Skip to content
Switch branches/tags
Go to file
12 contributors

Users who have contributed to this file

@WilcoFiers @jeeyyy @marcysutton @straker @SandersAaronD @nschonni @mohanraj-r @stephenmathieson @dylanb @chaals @grgr-dkrk @0xflotus

Axe Developer Guide

Axe runs a series of tests to check for accessibility of content and functionality on a website. A test is made up of a series of Rules which are, themselves, made up of Checks. Axe executes these Rules asynchronously and, when the Rules are finished running, runs a callback function which is passed a Result structure. Since some Rules run on the page level while others do not, tests will also run in one of two ways. If a document is specified, the page level rules will run, otherwise they will not.

Axe 3.0 supports open Shadow DOM: see our virtual DOM APIs and test utilities for developing axe-core moving forward. Note: we do not and cannot support closed Shadow DOM.

  1. Getting Started
    1. Environment Pre-requisites
    2. Building axe.js
    3. Running Tests
    4. API Reference
    5. Supported CSS Selectors
  2. Architecture Overview
    1. Rules
    2. Checks
    3. Common Functions
    4. Virtual Nodes
    5. Core Utilities
  3. Virtual DOM APIs
    1. API Name: axe.utils.getFlattenedTree
    2. API Name: axe.utils.getNodeFromTree
  4. Test Utilities
    1. Test Util Name: axe.testUtils.MockCheckContext
    2. Test Util Name: axe.testUtils.shadowSupport
    3. Test Util Name: axe.testUtils.fixtureSetup
    4. Test Util Name: axe.testUtils.checkSetup
  5. Using Rule Generation CLI

Getting Started

Environment Pre-requisites

  1. You must have NodeJS installed.
  2. Install npm development dependencies. In the root folder of your axe-core repository, run npm install

Building axe.js

To build axe.js, simply run npm run build in the root folder of the axe-core repository. axe.js and axe.min.js are placed into the root folder.

Running Tests

To run all tests from the command line you can run npm test, which will run all unit and integration tests using headless chrome and Selenium Webdriver.

You can also load tests in any supported browser, which is helpful for debugging. Tests require a local server to run, you must first start a local server to serve files. You can use Grunt to start one by running npm start. Once your local server is running you can load the following pages in any browser to run tests:

  1. Core Tests
  2. Commons Tests
  3. Check Tests
  4. Rule Matches
  5. Integration Tests
  6. There are additional tests located in test/integration/full/ for tests that need to be run against their own document.

API Reference

See API exposed on axe

Supported CSS Selectors

Axe supports the following CSS selectors:

  • Type, Class, ID, and Universal selectors. E.g div.main, #main
  • Pseudo selector not. E.g th:not([scope])
  • Descendant and Child combinators. E.g. table td, ul > li
  • Attribute selectors =, ^=, $=, *=. E.g a[href^="#"]

Architecture Overview

Axe tests for accessibility using objects called Rules. Each Rule tests for a high-level aspect of accessibility, such as color contrast, button labels, and alternate text for images. Each rule is made up of a series of Checks. Depending on the rule; all, some, or none of these checks must pass in order for the rule to pass.

Upon execution, a Rule runs each of its Checks against all relevant nodes. Which nodes are relevant is determined by the Rule's selector property and matches function. If a Rule has no Checks that apply to a given node, the Rule will result in an inapplicable result.

After execution, a Check will return true, false, or undefined depending on whether or not the tested condition was satisfied. The result, as well as more information on what caused the Check to pass or fail, will be stored in either the passes, violations, or incomplete arrays.


Rules are defined by JSON files in the lib/rules directory. The JSON object is used to seed the Rule object. A valid Rule JSON consists of the following:

  • id - String A unique name of the Rule.
  • selector - optional String which is a CSS selector that specifies the elements of the page on which the Rule runs. axe-core will look inside of the light DOM and open Shadow DOM trees for elements matching the provided selector. If omitted, the rule will run against every node.
  • excludeHidden - optional Boolean Whether the rule should exclude hidden elements. Defaults to true.
  • enabled - optional Boolean Whether the rule is enabled by default. Defaults to true.
  • pageLevel - optional Boolean Whether the rule is page level. Page level rules will only run if given an entire document as context.
  • matches - optional String The ID of the filtering function that will exclude elements that match the selector property. See the metadata-function-map file for all defined IDs.
  • impact - optional String (one of minor, moderate, serious, or critical). Override the impact defined by checks.
  • tags - optional Array Strings of the accessibility guidelines of which the Rule applies.
  • metadata - Object Consisting of:
    • description - String Text string that describes what the rule does.
    • helpUrl - String optional URL that provides more information about the specifics of the violation. Links to a page on the Deque University site.
    • help - String Help text that describes the test that was performed.
  • any - Array Checks that make up this Rule; one of these checks must return true for a Rule to pass.
  • all - Array Checks that make up this Rule; all these checks must return true for a Rule to pass.
  • none - Array Checks that make up this Rule; none of these checks must return true for a Rule to pass.

