Content Security Policy
Set a Content Security Policy from route or shell headers, then verify it on dynamic and prerendered pages.
On this page
This recipe adds a Content Security Policy to every page under a shell, then allows the inline code some pracht features generate.
1. Set a starter policy
Return the policy from the headers() export of the shell that wraps your pages:
export function headers() {
return {
"content-security-policy": [
"default-src 'self'",
"base-uri 'self'",
"object-src 'none'",
"frame-ancestors 'self'",
"form-action 'self'",
"script-src 'self'",
"style-src 'self'",
"img-src 'self' data:",
"font-src 'self'",
"connect-src 'self'",
].join("; "),
};
}Every page under the shell now carries the header. It allows pracht's module script and same-origin assets and blocks cross-origin scripts and plugin embeds.
2. Add the origins you use
List only the external origins your app loads from:
export function headers() {
return {
"content-security-policy": [
"default-src 'self'",
"script-src 'self'",
"style-src 'self'",
"img-src 'self' data: https://images.example.com",
"font-src 'self' https://fonts.example.com",
"connect-src 'self' https://api.example.com",
"object-src 'none'",
"base-uri 'self'",
"frame-ancestors 'self'",
"form-action 'self'",
].join("; "),
};
}Avoid 'unsafe-eval'. Avoid 'unsafe-inline' unless an audited integration requires it and the exception is documented.
3. Allow framework inline code
Some features emit inline <style> or <script> elements. On SSR routes, allow them with a nonce generated per request. Set it on the context in a middleware:
import type { MiddlewareFn } from "@pracht/core";
export const middleware: MiddlewareFn = ({ context }, next) => {
const bytes = crypto.getRandomValues(new Uint8Array(16));
context.cspNonce = btoa(String.fromCharCode(...bytes));
return next();
};Register it in defineApp({ middleware }) and add it to the group that uses the shell. See Middleware.
Framework-generated styles
defineFont(), pracht({ inlineCss: true }), and defineApp({ viewTransitions: true }) emit inline <style> elements. Return the nonce as styleNonce from the shell's head() and allow it in style-src:
import { inter } from "../fonts.ts";
export function head({ context }) {
return { fonts: [inter], styleNonce: context.cspNonce };
}
export function headers({ context }) {
return {
"content-security-policy": `default-src 'self'; style-src 'self' 'nonce-${context.cspNonce}'; font-src 'self'`,
};
}fontNonce overrides the nonce for font styles only; prefer styleNonce.
SSG and ISG output cannot reuse a request nonce. For static documents, keep CSS linked rather than inlined, or allow the exact inline content by hash. The viewTransitions style never changes, so allow it with 'sha256-SREix9zPMZHrSuo8zRSjb672r1gsHIh96MJuaZq6iJo='.
Framework-generated scripts
Buffered pages need no inline scripts. Two features emit them:
- Streaming routes (
streaming: true) emit scripts for deferred data and Suspense boundary swaps. - Routes with
speculationemit a speculation rules script.
Return the nonce as scriptNonce from the shell's head() and allow it in script-src:
export function head({ context }) {
return { scriptNonce: context.cspNonce, styleNonce: context.cspNonce };
}
export function headers({ context }) {
return {
"content-security-policy": `default-src 'self'; script-src 'self' 'nonce-${context.cspNonce}'; style-src 'self' 'nonce-${context.cspNonce}'`,
};
}Server islands need nothing extra: the script that fills them is a same-origin module covered by script-src 'self' and connect-src 'self'.
4. Verify
- Load an SSR page and an SSG or ISG page in a browser.
- Navigate client-side between routes.
- Check the console. A blocked resource logs a
Content-Security-Policyviolation naming the directive; widen only that directive.
Inline script entries in head()
pracht renders pages without app-authored inline scripts. If a route's head() returns inline script entries, such as JSON-LD, test that route with the CSP enabled and allow the exact content with a hash.
Headers on SSG and ISG pages
Headers on SSG and ISG pages are stored with the output and replayed to every visitor, so keep them public and replay-safe. Prerendering fails if they include Set-Cookie, authentication headers such as Authorization, or secret-shaped custom x-* headers; set those from API routes, middleware responses, or SSR routes.