rehype plugin to sanitize HTML.

• What is this?
• When should I use this?
• Install
• Use
• API
  • defaultSchema
  • unified().use(rehypeSanitize[, schema])
  • Options
• Example
  • Example: headings (DOM clobbering)
  • Example: math
  • Example: syntax highlighting
• Types
• Compatibility
• Security
• Related
• Contribute
• License

This package is a unified (rehype) plugin to make sure HTML is safe. It drops anything that isn’t explicitly allowed by a schema (defaulting to how github.com works).

unified is a project that transforms content with abstract syntax trees (ASTs). rehype adds support for HTML to unified. hast is the HTML AST that rehype uses. This is a rehype plugin that transforms hast.

It’s recommended to sanitize your HTML any time you do not completely trust authors or the plugins being used.

This plugin is built on hast-util-sanitize, which cleans hast syntax trees. rehype focusses on making it easier to transform content by abstracting such internals away.

This package is ESM only. In Node.js (version 16+), install with npm:

npm install rehype-sanitize

In Deno with esm.sh:

import rehypeSanitize from 'https://esm.sh/rehype-sanitize@6'

In browsers with esm.sh:

<script type="module">
  import rehypeSanitize from 'https://esm.sh/rehype-sanitize@6?bundle'
</script>

Say we have the following file index.html:

<div onmouseover="alert('alpha')">
  <a href="jAva script:alert('bravo')">delta</a>
  <img src="x" onerror="alert('charlie')">
  <iframe src="javascript:alert('delta')"></iframe>
  <math>
    <mi xlink:href="data:x,<script>alert('echo')</script>"></mi>
  </math>
</div>
<script>
require('child_process').spawn('echo', ['hack!']);
</script>

…and our module example.js looks as follows:

import rehypeParse from 'rehype-parse'
import rehypeSanitize from 'rehype-sanitize'
import rehypeStringify from 'rehype-stringify'
import {read} from 'to-vfile'
import {unified} from 'unified'

const file = await unified()
  .use(rehypeParse, {fragment: true})
  .use(rehypeSanitize)
  .use(rehypeStringify)
  .process(await read('index.html'))

console.log(String(file))

Now running node example.js yields:

<div>
  <a>delta</a>
  <img src="x">




</div>

This package exports the identifier defaultSchema. The default export is rehypeSanitize.

Default schema (Options).

Follows GitHub style sanitation.

Sanitize HTML.

• options (Options, optional) — configuration

Transform (Transformer).

Schema that defines what nodes and properties are allowed (TypeScript type).

This option is a bit advanced as it requires knowledge of syntax trees, so see the docs for Schema in hast-util-sanitize.

DOM clobbering is an attack in which malicious HTML confuses an application by naming elements, through id or name attributes, such that they overshadow presumed properties in window (the global scope in browsers). DOM clobbering often occurs when user content is used to generate heading IDs. To illustrate, say we have this browser.js file:

console.log(current)

And our module example.js contains:

/**
 * @typedef {import('hast').Root} Root
 */

import fs from 'node:fs/promises'
import rehypeParse from 'rehype-parse'
import rehypeStringify from 'rehype-stringify'
import {unified} from 'unified'

const browser = String(await fs.readFile('browser.js'))
const document = `<a name="old"></a>
<h1 id="current">Current</h1>
${`<p>${'Lorem ipsum dolor sit amet. '.repeat(20)}</p>\n`.repeat(20)}
<p>Link to <a href="#current">current</a>, link to <a href="#old">old</a>.`

const file = await unified()
  .use(rehypeParse, {fragment: true})
  .use(function () {
    /**
     * @param {Root} tree
     */
    return function (tree) {
      tree.children.push({
        type: 'element',
        tagName: 'script',
        properties: {type: 'module'},
        children: [{type: 'text', value: browser}]
      })
    }
  })
  .use(rehypeStringify)
  .process(document)

await fs.writeFile('output.html', String(file))

This code processes HTML, inlines our browser script into it, and writes it out. The input HTML models how markdown often looks on platforms like GitHub, which allow heading IDs to be generated from their text and embedded HTML (including <a name="old"></a>, which can be used to create anchors for renamed headings to prevent links from breaking). The generated HTML looks like:

