Autoconsent

This is a library of rules for navigating through common consent popups on the web. These rules can be run in a Chrome extension, or in a Playwright-orchestrated headless browser. Using these rules, opt-in and opt-out options can be selected automatically, without requiring user-input.

Using the library

Autoconsent is meant to be used in browser apps and extensions. DuckDuckGo browser apps use this library to automatically handle cookie consent popups.

To integrate Autoconsent, you'll need to instantiate the main AutoConsent class in a content script (running in isolated page context), and implement some configuration hooks in a background script. See this document for more details on internal APIs and data flows.

import AutoConsent from '@duckduckgo/autoconsent'; // or '@duckduckgo/autoconsent/extra' for the version with filterlists
import * as rules from '@duckduckgo/autoconsent/rules/rules.json';

const autoconsent = new AutoConsent(
    chrome.runtime.sendMessage, // provide a callback to send messages to the background script
    null, // optionally provide a config object here if you don't want to implement a background script
    rules,
);

// connect the message receiver callback to handle messages from the background script
chrome.runtime.onMessage.addListener((message) => {
  return Promise.resolve(consent.receiveMessageCallback(message));
});

Browser extension

Autoconsent comes with a reference extension implementation. It is not published in stores since the feature is available in all DuckDuckGo apps, but you can build it yourself and use for testing.

To build the extension:

# Download dependencies
npm install
# Build the extension
npm run prepublish

The extension-specific code can be found in the addon directory. There are two versions of the addon (found under dist/addon after building), one for mv3 version for Chromium-based browsers, and a firefox version for Firefox. You can load these in Chrome in developer mode, and in Firefox as a temporary addon.

Watch mode

For development, you can run in watch mode

npm run watch

This will rebuild the extension on every source file change. You still need to refresh the extension in the browser to see the changes.

Rules

The library's functionality is implemented as a set of rules that define how to manage consent on a subset of sites. These generally correspond to specific Consent Management Providers (CMPs) that are installed on multiple sites. Each CMP ruleset defines:

If the site is using that CMP
If a popup is displayed
Steps to specify an 'opt-in' or 'opt-out' consent for the CMP.
Optionally, a test if the consent was correctly applied.

There are currently three ways of implementing a CMP:

As a JSON ruleset, intepreted by the AutoConsent class.
As a class implementing the AutoCMP interface. This enables more complex logic than the linear AutoConsent rulesets allow.
As a Consent-O-Matic rule. The ConsentOMaticCMP class implements compability with rules written for the Consent-O-Matic extension.

Intermediate rules

Sometimes the opt-out process requires actions that span across multiple pages or iframes. In this case it is necessary to define stages (each corresponding to a separate page context) as separate rulesets. Each one, except the very last stage, must be marked as intermediate using the intermediate: true flag. If the intermediate flag is not set correctly, autoconsent may report a successful opt-out even if it is not yet finished.

Cosmetic rules

Some rules do not interact with the page, and only hide the cookie pop-ups with CSS. These rules are marked with the cosmetic: true flag. They are useful for pop-ups that do not provide a Reject button. Cosmetic rules can be disabled with the enableCosmeticRules config option.

Filterlist

Autoconsent supports cosmetic filters in common ABP/uBO format. For performance reasons, it needs to be bundled at build time for performance reasons. At the moment we include cosmetic filters from Easylist Cookie. Note that by default filterlist rules are not included, as this significantly increases the resulting bundle size. To use filterlist rules, you need to explicitly import the "extra" version of the library (@duckduckgo/autoconsent/extra), and set the enableFilterlist config option to true.

// import the library version with bundled filterlist rules
import AutoConsent from '@duckduckgo/autoconsent/extra'

// ...

new AutoConsent({
  enableFilterlist: true,
  // other options
})

Context filters

By default, rules will be executed in all top-level documents. Some rules are designed for specific contexts (e.g. only nested iframes, or only specific URLs). This can be configured in runContext field (see the syntax reference below).

Rule Syntax Reference

An autoconsent CMP rule can be written as either:

a JSON file adhering to the AutoConsentCMPRule type.
a class implementing the AutoCMP interface, or
- common JSON rules are available as reusable functions in dom-actions.ts. You can also use existing class-based rules as reference.

In most cases the JSON syntax should be sufficient, unless some complex non-linear logic is required, in which case a class is required.

