what several definitions are for by putting an attribute on a container. In this case, just add dfn-for to the container. This is especially useful for property/descriptor values, as they’re usually defined in a dl, or IDL definitions, as you can just put a dfn-for on the <pre class='idl'>.

If a single definition is "for" multiple things, you can provide a comma-separated list of values in the attribute.

6.6. Exporting Definitions

Most definitions are automatically "exported", which means they’re made available for other specs to autolink to. The only exception is "dfn" type definitions, which aren’t exported by default.

To force a link to be exported, add an export boolean attribute to it. To force a link not to be exported, add a noexport boolean attribute instead. Like the other attributes, you can instead add this to a container to have it be a default for the definitions inside.

6.7. Link Attributes On Ancestors

All of the dfn attributes can be specified on an ancestor of the dfn instead, if you have a group of dfns that all share a particular attribute.

They can be spelled out fully (like data-dfn-type), or without the data- prefix, like dfn-type or dfn-for. As there’s no overlap with a similar data-link-* attribute, dfn-force can omit dfn- as well, being spelled as merely force and spec.

A link will take the "closest" version of each attribute it can find, so specifying an attribute directly on the link will override anything coming from an ancestor.

6.8. Providing Custom Definitions

If you want to link to dfns in specs that aren’t yet part of the autolinking database, you can provide your own definition data that Bikeshed can use. Within a <pre class='anchors'> element, define the anchors you need in InfoTree format, with the following keys:

To generate the url for the anchor, first all of the urlPrefix entries are concatenated. If a url is provided, it’s appended to the prefixes; otherwise, the text is url-ified and appended. (Lowercased, spaces converted to dashes, non-alphanumeric characters dropped.) If neither urlPrefix nor url had a "#" character in them, one is inserted between them.

The spec attribute is used only for index generation, and has no effect on URL generation.


<pre class="anchors">
urlPrefix:; type: dfn; spec: ENCODING
  text: ascii whitespace
  text: decoder
url:; type: dfn; text: toascii

<a>ascii whitespace</a> links now!

Alternately, this data can be provided in a file named anchors.bsdata, in the same folder as the spec source, but this prevents you from using the web service.

6.9. Definitions data model

Bikeshed’s most important feature is its powerful cross-spec autolinking. This is possible due to each definition being annotated with a rich set of metadata, which is then exposed via custom attributes and picked up by specialized scrapers (such as Shepherd) that then compile a definition database that Bikeshed relies on.

If you’re writing a spec processor or related tool and would like to interoperate with the Bikeshed ecosystem, here’s the full definition data model and how to properly expose it.

  1. The defining element MUST be a dfn or h2...h6. No other element is recognized as defining a term.

  2. The element MUST have an id attribute.

  3. The linking text defaults to the text content of the dfn/heading. If the desired text content isn’t suitable for linking text, or you wish to provide multiple linking texts, a [lt attribute](#changing-lt) containing one or more pipe-separated linking texts will override the text content. See also § 6.3 Defining Extra-Short "Local" Linking Texts

  4. data-dfn-type MUST be provided, and set one of the accepted values.

    (In an actual Bikeshed document, this can be omitted from dfn elements and it will be inferred from context, usually defaulting to "dfn" type. But non-Bikeshed documents must specify this, or else they’ll get their dfn type inferred heuristically, which may or may not be correct.)

  5. Either data-export or data-noexport MAY be provided (both boolean attributes). If neither is provided, "dfn" type definitions default to noexport, while all others default to export. Unexported definitions aren’t linkable by default.

  6. Several types of definitions are namespaced to another construct; for example, attribute names are namespaced to an interface. These definitions MUST contain a data-dfn-for attribute, containing a comma-separated list of one or more definitions they’re namespaced to.

If you have written a web spec and it conforms to this definition syntax, contact the project maintainer and ask them to register your spec in Shepherd, so its definitions will be available to everyone else.

7. Autolinking

The processor supports "autolinks" for easy linking to terms without having to fiddle around with urls. Instead, just match up the text of the link to the text of the definition!

There are three types of autolink syntaxes you might see/use:

  1. Dfn autolinks, like {{foo}} or [=foo=], which link to a definition. See § 7.1 Dfn Autolinks for details.

  2. Spec/section autolinks, like [[FOO]] or [[#foo]], which link to a document or heading. See § 7.2 Spec/Section Autolinks for details.

  3. Manual a elements without an href attribute, which can create either of the preceding two types, and are occasionally useful when the preceding syntaxes can’t express what you need. See § 7.3 Manual Autolinks for details.

You may also see an l element, which is used to help augment dfn autolinks, and can avoid the need to use an a in many cases. See § 5.4.1 Autolink Shortcuts That Work Anywhere: the l element for details.

Dfn autolinks create links to definitions.

There are two parts to the syntax:

  1. The wrapper, like {{...}} or [=...=], which specifies what type of definition you’re linking to.

  2. The inside syntax, which gives details on what you’re trying to link to.

The characters used to wrap an autolink determine what type of definition (see § 6.4 Definition Types) you’re trying to link to. They’re arranged into categories that can be individually turned on or off, if you don’t think you’ll use a particular type of autolink, and want to avoid possibly triggering an autolink accidentally:

The Dfn varieties (controlled by Markup Shorthands: dfn yes):

The HTTP varieties (controlled by Markup Shorthands: http yes):

The IDL variety (controlled by Markup Shorthands: idl yes):

The markup (HTML/SVG/etc) varieties (controlled by Markup Shorthands: markup yes):

The CSS varieties (controlled by Markup Shorthands: css yes):

All together, the order of pieces is:

  1. for value

  2. linking text

  3. link type

  4. visible text

like {{Document/ready!!attribute|the Document.ready attribute}}.

Biblio/spec/section autolinks create links to other documents as a whole, or sections of those documents, rather than to particular definitions. They come in three closely-related syntaxes:

  1. Biblio/document autolinks, like [[FOO]] or [[!FOO]], which specify the shortname of a document (from SpecRef). See § 8 Bibliography for more details.