<a name="old"></a>
<h1 id="current">Current</h1>
<p>Lorem ipsum dolor sit amet.<!--…--></p>
<p>Link to <a href="#current">current</a>, link to <a href="#old">old</a>.</p>
<script type="module">console.log(current)</script>

When you run this code locally and open the generated output.html, you can observe that the links at the bottom work, but also that the <h1> element is printed to the console (the clobbering).

rehype-sanitize solves the clobbering by prefixing every id and name attribute with 'user-content-'. Changing example.js:

@@ -15,6 +15,7 @@ ${`<p>${'Lorem ipsum dolor sit amet. '.repeat(20)}</p>\n`.repeat(20)}

   const file = await unified()
     .use(rehypeParse, {fragment: true})
+    .use(rehypeSanitize)
     .use(function () {
       /**
        * @param {Root} tree

Now yields:

-<a name="old"></a>
-<h1 id="current">Current</h1>
+<a name="user-content-old"></a>
+<h1 id="user-content-current">Current</h1>

This introduces another problem as the links are now broken. It could perhaps be solved by changing all links, but that would make the links rather ugly, and we’d need to track what IDs we have outside of the user content on our pages too. Alternatively, and what arguably looks better, we could rewrite pretty links to their safe but ugly prefixed elements. This is what GitHub does. Replace browser.js with the following:

/// <reference lib="dom" />
/* eslint-env browser */

// Page load (you could wrap this in a DOM ready if the script is loaded early).
hashchange()

// When URL changes.
window.addEventListener('hashchange', hashchange)

// When on the URL already, perhaps after scrolling, and clicking again, which
// doesn’t emit `hashchange`.
document.addEventListener(
  'click',
  function (event) {
    if (
      event.target &&
      event.target instanceof HTMLAnchorElement &&
      event.target.href === location.href &&
      location.hash.length > 1
    ) {
      setImmediate(function () {
        if (!event.defaultPrevented) {
          hashchange()
        }
      })
    }
  },
  false
)

function hashchange() {
  /** @type {string | undefined} */
  let hash

  try {
    hash = decodeURIComponent(location.hash.slice(1)).toLowerCase()
  } catch {
    return
  }

  const name = 'user-content-' + hash
  const target =
    document.getElementById(name) || document.getElementsByName(name)[0]

  if (target) {
    setImmediate(function () {
      target.scrollIntoView()
    })
  }
}

Math can be enabled in rehype by using the plugins rehype-katex or rehype-mathjax. The operate on elements with certain classes and inject complex markup and of inline styles, most of which this plugin will remove. Say our module example.js contains:

import rehypeKatex from 'rehype-katex'
import rehypeParse from 'rehype-parse'
import rehypeSanitize from 'rehype-sanitize'
import rehypeStringify from 'rehype-stringify'
import {unified} from 'unified'

const file = await unified()
  .use(rehypeParse, {fragment: true})
  .use(rehypeKatex)
  .use(rehypeSanitize)
  .use(rehypeStringify)
  .process('<code class="math-inline">L</code>')

console.log(String(file))

Running that yields:

<span><span>LL</span><span><span><span></span><span>L</span></span></span></span>

It is possible to pass a schema which allows MathML and inline styles, but it would be complex and allows all inline styles, which is unsafe. Alternatively, and arguably better, would be to first sanitize the HTML, allowing only the specific classes that rehype-katex and rehype-mathjax use, and then using those plugins:

@@ -1,13 +1,20 @@
 import rehypeKatex from 'rehype-katex'
 import rehypeParse from 'rehype-parse'
-import rehypeSanitize from 'rehype-sanitize'
+import rehypeSanitize, {defaultSchema} from 'rehype-sanitize'
 import rehypeStringify from 'rehype-stringify'
 import {unified} from 'unified'

 const file = await unified()
   .use(rehypeParse, {fragment: true})
+  .use(rehypeSanitize, {
+    ...defaultSchema,
+    attributes: {
+      ...defaultSchema.attributes,
+      // The `language-*` regex is allowed by default.
+      code: [['className', /^language-./, 'math-inline', 'math-display']]
+    }
+  })
   .use(rehypeKatex)
-  .use(rehypeSanitize)
   .use(rehypeStringify)
   .process('<code class="math-inline">L</code>')