The any, all and none arrays must contain either a String which references the id of the Check; or an object of the following format:

  • id - String The unique ID of the Check.
  • options - Mixed Any options the Check requires that are specific to the Rule.

There is a Grunt target which will ensure each Rule has a valid format, which can be run with npx grunt validate.

Matches Function

A Rule's matches function is executed against each node which matches the Rule's selector and receive two parameters:

  • node – node, the DOM Node to test
  • virtualNode– object, the virtual DOM representation of the node. See virtualNode documentation for more.

The matches function must return either true or false. Common functions are provided as commons. See the data-table matches function for an example.


Similar to Rules, Checks are defined by JSON files in the lib/checks directory. The JSON object is used to seed the Check object. A valid Check JSON consists of the following:

  • id - String A unique name of the Check
  • evaluate - String The ID of the function that implements the check's functionality. See the metadata-function-map file for all defined IDs.
  • after - optional String The ID of the function that gets called for checks that operate on a page-level basis, to process the results from the iframes.
  • options - optional Object Any information the Check needs that you might need to customize and/or is locale specific. Options can be overridden at runtime (with the options parameter) or config-time. For example, the valid-lang Check defines what ISO 639-1 language codes it should accept as valid. Options do not need to follow any specific format or type; it is up to the author of a Check to determine the most appropriate format.
  • metadata - Object Consisting of:
    • impact - String (one of minor, moderate, serious, or critical)
    • messages - Object These messages are displayed when the Check passes or fails
      • pass - String doT.js template string displayed when the Check passes
      • fail - String doT.js template string displayed when the Check fails
      • incompleteString|ObjectdoT.js template string displayed when the Check is incomplete OR an object with missingData on why it returned incomplete. Refer to

Check evaluate

A Check's evaluate function is run a special context in order to give access to APIs which provide more information. Checks will run against a single node and do not have access to other frames. A Check must either return true or false.

The following variables are defined for Check#evaluate:

  • node - HTMLElement The element that the Check is run against
  • options - Mixed Any options specific to this Check that may be necessary. If not specified by the user at run-time or configure-time; it will use options as defined by the Check's JSON file.
  • virtualNodeObject The virtualNode object for use with Shadow DOM. See virtualNode documentation.
  • - Function Free-form data that either the Check message requires or is presented as data in the CheckResult object. Subsequent calls to will overwrite previous. See aria-valid-attr for example usage.
  • this.relatedNodes() - Function Array or NodeList of elements that are related to this Check. For example the duplicate-id Check will add all Elements which share the same ID.

Check after

You can use the after function to evaluate nodes that might be in other frames or to filter the number of violations or passes produced. The after function runs once for each Rule in the top-most (or originating) frame. Due to this, you should not perform DOM operations in after functions, but instead operate on data defined by the Check.

For example, the duplicate-id Check include an after function which reduces the number of violations so that only one violation per instance of a duplicate ID is found.

The following variables are defined for Check#after:

  • results - Array Contains CheckResults for every matching node.

The after function must return an Array of CheckResults, due to this, it is a very common pattern to just use Array#filter to filter results:

var uniqueIds = [];
return results.filter(function(r) {
	if (uniqueIds.indexOf( === -1) {
		return true;
	return false;

Pass, Fail and Incomplete Templates

Occasionally, you may want to add additional information about why a Check passed, failed or returned undefined into its message. For example, the aria-valid-attr will add information about any invalid ARIA attributes to its fail message. The message uses a custom message format. In the Check message, you have access to the data object as data.

// aria-valid-attr check
"messages": {
  "pass": "ARIA attributes are used correctly for the defined role",
  "fail": {
    "singular": "ARIA attribute is not allowed: ${data.values}",
    "plural": "ARIA attributes are not allowed: ${data.values}"
  "incomplete": "axe-core couldn't tell because of ${data.missingData}"

See Developing Axe-core Rules for more information on writing rules and checks, including incomplete results.