  2. Local section autolinks, like [[#foo]], which link to another section in your current document by its ID.

  3. Cross-spec section autolinks, like [[FOO#bar]], which link to a section of another document, and mixes the previous two syntaxes.

See § 7.9 Section Links for details on the latter two.

Like dfn autolinks, if you want to display a specific visual text rather than what the autolink would normally display, you can append |visible text to the internals, like [[CSS-FOO|the CSS Foo spec]].

These are all controlled by Markup Shorthands: biblio yes.

If one of the shorthand syntaxes don’t work for whatever reason, even if you use the l element, you can use an ordinary a element to specify a manual autolink.

Any a without an href attribute is an autolink.

All together, it might look like <a attribute for=Document lt=ready>the Document.ready attribute</a>.

Links have the same types as definitions, with a few additional "union" types that are used by the shortcut forms. While you shouldn’t specify them explicitly, they’ll show up in error messages sometimes, so here’s a list of them:

When you actually run the processor, you may get errors about there being too many possible references to choose from. The processor will continue to run anyway, but its default choice might be the wrong definition. There are three things you might have to do to fix these:

  1. Specify the type explicitly, if the link isn’t being processed as the correct type. Like definitions, this can be done by just adding the type as a boolean attribute on the link, or by adding a link-for attribute to a container. If the link is using shorthand syntax, you can use the !!type suffix to specify the type.

  2. If the link type corresponds to one of the definition types that needs for to be specified, you may need to specify for on the link as well to narrow down which definition you’re referring to. For example, many CSS properties define an "auto" value; to link to the "auto" value of the 'width' property in particular, specify <a value for=width>auto</a>, or the shorthand syntax ''width/auto''. To refer to a value of a descriptor, you can be completely explicit and specify the at-rule as well, like <a value for='@counter-style/system'>numeric</a>, but you’re allowed to omit the at-rule if there are no other properties or descriptors with the same name, like ''system/numeric''. This might trigger errors in the future if a conflicting term gets added later, but it keeps your links shorter for now.

    Again, you can specify a link-for attribute on a container to default it for all the autolinks inside the container. Alternately, you can specify link-for-hint on a container, which’ll use the hint as the for value if possible (if doing so wouldn’t eliminate all the possible links). This is useful if some container has a bunch of links for a given property, say, but some of the links are to other things entirely; using link-for means you have to manually specify the other links aren’t for anything, but link-for-hint is more "do what I mean".

  3. If multiple specs define the same term, you may need to declare which spec you’re referring to. (The processor is smart enough to automatically figure out which one you probably want in many cases.) Just add a spec attribute with the spec’s shortname to either the link or a container. This can also be specified document-wide, as described in § 7.7 Configuring Linking Defaults. (You can add this to a dfn autolink by wrapping it in an l element, or by fully converting to an a element, and adding a spec attribute.)

As a final note, the autolinking algorithm will link differently based on whether the spec being processed is a "current" (up-to-date) or "snapshot" (generated for a past date) draft. If "current" (ED, UD, etc.), it’ll prefer to link to other current drafts, and will only link to "snapshot" if no "current" version of that spec exists. (If a definition only exists in the "snapshot" draft but not the "current" draft, that almost certainly means it’s been deleted since the "snapshot" draft was last published, and thus shouldn’t be linked to.) On the other hand, "official" (WD, CR, etc.) specs will preferentially link to other official specs. A future version of the processor will likely enforce the W3C’s linking policy more strongly: preventing CRs from linking to EDs at all, preventing RECs from linking to anything below CR, etc.

If you need to override the processor’s choice for which status to link to for a particular link, provide a status attribute containing either "ED" or "TR" on the link or a container.

7.5. Linking to Unexported Definitions

Most definition types are automatically exported and made available for cross-linking, but "dfn" type definitions aren’t, because specs often define terms for their own internal use that aren’t meant to be used outside the spec (and in particular, aren’t named in a way so as to avoid collisions).

If a spec contains a "dfn" type definition that you want to link to, but it’s not marked for export (either intentionally, or because it was accidentally missed and fixing the spec would be time-consuming), using the spec attribute (defined above) will override the lack of an export declaration, and go ahead and link to it anyway.

All of the autolinking attributes can be specified on an ancestor of the autolink instead, if you have a group of links that all share a particular attribute.

They can be spelled out fully (like data-link-type), or without the data- prefix, like link-type. As there’s no overlap with a similar data-dfn-* attribute, link-spec and link-status can omit link- as well, being spelled as merely status and spec.

A link will take the "closest" version of each attribute it can find, so specifying an attribute directly on the link will override anything coming from an ancestor.

When there are multiple definitions for a given term and Bikeshed can’t automatically tell which one you want, it’ll emit a warning asking you to specify more explicitly. You can do this per-link, but you typically want to make the same choice every time the term is autolinked; this can be done by adding a <pre class='link-defaults'> block, written in the InfoTree format. Each piece of info must have a spec, type, and text line, and optionally a for line if necessary to further disambiguate.

Sometimes this is too fine-grained, and you’d actually like to completely ignore a given spec when autolinking, always preferring to link to something else. To do this, add a <pre class='ignored-specs'> block, written in the InfoTree format. Each piece of info must have a spec line, and optionally a replacedBy line, both naming specs. If the info has just a spec line, that spec is ignored totally by default; linking to it requires you to manually specify a spec="" attribute on the autolink. If the info has a replacedBy line, then whenever an autolink has a choice between the two specs, it’ll delete the spec value from consideration, leaving only the replacedBy value (plus any other specs that might be providing a definition).

7.8. Potentially-ambiguous for-less Links

If you were directed here by a Bikeshed error message, put an explicit for value on your autolink. (Use for="/" to link to a definition without a for value of its own).

