CVE-2026-114128

In Astro from 7.1.0 until 7.3.3, the server-side View Transition CSS generator does not keep string animation properties within their generated declarations. An application passing attacker-controlled data into properties such as duration, delay, easing, direction, fillMode, or animation names can allow CSS delimiters to terminate the declaration and add attacker-selected rules to the server-rendered page. The injected CSS can conceal or reposition content, add generated text, and otherwise misrepresent the interface. This is fixed in 7.3.3.

Astro is a web framework for content-driven websites. Prior to version 11.1.3, the @astrojs/node adapter builds a request URL from the Host header, and a malformed port can make that URL invalid. The recovery path reuses the same malformed host and throws an uncaught TypeError: Invalid URL before routing begins. In the default standalone configuration, the request returns an HTTP 500 response and the server continues running, but when staticHeaders is enabled the synchronous handler does not catch the exception and the Node process terminates. Proxies and CDNs that reject malformed Host headers prevent this path from reaching the origin. The issue affects availability only and does not expose data or permit code execution. This issue is fixed in version 11.1.3.

Astro from 5.2.0 to 8.2.4 has a vulnerability in the @astrojs/netlify adapter that generates regexes for Netlify Image CDN remote-image allowlists without anchoring them to the URL start. This allows an attacker to control the actual host while an allowed origin appears only in the path or query. An unauthenticated request to /.netlify/images can cause the Image CDN to request attacker-selected URLs, potentially probing internal services. Fixed in version 8.2.4.

Astro is a web framework for content-driven websites. Prior to 7.2.4, Astro stripped a configured non-root base path from request pathnames using a string-prefix check without verifying a path-segment boundary. With base "/app", a request to "/appX/admin" resolved internally to the protected "/admin" route while middleware observed "/appX/admin" in context.url.pathname. In applications that authorize base-prefixed routes by inspecting context.url.pathname, an unauthenticated remote attacker could bypass pathname-based middleware authorization and reach protected routes. This issue is fixed in version 7.2.4.

Astro from 10.0.3 until 11.0.3 has a vulnerability in the Vercel adapter where the public /_isr function accepts x_astro_path based only on the x-vercel-isr header, allowing unauthenticated GET requests to render routes protected only by Vercel edge path rules or split edge middleware.

Astro prior to 8.1.2, the Netlify adapter converts each image.remotePatterns entry into a regular expression written to .netlify/v1/config.json under images.remote_images for Netlify's Image CDN allowlist. In packages/integrations/netlify/src/index.ts, remotePatternToRegex() escapes dots in hostname values but interpolates literal pathname values without escaping regular expression metacharacters such as ., +, ?, (, and [, so a restrictive pathname such as /img/v1.0/file also matches metacharacter-adjacent paths, including paths that cross a segment. Netlify enforces the generated regular expression directly and Astro's matchPattern() helper does not revalidate the request, allowing optimization of images on an already-allowed host that the declared pathname was intended to exclude. This issue is fixed in version 8.1.2.

Astro is a web framework for content-driven websites. From 7.0.0 until 7.0.6, the composable astro/hono pipeline installs security.checkOrigin only through the middleware() primitive, while actions() and pages() can dispatch to user code independently. Mounting actions() before middleware(), as in the examples/advanced-routing example and Cloudflare Hono documentation, allows cross-origin form-encoded action requests to execute before the origin check, and using pages() without middleware() drops the check for on-demand endpoints and pages. The flaw enables blind write-only cross-site request forgery using the victim's cookies against ActionHandler.handle and PagesHandler.handleWithErrorFallback when manifest.checkOrigin is enabled; the attacker can trigger a state-mutating action or endpoint handler but cannot read the cross-origin response. The default non-composable astro() pipeline is not affected. This issue is fixed in version 7.0.6.

Astro is a web framework for content-driven websites. From 2.9.0 until 7.1.0, Astro's server-side View Transition CSS generator interpolates animation properties into an inline style element without escaping them for CSS and HTML contexts. An attacker-controlled View Transition animation value such as duration can terminate the generated style element and inject arbitrary HTML or JavaScript. The affected code is packages/astro/src/runtime/server/transition.ts; renderTransition passes sheet.toString() into markHTMLString(), while addAnimationProperty serializes duration through toTimeValue() and also handles easing, direction, delay, fillMode, and name. Exploitation requires an on-demand or server-rendered route to pass attacker-controlled data into a View Transition animation definition and can execute arbitrary JavaScript in the affected application's origin, allowing access to sensitive page data and authenticated actions available to the victim. This issue is fixed in version 7.1.0.

In Astro versions 8.1.0 through 11.0.1, when trailingSlash: 'always' is configured, the @astrojs/node standalone server's static file handler appends a trailing slash to request paths and issues a 301 redirect. Paths beginning with /\ (slash-backslash) were not recognized as internal paths, so the handler would echo the raw path back in the Location header. Because browsers treat \ as / per the WHATWG URL specification, the resulting redirect could resolve to an external host.

Astro versions 1.0.0 through 4.0.18 have a vulnerability where source.title and enclosure.type fields in the RSS generator are not XML-escaped. An attacker controlling these values can inject arbitrary XML into the RSS feed, leading to fake metadata, malicious redirects, or misparsing by feed readers. In SSR mode, the poisoned feed is served to all subscribers. Fixed in version 4.0.19.