Running that yields:

<span class="katex"><span class="katex-mathml"><math xmlns="https://www.w3.org/1998/Math/MathML">…</math></span><span class="katex-html" aria-hidden="true">…</span></span>

Highlighting, for example with rehype-highlight, can be solved similar to how math is solved (see previous example). That is, use rehype-sanitize and allow the classes needed for highlighting, and highlight afterwards:

import rehypeHighlight from 'rehype-highlight'
import rehypeParse from 'rehype-parse'
import rehypeSanitize, {defaultSchema} from 'rehype-sanitize'
import rehypeStringify from 'rehype-stringify'
import {unified} from 'unified'

const file = await unified()
  .use(rehypeParse, {fragment: true})
  .use(rehypeSanitize, {
    ...defaultSchema,
    attributes: {
      ...defaultSchema.attributes,
      code: [
        ...(defaultSchema.attributes.code || []),
        // List of all allowed languages:
        ['className', 'language-js', 'language-css', 'language-md']
      ]
    }
  })
  .use(rehypeHighlight, {subset: false})
  .use(rehypeStringify)
  .process('<pre><code className="language-js">console.log(1)</code></pre>')

console.log(String(file))

Alternatively, it’s possible to make highlighting safe by allowing all the classes used on tokens. Modifying the above code like so:

 const file = await unified()
   .use(rehypeParse, {fragment: true})
+  .use(rehypeHighlight, {subset: false})
   .use(rehypeSanitize, {
     ...defaultSchema,
     attributes: {
       ...defaultSchema.attributes,
-      code: [
-        ...(defaultSchema.attributes.code || []),
-        // List of all allowed languages:
-        ['className', 'hljs', 'language-js', 'language-css', 'language-md']
+      span: [
+        ...(defaultSchema.attributes.span || []),
+        // List of all allowed tokens:
+        ['className', 'hljs-addition', 'hljs-attr', 'hljs-attribute', 'hljs-built_in', 'hljs-bullet', 'hljs-char', 'hljs-code', 'hljs-comment', 'hljs-deletion', 'hljs-doctag', 'hljs-emphasis', 'hljs-formula', 'hljs-keyword', 'hljs-link', 'hljs-literal', 'hljs-meta', 'hljs-name', 'hljs-number', 'hljs-operator', 'hljs-params', 'hljs-property', 'hljs-punctuation', 'hljs-quote', 'hljs-regexp', 'hljs-section', 'hljs-selector-attr', 'hljs-selector-class', 'hljs-selector-id', 'hljs-selector-pseudo', 'hljs-selector-tag', 'hljs-string', 'hljs-strong', 'hljs-subst', 'hljs-symbol', 'hljs-tag', 'hljs-template-tag', 'hljs-template-variable', 'hljs-title', 'hljs-type', 'hljs-variable'
+          ]
       ]
     }
   })
-  .use(rehypeHighlight, {subset: false})
   .use(rehypeStringify)
   .process('<pre><code className="language-js">console.log(1)</code></pre>')

This package is fully typed with TypeScript. It exports the additional type Options.

Projects maintained by the unified collective are compatible with maintained versions of Node.js.

When we cut a new major release, we drop support for unmaintained versions of Node. This means we try to keep the current release line, rehype-sanitize@^6, compatible with Node.js 16.

This plugin works with rehype-parse version 3+, rehype-stringify version 3+, rehype version 5+, and unified version 6+.

The defaults are safe but improper use of rehype-sanitize can open you up to a cross-site scripting (XSS) attack.

Use rehype-sanitize after the last unsafe thing: everything after rehype-sanitize could be unsafe (but is fine if you do trust it).

• hast-util-sanitize — utility to sanitize hast
• rehype-format — format HTML
• rehype-minify — minify HTML

See contributing.md in rehypejs/.github for ways to get started. See support.md for ways to get help.

This project has a code of conduct. By interacting with this repository, organization, or community you agree to abide by its terms.

MIT © Titus Wormer