Bikeshed’s autolinking functions as a series of filters, some of which are optional. For example, if you don’t specify a for value, then Bikeshed simply doesn’t care about the for values of definitions as it searches. However, this doesn’t always match people’s mental models—​in particular, people sometimes implicitly assume that an autolink without a for will specifically match a definition without a for.

Usually this is fine; if there are multiple possible definitions, Bikeshed will just throw up a linking error informing you of how to be more specific. But if you have a definition of that term with a for value in your local spec (or your anchors block), Bikeshed can silently select that as the correct definition to link to, causing accidental spec corruption if you meant to link to a cross-spec definition without a for.

In other words, if some other spec defines <dfn>term</dfn>, and then your spec both defines <dfn for=foo>term</dfn> and links to <a>term</a> (expecting the link to go to the cross-spec definition), you’ll be disappointed, but won’t know it’s wrong unless you manually check your links.

Bikeshed looks out for this situation, and flags it as a potentially-ambiguous link, requiring you to specify the for value explicitly. If you want to link to the for-less cross-spec definition, simply add for="/" to your autolink, to indicate explicitly that you want the definition without a for value. Otherwise, add the appropriate for value as normal.

Sometimes you want to link to a section of a document, rather than a specific definition. Bikeshed has section links to handle this case more easily:


renders as:

<a href="#heading-id">§6.1 The Example Section</a>

Note that this is quite different from normal autolinks; rather than matching on text and letting Bikeshed fill in the href, you match on href and let Bikeshed fill in the text. This is because section titles change much more often than definition texts, so using text-based matching is fragile; on the other hand, their IDs tend to be stable, as they’re often linked to. Also, the section titles are often long and annoying to type, and they move around, so numbering isn’t stable.

You can also use cross-spec section links, as long as the spec is either in Bikeshed’s linking database, or the biblio database. The syntax is a mixture of a biblio reference and a section link:


which renders as:

<a href="">CSS Flexbox 1 §8.1 Aligning with auto margins</a>
<a href="">Accessibility Features of CSS §Features</a>

If Bikeshed knows about the spec, it link-checks you, and fills in the section number and heading in the generated text. If the spec is only in the bibliography database, Bikeshed just assumes that the link target exists and uses it directly in the text, because it has no way to tell what the section is named.

If the spec is multipage, like SVG, and Bikeshed knows about it, most of the time you don’t need to do anything different - Bikeshed will find the correct page for the heading you’re linking to. On the rare occasions that the same heading id exists in multiple pages of the same spec, tho, specify the page like [[svg/intro#toc]] (which indicates the #toc heading on the intro.html page). If the desired heading is on the top-level page, use an empty page name, like [[html/#living-standard]]. In any case, Bikeshed will throw an error, and tell you what names it knows about so you can easily correct your link.

8. Bibliography

Bibliographical references form a special class of autolinks. They’re typically added only via the shorthands [[FOO]] for informative references and [[!FOO]] for normative references.

(A "normative reference" is a reference that your document depends on; something in the other document is required in order to implement your document correctly. An "informative" reference is anything else. Informative references might be to normative documents, if your feature doesn’t actually depend on the referenced document.)

Unlike regular autolinks, which link to dfn elements, biblio autolinks cause the spec to generate entries in its "References" section, and then link to that instead.

The bibliography database is completely separate from the autolinking database, and comes from multiple sources. The default data comes from the SpecRef project and the CSSWG’s own biblio file (preferring SpecRef’s information when the same name appears in both).

8.1. Biblio Link Modifiers

Several aspects of how bibliography autolinks work can be controlled by adding keywords to the autolink, after the biblio name.

All of these keywords can be combined, in any order, such as [[FOO inline obsolete]] or [[FOO index current]].

8.2. Autolinking and Automatic Bibliography Entries

Whenever you autolink to a term from another spec using a standard dfn autolink, Bikeshed automatically adds a bibliography entry for that document if one doesn’t already exist.

By default, it assumes these are normative references. However, if the link is in a "non-normative" element, Bikeshed doesn’t put it in the bibliography. This includes any notes (class=note), examples (class=example), or sections manually marked with class=non-normative. (You can override this and get a normative reference by adding a class=normative to the link or an ancestor.)

8.3. Manual Biblio Links

If, for whatever reason, you need to craft a bibliography link manually, just create an a element with data-link-type=biblio and the biblio tag either in the text content of the link or in a data-lt attribute, similar to a manually-crafted autolink.

If you need one of the biblio link modifiers, add data-biblio-type=[normative | informative], data-biblio-status=[current | snapshot], data-biblio-display=[index | inline], and/or data-biblio-obsolete="" to the link as well.

8.4. Custom Biblio Entries

You can also add your own bibliography data, following the SpecRef JSON format:

  "foo-bar": {
    "authors": [
      "Tab Atkins",
      "Bat Snikta"
    "href": "",
    "title": "Foo Bar Level 1",
    "status": "CR",
    "publisher": "W3C",
    "deliveredBy": [

This would allow you to write [[FOO-BAR]] into your document, and have it be recognized as a valid autolink, with a bibliography entry generated in the index just like any of the predefined biblio references.

Note: Only the "title" field is strictly necessary; the rest can be omitted if desired.

This JSON should be inline, in a <pre class=biblio> block. It can also be in a biblio.json file in the same folder as the Bikeshed file, but this is incompatible with the web service.

If you find yourself adding this to more than one document, it might be worth adding to SpecRef itself so others can benefit from it as well; file an issue on that project if you’re interested.

9. Definition Tables

Several types of definitions have well-established presentations, as a simple "key/value" table providing all of the details in a standardized format.

Bikeshed handles these specially, with a dedicated set of <pre class=xxxdef> blocks. The information in the block is presented in a simple line-based Key: value format, similar to <pre class=metadata>, and this is transformed into a standardized table.