Both JSON and class implementations have the following components:

name - to identify this CMP.
detectCMP - which determines if this CMP is included on the page.
detectPopup - which determines if a popup is being shown by the CMP.
optOut - a list of actions to do an 'opt-out' from the popup screen. i.e. denying all consents possible.
optIn - a list of actions for an 'opt-in' from the popup screen.
(optional) prehideSelectors - a list of CSS selectors to "pre-hide" early before detecting a CMP. This helps against flickering. Pre-hiding is done using CSS opacity and z-index, so be it should be used with care to prevent conflicts with the opt-out process.
(optional) intermediate - a boolean flag indicating that the ruleset is part of a multi-stage process, see the Intermediate rules section. This is false by default.
(optional) vendorUrl - link to the CMP vendor site, for reference.
(optional) cosmetic - a boolean flag indicating that the rule is purely cosmetic and does not affect the consent state. This is false by default.
(optional) runContext - an object describing when this rule should be tried:
- main - boolean, set to true if the rule should be executed in top-level documents (default: true)
- frame - boolean, set to true if the rule should be executed in nested frames (default: false)
- urlPattern - string, specifies a regular expression that should match the page URL (default: empty)
(optional) test - a list of actions to verify a successful opt-out. This is currently only used in Playwright tests.

detectCMP, detectPopup, optOut, optIn, and test are defined as a set of checks or actions on the page. In the JSON syntax this is a list of AutoConsentRuleStep objects. For detect checks, we return true for the check if all steps return true. For opt in and out, we execute actions in order, exiting if one fails. The following checks/actions are supported:

Element selectors

Many rules use ElementSelector to locate elements in a page. ElementSelector can be a string, or array of strings, which are used to locate elements as follows:

By default, strings are treated as CSS Selectors via the querySelector API. e.g. #reject-cookies to find an element whose id is 'reject-cookies'.
Strings prefixed with xpath/ are Xpath selectors which can locate elements in the page via document.evaluate. e.g. xpath///*[@id="reject-cookies"] can find an element whose id is 'reject-cookies'.
If an array of strings is given, the selectors are applied in array order, with the search scope constrained each time but the first match of the previous selector. e.g. ['#reject-cookies', 'button'] first looks for an element with id="reject-cookies", then looks for a match for button that is a descendant of that element. If one of the selectors returns an element that has a shadowRoot property, the next selector will run within that element's shadow DOM. This is the main difference from nested CSS selectors, which do not cross shadow DOM boundaries.

For example, consider the following DOM fragment:
```
<open-shadow-root-element>
<button>X</button>
</open-shadow-root-element>
```

Then ['open-shadow-root-element', 'button'] will find the button, but a usual CSS selector 'open-shadow-root-element button' will not.

Element exists

{
  "exists": ElementSelector
}

Returns true if the given selector matches one or more elements.

Element visible

{
  "visible": ElementSelector,
  "check": "any" | "all" | "none"
}

Returns true if elements matched by ElementSelector are currently visible on the page. If check is all, every element must be visible. If check is none, no element should be visible. Visibility check is a CSS-based heuristic.

Wait for element

{
  "waitFor": ElementSelector,
  "timeout": 1000
}

Waits until selector exists in the page. After timeout ms the step fails.

Wait for visibility

{
  "waitForVisible": ElementSelector,
  "timeout": 1000,
  "check": "any" | "all" | "none"
}

Waits until element is visible in the page. After timeout ms the step fails.

Click an element

{
  "click": ElementSelector,
  "all": true | false,
}

Click on an element returned by selector. If all is true, all matching elements are clicked. If all is false, only the first returned value is clicked.

Wait for then click

{
  "waitForThenClick": ElementSelector,
  "timeout": 1000,
  "all": true | false
}

Combines waitFor and click.

Unconditional wait

{
  "wait": 1000,
}

Wait for the specified number of milliseconds.

Hide

{
  "hide": "CSS selector",
  "method": "display" | "opacity"
}

Hide the elements matched by the selectors. method defines how elements are hidden: "display" sets display: none, "opacity" sets opacity: 0. Method is "display" by default. Note that only a single string CSS selector is supported here, not an array.

Eval

{
  "eval": "SNIPPET_ID"
}

Evaluates a code snippet in the context of the page. The rule is considered successful if it evaluates to a truthy value. Snippets have to be explicitly defined in snippets.ts. Eval rules are not 100% reliable because they can be affected by the page scripts, or blocked by a CSP policy on the page. Therefore, they should only be used as a last resort when none of the other rules are sufficient.

Conditionals

{
  "if": { "exists": ElementSelector },
  "then": [
    { "click": ".button1" },
    { "click": ".button3" }
  ],
  "else": [
    { "click": ".button2" }
  ]
}

Allows to do conditional branching in JSON rules. The if section can contain either a "visible" or "exists" rule. Depending on the result of that rule, then or else sequences will be executed. else section is optional. The "if" rule is considered successful as long as all rules inside the chosen branch are successful. The other branch, as well as the result of the condition itself, do not affect the result of the whole rule.

Any

{
  "any": [
    { "exists": ".button1" },
    { "exists": ".button2" }
  ]
}

Evaluates a list of steps in order. If any return true (success), then the step returns true. If all steps return false, the any step returns false.

Optional actions

All rules can include the "optional": true to ignore failure.

License

MPLv2.

Manual Testing

To test the extension / addon with Firefox, open the about:debugging, navigate to "This Firefox" on the menu and under "Temporary Extensions" click on "Load Temporary Addon". Select the manifest.json file from the dist/firefox directory. You will need to build the extension before as described above. The extension should then be active and you can test it manually by simply visiting websites.

duckduckgo / autoconsent

readme