When a Check is executed, its result is then added to a CheckResult object. Much like the RuleResult object, the CheckResult object only contains information that is required to determine whether a Check, and its parent Rule passed or failed. metadata from the originating Check is combined later and will not be available until axe reaches the reporting stage.

A CheckResult has the following properties:

  • id - String The ID of the Check this CheckResult belongs to.
  • data - Mixed Any data the Check's evaluate function added with Typically used to insert data from analysis into a message or to perform further tests in the post-processing function.
  • relatedNodes - Array Nodes that are related to the current Check as defined by check.evaluate.
  • result - Boolean The return value of check.evaluate.

Common Functions

Common functions are an internal library used by the rules and checks. If you have code repeated across rules and checks, you can use these functions and contribute to them. Documentation is available in source code.

Commons and Shadow DOM

To support open Shadow DOM while maintaining backwards compatibility, commons functions that query DOM nodes must operate on an in-memory representation of the DOM using axe-core’s built-in API methods and utility functions.

Commons functions should do the virtual tree lookup and call a virtual function including the rest of the commons code. The naming of this special function should contain the original commons function name with Virtual added to signify it expects to operate on a virtual DOM tree.

Let’s look at an example:

// lib/commons/text/accessible-text.js
import { getNodeFromTree } from '../../core/utils';
import accessibleTextVirtual from './accessible-text-virtual';

function accessibleText(element, inLabelledbyContext) {
	let virtualNode = getNodeFromTree(axe._tree[0], element); // throws an exception on purpose if axe._tree not correct
	return accessibleTextVirtual(virtualNode, inLabelledbyContext);

export default accessibleText;

// lib/commons/text/accessible-text-virtual.js
function accessibleTextVirtual(element, inLabelledbyContext) {
	// rest of the commons code minus the virtual tree lookup, since it’s passed in

accessibleTextVirtual would only be called directly if you’ve got a virtual node you can use. If you don’t already have one, call the accessibleText lookup function, which passes on a virtual DOM node with both the light DOM and Shadow DOM (if applicable).

Virtual Nodes

To support open Shadow DOM, axe-core has the ability to handle virtual nodes in rule matches and check evaluate functions. The full set of API methods for Shadow DOM can be found in the API documentation, but the general structure for a virtualNode is as follows:

  actualNode: <HTMLElement>,
  children: <Array>,
  parent: <VirtualNode>,
  shadowId: <String>,
  attr: <Function>,
  hasAttr: <Function>,
  props: <Object>,
  • A virtualNode is an object containing an HTML DOM element (actualNode).
  • Children contains an array of child VirtualNodes.
  • Parent is the VirtualNode parent
  • The shadowID indicates whether the node is in an open shadow root and if it is, which one it is inside the boundary.
  • Attr is a function which returns the value of the passed in attribute, similar to node.getAttribute() (e.g. vNode.attr('aria-label'))
  • HasAttr is a function which returns true if the VirtualNode has the attribute, similar to node.hasAttribute() (e.g. vNode.hasAttr('aria-label'))
  • Props is an object of HTML DOM element properties. The general structure is as follows:
      nodeName: <String>,
      nodeType: <Number>,
      id: <String>,
      nodeValue: <String>

Core Utilities

Core Utilities are an internal library that provides axe with functionality used throughout its core processes. Most notably among these are the queue function and the DqElement constructor.

Common Utility Functions

In addition to the ARIA lookupTable, there are also utility functions on the axe.commons.aria and axe.commons.dom namespaces:

  • axe.commons.aria.implicitRole - Get the implicit role for a given node
  • axe.commons.aria.label - Gets the accessible ARIA label text of a given element
  • axe.commons.dom.isVisible - Determine whether an element is visible

Queue Function

The queue function creates an asynchronous "queue", list of functions to be invoked in parallel, but not necessarily returned in order. The queue function returns an object with the following methods:

  • defer(func) Defer a function that may or may not run asynchronously
  • then(callback) The callback to execute once all "deferred" functions have completed. Will only be invoked once.
  • abort() Abort the "queue" and prevent then function from firing

DqElement Class

The DqElement is a "serialized" HTMLElement. It will calculate the CSS selector, grab the source outerHTML and offer an array for storing frame paths. The DqElement class takes the following parameters:

  • Element - HTMLElement The element to serialize
  • Spec - Object Properties to use in place of the element when instantiated on Elements from other frames
var firstH1 = document.getElementByTagName('h1')[0];
var dqH1 = new axe.utils.DqElement(firstH1);

Elements returned by the DqElement class have the following methods and properties:

  • selector - string A unique CSS selector for the element
  • source - string The generated HTML source code of the element
  • element - DOMNode The element which this object is based off or the containing frame, used for sorting.
  • toJSON() - Returns an object containing the selector and source properties

Virtual DOM APIs

Note: You shouldn’t need the Shadow DOM APIs below unless you’re working on the axe-core engine, as rules and checks already have virtualNode objects passed in. However, these APIs will make it easier to work with the virtual DOM.