All of the def tables look like this:

<pre class='propdef'>
Name: flex-basis
Value: content | <<'width'>>
Initial: auto
Inherited: no
Applies to: <a>flex items</a>
Computed value: as specified, with lengths made absolute
Percentages: relative to the <a>flex container's</a> inner <a>main size</a>
Animation type: length
Canonical Order: per grammar

9.1. Arbitrary Def Tables

If you want to display your own key/value rows in a similar way to the built-in types defined below, you can use a plain <pre class=simpledef> block.

<pre class=simpledef>
first row: some info
second row: different info
second row: merged with the previous row

will render as:

first row some info
second row different info merged with the previous row

Parsing is extremely simplistic:

9.2. CSS Property Tables

CSS properties are defined with a <pre class="propdef"> block.

Name, Value, Initial, and Inherited are all required.

All the other keys listed in the block above are optional. If left unspecified, they’ll be given a usually-reasonable default value:

All values are placed directly into the output HTML as written, so HTML tags, Bikeshed autolinks, etc can all be used as normal.

If defining a shorthand property, put a "shorthand" class on the block. This will change the block to only require Name and Value; Canonical Order is still allowed, but continues to default to "per grammar" as usual. All other keys will default to "see individual properties", which is usually all you need.

If adding some values to a property, rather than defining the property itself, put a "partial" class on the block. It will now require a New Values key, rather than Value, but is otherwise normal.

Additional keys beyond what is described here can be provided; they’re placed after all the built-in keys.

The end result looks like:

Name: flex-basis
Value: content | <<'width'>>
Initial: auto
Applies to: flex items
Inherited: no
Percentages: relative to the flex container’s inner main size
Computed value: as specified, with lengths made absolute
Canonical order: per grammar
Animation type: length

9.3. CSS Descriptor Tables

CSS descriptors (like properties, but for at-rules) have def blocks similar to, but simpler than, propdef blocks.

Name, Value, and Initial are required keys, just like propdef blocks. Additionally, For is required, giving the name of the at-rule the descriptor is for, like For: @counter-style. The rest of the propdef keys are not meaningful for descriptors.

Like propdef blocks, a "partial" class can be put on the block to indicate that you’re adding values to an existing definition; Name, For, and New Values must be provided.

While not technically descriptors, media features can be defined with this block as well, if an "mq" class is put on the block. It then only requires Name, For, and Value.

Additional keys beyond what is described here can be provided; they’re placed after all the built-in keys.

9.4. HTML Element Tables

HTML/SVG/etc elements can use a <pre class="elementdef"> block. (The content and presentation of this block is similar to that in the HTML standard, but is closer to that in the SVG standard.)

Additional keys beyond what is described here can be provided; they’re placed after all the built-in keys.

9.5. IDL Method Argument Tables

IDL methods can have their arguments defined & documented in a <pre class="argumentdef"> block:

<pre class=idl>
  interface Foo {
    undefined bar(long arg1, DOMString? arg2);

<pre class=argumentdef for="Foo/bar()">
arg1: here's the first arg
arg2: the second arg is different!

The argumentdef block must specify what method the arguments are for in a for attribute on the <pre>. The contents of the element must then, on each line, list the argument name, a colon, then a description of that argument.

It will produce a table listing each argument, its description, and automatically contain its type, nullability, and optionality (determined from the IDL block), like:

Arguments for the method.
Parameter Type Nullable Optional Description
arg1 long here’s the first arg
arg2 DOMString? the second arg is different!

Note: This automatically wraps the argument names in a dfn; this is meant to be the canonical location of the argument’s definition.

10. WebIDL Processing

Bikeshed can automatically process IDL blocks, marking up all relevant terms for you without any intervention, setting up definitions and autolinks as appropriate.

To activate this behavior, simply place the IDL in the <pre class='idl'> element. Bikeshed will consume the text content of the element (ignoring any markup you may currently have) and replace it with marked-up text containing dfn and a elements.

As mentioned in § 5.6.2 xmp To Avoid Escaping Markup, you can use <xmp class=idl> as well, so you don’t have to escape your sequence<Foo>/etc types. This prevents you from using other markup in your IDL, but that’s relatively rare.

For IDL specifically, you can also use <script type=idl>. This has similar effects to <xmp>, but it interacts better with the syntax-highlighting of some text editors, preventing the contents from being highlighted as HTML.

In the process of doing this, Bikeshed will also syntax-check your IDL, and report fatal errors for any mistakes. Bikeshed’s IDL parser, courtesy of Peter Linss, is intended to be forward-compatible with IDL changes, gracefully emitting unknown constructs unchanged and recovering as well as it can. If anything isn’t recognized when it should be, or the parser fails in a major, non-graceful way, please report it as an issue.

10.1. Putting Definitions Elsewhere

Quite often, you may want to have the actual definition of an IDL term (the thing that Bikeshed actually links to) somewhere in your prose near the full definition, rather than being in the IDL block.

Bikeshed will automatically produce an a in your IDL, rather than a dfn, if it can find a pre-existing definition of that IDL term, including local definitions in the current spec. However, you have to mark up the definition correctly to get this to work, or else Bikeshed will fail to recognize there’s an external definition and will mark up the IDL with a dfn as well.

In particular, method and attribute definitions need to have their for value set to the interface they’re a part of (and similar with dictionary members). Methods have some further complexity - they should have their definition text set to contain the names of all their arguments.

For example, take the following example IDL:

interface Foo {
  undefined bar(DOMString baz, optional long qux);

To have Bikeshed recognize a definition for the bar() method placed elsewhere, it must look something like <dfn method for=Foo title="bar(baz, qux)">bar(DOMString baz, optional long qux)</dfn>.

Additionally, it should define alternate linking texts for omittable arguments, like <dfn method for=Foo title="bar(baz, qux)|bar(baz)">bar(DOMString baz, optional long qux)</dfn>. This way any valid call signature can be used to autolink. Note that arguments are omittable if they’re marked with optional, or are variadic (like long... qux), or have a default value. Nullable arguments (like long? qux) are not omittable. (If you are fine with the dfn being in the IDL block, Bikeshed will do all of this for you.)

Unless all arguments can be omitted, the definition text should not have an alternative with empty args. For convenience, however, Bikeshed will allow autolinks with empty argument lists to work, as long as it can resolve the link unambiguously. For example, {{Foo/bar()}} will autolink to the method defined above, despite it not being a valid call signature, as long as there isn’t an overload of bar() that it might also apply to.

(The above applies to all functionish types: method, constructor, stringifier, etc.)

Marking up argument definitions is similar. To mark up the baz argument of the above method, for example, do <dfn argument for="Foo/bar(baz, qux)">baz</dfn>. You should use the full call signature of the method.

10.1.1. Free Type/Etc Documentation on Attributes/Dict Members

Bikeshed automatically gathers information from your IDL block about attributes and dictionary members, such as their type, their default value, and whether they’re readonly or nullable. You can have Bikeshed automatically output that information in your definition elsewhere, by using a specific markup structure:

<dl dfn-type=attribute dfn-for=FooInterface>


If an attribute or dictionary member is defined in a dt element, and is the sole content of that element, Bikeshed will automatically add useful information about the attribute following it.

You can also invoke this manually, by adding a <span data-attribute-info for="FooInterface/fooAttr"></span> element anywhere in your spec (or data-dict-member-info for dictionary members).

10.2. Linking to Stringifiers

Linking to a stringifier is a little complicated, because WebIDL allows four different syntaxes for it.

The stringifier keyword itself is always linkable; it’s a "dfn" type definition with for=MyInterface and linking text "stringification behavior". Like any other IDL construct, you can instead define the term yourself in the same way, and the IDL will link to your definition instead, like <dfn dfn for=MyInterface>stringification behavior</dfn>. This is generally what you should use to link to the stringifier, as it’ll maintain the links even if you change which syntax form you use.

If you use the "stringifier attribute" form, like stringifier attribute DOMString href;, you can also just link/dfn the attribute as normal.

If you use the "stringifier method" form, like stringifier DOMString foo(long bar);, you can also just link/dfn the method as normal, like <dfn stringifier for=MyInterface>foo(bar)</dfn>. (Note that it’s a "stringifier" type definition, not "method".)

If you use the "anonymous stringifier method" form, like stringifier DOMString(long bar), you can still technically link/dfn it as a stringifier method. It doesn’t have a name, so we invent one - it’s called __stringifier__(), a la Python’s magic methods. (Note the two underscores on each side.) You should almost never need to do this; the only reason to need to specify the method name (rather than just linking to the keyword, as described above) is if you’re linking/dfning an argument to the method, and need to specify a for value for it.

10.3. Turning Off Processing

If for whatever reason you don’t want your IDL block to be processed by Bikeshed, simply use another element, or another class. If you really want to use <pre class=idl>, you can add a data-no-idl attribute to the element. Bikeshed will leave these elements alone.

Alternately, if your block is IDL, but it’s not meant to be taken literally (for example, if it shows an example attribute, then explains in prose the set of actual attribute names to be used, as in the _camel_cased_attribute in CSSOM), put a class=extract on it.

11. Testing Integration With WPT

When writing specifications for the open web, the canonical way to test new features and ensure interoperability is to contribute the testsuite to the Web Platform Tests project. Bikeshed has some useful tools to interact with WPT information.

11.1. Annotating Specs with Tests: the wpt element

When writing tests, you can sometimes link to the section of the spec that you’re testing, to make it easier to review. But when you’re actually reading (or updating!) the spec, you can’t tell what sections are tested or untested, or what tests might need to be updated due to a change you’re making. The wpt element lets you easily link to your WPT testcases inline, right next to the text they’re testing, and helps make sure that the testsuite and the spec are kept in-sync.

The wpt element is a block-level element for parsing purposes; place it after or between paragraphs/etc. The contents of the element are line-based: each line contains a single test path (the path under the WPT repo, so not including the domain or the /web-platforms-tests/wpt/) pointing to the test covering some nearby text.

For example:
Implementations must FOO whenever they would also BAR.


If all or most of your tests have a common path prefix, you can specify it in WPT Path Prefix metadata, and then leave it off of all the individual test lines. An explicit pathprefix attribute can also provide the path prefix, and will override the metadata if specified.

If the preceding example used WPT Path Prefix, the wpt element could be shorter:
<pre class=metadata>
WPT Display: open
WPT Path Prefix: /foo-spec/


Implementations must FOO whenever they would also BAR.


For debugging purposes, a single <wpt-rest> element can be used in a spec, which will act like a wpt element containing all the tests under the current path prefix that aren’t already specified in a wpt element. As this defeats Bikeshed’s ability to tell you when you’re missing any tests, Bikeshed will emit a warning as long as a wpt-rest element is in use.

Bikeshed’s knowledge of valid tests allows some useful checks:

By default, wpt elements don’t display in the output document; they’re intended to annotate the source file to help with spec authoring, and trigger Bikeshed’s checks. If you want to produce a test-annotated version of the output, specify the WPT Display metadata with the value "open" or "closed"; all of the wpt elements will become usefully-formatted lists of their contained tests, with links to, the live test on, and the source code on GitHub. The difference between "open" and "closed" is that with the former, all wpt elements are initially displayed in a collapsed state to avoid getting in the way of reading the document, while "open" causes them to be expanded initially. In either case, the use can interactively collapse or expand each element.

When WPT Display: open is set, the wpt element turns into a "TESTS" block.
When WPT Display: closed is set, the wpt element also turns into a "TESTS" block, but it is initially collapsed.

Note: For compatibility with earlier versions of Bikshed, "inline" is accepted as a synonym of "open".

When the hidden attribute is set on the wpt element, Bikeshed will take the tests listed in that element into account, but exclude them from the output, even when producing a test-annotated version.

Note: This can be useful when tests for all versions or levels of a single specification are stored together under a single directory. Using WPT Path Prefix would cause Bikeshed to complain if the tests are not listed, but the tests may only be relevant for a later version of the specification. In such cases, listing the tests in a <wpt hidden> element silences the warnings, without adding anything irrelevant to the output.