API Name: axe.utils.getFlattenedTree


Recursively return an array containing the virtual DOM tree for the node specified, excluding comment nodes and shadow DOM nodes <content> and <slot>. This method will return a flattened tree containing both light and shadow DOM, if applicable.


var element = document.body;
axe.utils.getFlattenedTree(element, shadowId);


  • node – HTMLElement. The current HTML node for which you want a flattened DOM tree.
  • shadowId – string(optional). ID of the shadow DOM that is the closest shadow ancestor of the node


An array of objects, where each object is a virtualNode:

		actualNode: body,
		children: [virtualNodes],
		shadowId: undefined

API Name: axe.utils.getNodeFromTree


Recursively return a single node from a virtual DOM tree. This is commonly used in rules and checks where the node is readily available without querying the DOM.


axe.utils.getNodeFromTree(axe._tree[0], node);


  • vNode – object. The flattened DOM tree to fetch a virtual node from
  • node – HTMLElement. The HTML DOM node for which you need a virtual representation


A virtualNode object:

  actualNode: div,
  children: [virtualNodes],
  shadowId: undefined

Test Utilities

All tests must support open Shadow DOM, so we created some test utilities to make this easier.

Test Util Name: MockCheckContext

Create a check context for mocking and resetting data and relatedNodes in tests.


describe('region', function() {
	var fixture = document.getElementById('fixture');

	var checkContext = new axe.testUtils.MockCheckContext();

	afterEach(function() {
		fixture.innerHTML = '';

	it('should return true when all content is inside the region', function() {
		assert.isTrue(checks.region.evaluate.apply(checkContext, checkArgs));
		assert.equal(checkContext._relatedNodes.length, 0);




An object containing the data, relatedNodes, and a way to reset them.

  data: (){},
  relatedNodes: (){},
  reset: (){}

Test Util Name: shadowSupport

Provides an API for determining Shadow DOM v0 and v1 support in tests. For example: PhantomJS doesn't have Shadow DOM support, while some browsers do.


(axe.testUtils.shadowSupport.v1 ? it : xit)(
	'should test Shadow tree content',
	function() {
		// The rest of the shadow DOM test




An object containing booleans for the following Shadow DOM supports: v0, v1, or undefined.

Test Util Name: fixtureSetup

Method for injecting content into a fixture and caching the flattened DOM tree (light and Shadow DOM together).


	'should return true if there is only one ' +
		type +
		' element with the same name',
	function() {
			'<input type="' +
				type +
				'" id="target" name="uniqueyname">' +
				'<input type="' +
				type +
				'" name="differentname">'

		var node = fixture.querySelector('#target');
		assert.isTrue(, node));


  • content – Node|String. Stuff to go into the fixture (html or DOM node)


An HTML Element for the fixture

Test Util Name: checkSetup

Create check arguments.


it('should return true when all content is inside the region', function() {
	var checkArgs = checkSetup(
		'<div id="target"><div role="main"><a href="a.html#mainheader">Click Here</a><div><h1 id="mainheader" tabindex="0">Introduction</h1></div></div></div>'

	assert.isTrue(checks.region.evaluate.apply(checkContext, checkArgs));
	assert.equal(checkContext._relatedNodes.length, 0);


  • content – String|Node. Stuff to go into the fixture (html or node)
  • options – Object. Options argument for the check (optional, default: {})
  • target – String. Target for the check, CSS selector (default: '#target')


An array with the DOM Node, options and virtualNode

[node, options, virtualNode];

Using Rule Generation CLI

Axe provides a CLI for generating the necessary files and configuration assets for authoring a rule.

To invoke the rule generator, run:

npm run rule-gen

The CLI acts a wizard, by asking a series of questions related to generation of the rule, for example:

- What is the name of the RULE? (Eg: aria-valid): sample-rule
- Does the RULE need a MATCHES file to be created?: Yes
- Would you like to create a CHECK for the RULE?: No
- Would you like to create UNIT test files? Yes
- Would you like to create INTEGRATION test files? Yes

Upon answering of which the assets are created in the respective directories.