Information about a set of tests described in a wpt element can be provided to the reader by writing it in the title attribute of the wpt element. This will be inserted in the output as an introductory paragraph before the test list. A lang or dir set on the wpt element will apply to the introductory paragraph if there is one.

<h1 id=intro>Introduction</h1>

This specification introduces…

<wpt title="The following tests are crash tests
            that relate to general usage
            of the features described in this specification
            but are not tied to any particular normative statement.">

The test block in the example above would be rendered as follows:

A WPT element with a list of tests, preceded by an introductory paragraph

If multiple wpt follow each other in the source, with nothing between them other than whitespace, they will be fused together in the output, visually taking less space than if they remained as separate elements. This will work even if they have different pathprefix or title attributes. When merging elements with title attributes, the tests listed in each element will be preceded by the relevant introductory paragraph, and visually separated from tests listed in different wpt elements.

Note: More tools and options for integrating with WPT will be developed in the future.

11.2. Updating Lists of Tests

If Bikeshed reports that a test doesn’t exist, but you’re sure that it does (and have checked the spelling!), make sure your data files are up-to-date with bikeshed update (see § 3.6 bikeshed update) for details).

If you’ve just pushed a new test to WPT and Bikeshed’s update hasn’t picked it up yet, try doing a manual update with bikeshed update --wpt --skip-manifest.

12. Boilerplate Generation

The processor automatically generates nearly all of a spec’s boilerplate, the text that is repeated nearly identically across all specs.

Generally, you won’t need to understand what’s going on here in order to use the processor - it’ll just automatically do the right thing.

For help in creating new boilerplate files for your organization, see § 12.8 Creating New Boilerplate Files For Your Organization.

12.1. Groups

Much of the boilerplate is determined based on the Group metadata. If unspecified, it defaults to a generic set of boilerplate that is generally appropriate for most things, without making reference to any particular standards body or the like. However, to obtain correct boilerplate for a given standards body, "Group" can be used.

Several groups are already accommodated with appropriate inclusion files:

You can put whatever value you want into the "Group" value, though. Unrecognized values will just use the default boilerplate files. If you want to add specialized boilerplate files for your group, check out the File-Based Includes section, later in this document, and write your own files.

12.2. Rearranging and Excluding "Spec Metadata"

An important part of the boilerplate is the "spec-metadata" section. This will likely be at the top of your header boilerplate, as it contains a bunch of useful information about your spec.

Bikeshed generates a lot of these automatically for you, based on the metadata you provide and other things detected from your document, and you can supply "custom" items as specified near the end of § 4 Metadata. There’s a predefined ordering of these, but if you’d like a slightly different order, or to omit some of the automatically-generated items, you can use Metadata Include and Metadata Order to control this.

Metadata Include takes a comma-separated list of names and boolish values, where the names are the strings that show up in the dt in the spec metadata. Everything defaults to "on" currently, but you can explicitly turn them "off" to omit them from the spec metadata section.

Metadata Order instead controls the ordering of the spec metadata section. It’s a comma-separated list of names, where the names are the strings that show up in the dt in the spec metadata (same as Metadata Include). Two special "names" are recognized as well: the * name stands in for "all the standard keys that aren’t otherwise explicitly specified", while !* stands in for "all the custom keys that aren’t otherwise explicitly specified". (The default value is thus Metadata Order: *, !*, listing all the standard keys in the default order, followed by all the custom keys in the default order.)

If a name is specially pluralized when there are multiple entries (such as "Editor" vs "Editors"), use the singular version in either of these metadatas.

For example, Metadata Include: This version off will make Bikeshed omit the "This version" entry from the spec-metadata section, which is otherwise auto-generated by the ED metadata.

If you wanted to make sure that editors were listed before anything else, you could set Metadata Order: Editor, *, !*.

12.3. Text Macros

Several text "macros" are defined by the spec’s metadata, and can be used anywhere in the spec to substitute in the spec they stand for by using the syntax [FOO]. Note that this is similar to the syntax for bibliography references, but it has only a single set of [] characters, and the text must be uppercase. The following macros are defined:

As these are substituted at the text level, not the higher HTML level, you can use them anywhere, including in attribute values.

You can mark a macro as "optional" by appending a ? to its name, like [DATE?]. This will cause Bikeshed to just remove it (replace it with the empty string) if it can’t find a definition, rather than throwing an error.

Like most other markup shorthands, text macros can be "escaped" by prepending a backslash, like \[TITLE]. When Bikeshed sees this, it will remove the slash and leave the text alone. This is sometimes necessary when code examples in your doc (such as a regex) accidentally look like text macros.

12.4. Conditional Inclusion

If you have multiple boilerplate files for publishing with different document statuses (like WD vs CR), but the only difference is that some sections are added/omitted between them, you can greatly simplify things by instead making a generic boilerplate for everything, and then adding/removing the variant bits with a conditional inclusion.

To mark an element as "conditionally included", put an include-if and/or exclude-if attribute on it, each of which takes a comma-separated list of inclusion conditions.

An element that fails will be removed from the output entirely, along with its contents.

Inclusion conditions can be:

Normally the conditional attributes go on an existing element in your page, which’ll show up in the output assuming it passes the tests. If there’s not a natural wrapper element, or adding such an element would mess with the semantics or styling of the page, you can instead wrap the conditional content in an if-wrapper element.

The if-wrapper element is a standard block-level element that can contain any flow markup. It must have one of the conditional attributes. If it fails the conditions, it and its contents are removed from the document, as normal; if it passes the conditions, it is still removed from the document, but its children are left behind in its place.

if-wrapper is still parsed as a standard unknown "custom" element by the HTML parser, and so can only appear in places where such elements are allowed; it cannot, for example, be used in the head of a document, or as a child of table.

It’s also considered "block level" by the Markdown parser. It can be used inline in text (tho you probably want to instead make the entire block conditional, for readability/editability), but it won’t parse as intended if it’s the very first content in a paragraph.

12.5. Boilerplate Sections

The location of the boilerplate sections are indicated by elements with data-fill-with='' attributes. If the elements contain anything, they’re emptied before being filled with the appropriate boilerplate. The valid data-fill-with='' values are:

Additionally, "header" and "footer" boilerplate files are used to put content at the start and end of your document. Most or all of the above boilerplate sections should actually show up here, in the header and footer, rather than being manually specified in your source file.

12.5.1. Default Boilerplate

Some sections listed above are generated by default; if you don’t put an explicitly data-fill-with container in your document, they’ll generate anyway (if they have anything to fill themselves with), appending themselves to the end of the body. These sections are:

Again, these will only auto-generate if there is something for them to do; if your spec doesn’t define any CSS properties, for example, the "property-index" boilerplate won’t generate. If you want to suppress their generation even when they do have something to do, use the Boilerplate metadata, like:

<pre class="metadata">
Boilerplate: idl-index no, property-index no

12.5.2. Custom/Overriding Boilerplate

Sometimes a file-based boilerplate (see below) that is appropriate for most of the specs in your group isn’t quite right for your specific spec. Any boilerplate, file-based or Bikeshed-generated, can be overridden by custom text of your choosing. Just add an element to your source document with the content you’d like to show up in place of the offending boilerplate, and add a boilerplate="foo" attribute to the container, specifying which boilerplate section is being replaced.

Bikeshed will automatically remove that element from your document, and instead inject its contents in place of the boilerplate that it would normally provide.

The boilerplate attribute isn’t limited to just the predefined boilerplate section names; you can supply whatever section name you want, and if there’s a data-fill-with attribute somewhere in the document with the same name, the element’s contents will be moved there.

12.6. Table of Contents

The headings in the spec are automatically numbered, and a table of contents automatically generated.

Any heading h2 to h6 (that is, skipping only the document-titling h1) is automatically numbered by having a <span class='secno'>...</span> prepended to its contents. You can avoid this behavior for a heading and all of its subsequent subheadings by adding class="no-num" to the heading.

Similarly, a ToC is generated to match. Headings and their subheadings can be omitted from the ToC by adding class="no-toc" to them.

The processor assumes that your headings are numbered correctly. It does not yet pay attention to the HTML outline algorithm, so using a bunch of h1s nested in sections will have very wrong effects.

Headings also automatically gain a self-link pointing to themselves, to enable people to easily link to sections without having to return to the ToC.

12.7. File-based Includes

Several of the data-fill-with values (those that are static, rather than generated from in-document data) actually come from sets of .include files in the include/ directory.

The base files are simply named "foo.include", where "foo" is the name of the data-fill-with value. They can be specialized, however, to particular working groups, and to particular document statuses.

Putting the boilerplate in a folder named after the group, like csswg/header.include, specializes it for that group (specified in the spec’s metadata). Adding a "-STATUS" to the filename specializes it for the status (same). These can be used together, like "csswg/status-CR.include".

The processor will first look for the "group/foo-STATUS.include" file, failing over to "group/foo.include", then "foo-STATUS.include", and finally "foo.include".

12.8. Creating New Boilerplate Files For Your Organization

Bikeshed’s default boilerplate generates a functional and reasonably attractive spec, but if your group has specific style requirements, you can produce your own boilerplate files. This section is a basic guide to developing these files.

The most important part of the boilerplate is the header.include and footer.include file. These define the parts of the spec HTML that precede and follow your actual spec content, so the source file can contain only the actual spec text, and all specs in the same organization can look similar.

Here is a basic example header.include file:

<!DOCTYPE html>
<html lang="en">
  <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
<body class="h-entry">
<div class="head">
  <p data-fill-with="logo"></p>
  <h1 id="title" class="p-name no-ref">[TITLE]</h1>
  <h2 id="subtitle" class="no-num no-toc no-ref">[LONGSTATUS],
    <span class="dt-updated"><span class="value-title" title="[CDATE]">[DATE]</span>
  <div data-fill-with="spec-metadata"></div>
  <div data-fill-with="warning"></div>
  <p class='copyright' data-fill-with='copyright'></p>
  <hr title="Separator for header">

<div class="p-summary" data-fill-with="abstract"></div>
<div data-fill-with="at-risk"></div>

<nav data-fill-with="table-of-contents" id="toc"></nav>

This uses several of Bikeshed’s boilerplating features:

13. Railroad Diagrams

A railroad diagram is a particular way of visually representing a structure roughly equivalent to regular expressions, or simple grammars. They tend to be more readable and easier to grok than their equivalents written in terse regexps, and smaller than their equivalents written in explicit parsers.

Here’s an example of a railroad diagram, this one describing the syntax of valid IDENT tokens in CSS:

-- - a-z A-Z _ or non-ASCII escape a-z A-Z 0-9 _ - or non-ASCII escape

Bikeshed supports the automatic generation of railroad diagrams from a simplified DSL. To use, simply embed a diagram description in a <pre class='railroad'> element - it’ll get replaced by an appropriate svg element.

13.1. The Diagram Language

Diagrams are described by a custom DSL that somewhat resembles Python.

A railroad diagram consists of a number of nested elements, each of which may contain multiple children. Each element is specified as a command followed by a colon, possibly followed by additional data (the prelude), and the element’s children indented on following lines, like:

T: /*
  N: anything but * followed by /
T: */

This draws the following diagram:

/* anything but * followed by / */

The top-level elements are assumed to be a sequence of elements in the diagram. Inside of a diagram, any of the elements may be used. Elements are split into two groups: containers and text.

The containers hold other elements, and modify their semantics:

14. Source-File Processing: bikeshed source

Sometimes it’s the source file you want to preprocess, if there is some feature you want literally in your source that is hard or annoying to type in yourself. Bikeshed has some options for doing this as well.

All of these commands are accessed from the source sub-command, like bikeshed source. You can run individual commands by specifying their relevant flag (see bikeshed source -h for a list), or run all of them by not passing any flags.

14.1. Big Text

When editing a large spec, it’s easy to get lost in its length, and have to spend some time scrolling back and forth to find particular sections.

The Sublime Text editor has a special feature, the minimap, which shows an extremely-zoomed out version of your document while you scroll, so you can recognize where you are in the file by the shape of your code. This can be made even easier by putting extra-large "ASCII art" text in your source to label major sections, so they show up visibly in the minimap as section markers.

Bikeshed can auto-generate this "ASCII art" text for you with its --big-text command. Just add an HTML comment to your document on its own line that looks like:

<!-- Big Text: Your Text -->

If you run bikeshed source --big-text, Bikeshed will replace it with a comment that looks like:

<!-- Big Text: Your Text

█   ▐▌  ███▌  █▌  █▌ ████▌      █████▌ █████▌ █     █ █████▌
▐▌  █  █▌  █▌ █▌  █▌ █▌  █▌       █▌   █▌      █   █    █▌
 █ ▐▌  █▌  █▌ █▌  █▌ █▌  █▌       █▌   █▌       █ █     █▌
 ▐▌█   █▌  █▌ █▌  █▌ ████▌        █▌   ████      █      █▌
  █▌   █▌  █▌ █▌  █▌ █▌▐█         █▌   █▌       █ █     █▌
  █▌   █▌  █▌ █▌  █▌ █▌ ▐█        █▌   █▌      █   █    █▌
  █▌    ███▌   ███▌  █▌  █▌       █▌   █████▌ █     █   █▌

Which is clearly visible from Sublime’s minimap!

You can also edit the first line to change the text, re-run bikeshed source --big-text, and it’ll swap out the text art with your new text.

Appendix A: Bikeshed’s "InfoTree" Format

Bikeshed’s custom text formats attempt to be fairly regular; most of them involve specifying key/value pairs, and are line-based. For example, Bikeshed’s metadata format is one key/value pair per line, with a colon between the key and the value.

The InfoTree format, used by several things in Bikeshed, is similar. It’s used when you need to specify data consisting of multiple key/value pairs, where it’s common that multiple entries share some of that data. The InfoTree format makes this easy to read, write, and maintain.

Specifying Information on a Single Line

The simplest way to provide a piece of information is by putting all the key/value pairs on a single line. In the InfoTree format, this is done by putting a colon between the key and value, and separating the pairs with semicolons. For example, here is an example of two "anchor" entries:

urlPrefix:; type: dfn; text: ascii whitespace
urlPrefix:; type: dfn; text: utf-8

This specifies two entries, each with three keys: urlPrefix, type, and text.

Nesting Information to Share Pieces

When multiple pieces of information share some key/value pairs, you can use nesting to indicate this, so you don’t have to repeat yourself. Here’s the same two entries as before, but using nesting to share their common information:

urlPrefix:; type: dfn
  text: ascii whitespace
  text: utf-8

Just like the previous, this defines two entries, each with three key/value pairs. Now it’s clearer, though, that the two entries share their urlPrefix and type data, and you only have to maintain the common data in one place.


A line that starts with a # character (with any amount of preceding whitespace) is a comment; it will be completely ignored for the purpose of parsing.

Additional Details

The order that keys are specified in is irrelevant. Feel free to rearrange them for readability or more effective nesting.

You can specify the same key multiple times; the values will be collected into an array for later processing. (Each user of InfoTree will define whether multiple values for a key is valid or not, and what it means.) The order that the values appear in is preserved, as it might be important. (For example, in the anchor format, multiple urlPrefix values are concatenated together, to help specify urls in multipage specs.)

Additional semicolons are silently ignored; in other words, empty entries get dropped, so you can put a final semicolon at the end of the line or not, as you prefer.


Conformance requirements are expressed with a combination of descriptive assertions and RFC 2119 terminology. The key words “MUST”, “MUST NOT”, “REQUIRED”, “SHALL”, “SHALL NOT”, “SHOULD”, “SHOULD NOT”, “RECOMMENDED”, “MAY”, and “OPTIONAL” in the normative parts of this document are to be interpreted as described in RFC 2119. However, for readability, these words do not appear in all uppercase letters in this specification.

All of the text of this specification is normative except sections explicitly marked as non-normative, examples, and notes. [RFC2119]

Examples in this specification are introduced with the words “for example” or are set apart from the normative text with class="example", like this:

This is an example of an informative example.

Informative notes begin with the word “Note” and are set apart from the normative text with class="note", like this:

Note, this is an informative note.


Terms defined by this specification

Terms defined by reference


Normative References

Tab Atkins Jr.; et al. CSS Flexible Box Layout Module Level 1. URL:
Rossen Atanassov; Elika Etemad. CSS Logical Properties and Values Level 1. URL:
Tab Atkins Jr.; Elika Etemad. CSS Values and Units Module Level 4. URL:
Anne van Kesteren; et al. HTML Standard. Living Standard. URL:
Anne van Kesteren; Domenic Denicola. Infra Standard. Living Standard. URL:
S. Bradner. Key words for use in RFCs to Indicate Requirement Levels. March 1997. Best Current Practice. URL:
Amelia Bellamy-Royds; et al. Scalable Vector Graphics (SVG) 2. URL:
Brian Birtles; et al. Web Animations. URL:
Edgar Chen; Timothy Gu. Web IDL Standard. Living Standard. URL:

Property Index

Name Value Initial Applies to Inh. %ages Anim­ation type Canonical order Com­puted value
flex-basis content | <<'width'>> auto flex items no relative to the flex container’s inner main size length per grammar as specified, with lengths made absolute

IDL Index

interface Foo {
  undefined bar(long arg1, DOMString? arg2);

Issues Index

Define the issues-list format. The -t output is already more than enough to actually work with, but it would still be good to describe it more fully.