HTML pages
An HTML artifact is a single file. Everything the page needs travels inside it: markup, styles and
scripts. Publish with kind: "html", which is the default.
How the page is served
Section titled “How the page is served”The viewer is rendered by Artiflow and puts your bytes into an iframe:
sandbox="allow-scripts"withoutallow-same-origin, so the document runs in an opaque origin. It cannot read the dashboard’s cookies, storage or DOM, and it cannot claim the application origin.- The bytes travel in
srcdoc, escaped by the server, so a page can never execute on the application origin itself. - The response carries this policy:
default-src 'none'; script-src 'unsafe-inline'; style-src 'unsafe-inline'; frame-src 'self'; base-uri 'none'; form-action 'none'; connect-src 'self'The last directive belongs to the viewer page, not to your document: the page polls the artifact’s own
metadata so it can notice a new revision and swap the document in place. Your document inherits the same
list, but it cannot read an authenticated answer: in an opaque origin a same-origin call matches nothing
under CSP3 'self', and an authenticated response is unreadable from there in any case. It reaches no
parent page, no storage and no other origin.
Allowed and blocked
Section titled “Allowed and blocked”| Inside the artifact | Result |
|---|---|
Inline <style>, inline <script> |
Runs |
Inline SVG, CSS shapes, <canvas> 2D drawing |
Runs |
fetch, XMLHttpRequest, WebSocket, sendBeacon |
Blocked: no remote host; a same-origin call carries no cookies |
| External scripts, stylesheets, fonts or images | Blocked: each is a network request |
localStorage, sessionStorage, IndexedDB, cookies |
Unavailable: the origin is opaque |
| Form submission | Blocked |
| Replacing the top-level page, opening popups | Blocked |
| A link inside the page | Can navigate the frame itself away from the page |
So there are no CDN links, no web fonts, no remote images and no analytics. Use system font stacks, inline SVG and CSS for visuals, and computed values for anything that would normally be fetched. A page that depends on a network request renders incomplete rather than failing loudly.
Inline scripts may do real work: compute numbers, draw charts, toggle a small amount of state. That is enough for dashboards, reports, comparisons and one-page explainers, which is the kind of artifact this viewer is built for.
Viewer controls
Section titled “Viewer controls”The bar above the page shows the title and the version, and offers Focus, which gives the page the
whole window; the F key toggles it and Esc leaves it. The version badge increments with every new
revision, so a reader can tell that the page changed.
The public link renders the same page in the same sandbox without the bar and without a back link.
Updating a page
Section titled “Updating a page”There is no in-browser editor for HTML. To change a page, publish a new revision through the agent, naming the artifact and the revision you are replacing:
{ "artifactId": "3f1c0e6a-8b2d-4d3f-9a71-0c5b8e2d4a19", "expectedRevisionId": "9d2b7c44-1e3a-4b8f-8d0e-6a1f2c3b4d5e", "title": "Pricing comparison", "text": "<!doctype html>…"}artifactId must be the artifact’s canonical lowercase UUID; anything else is refused before any read
or write. expectedRevisionId is optional there, but sending it is what turns a blind overwrite into a
conflict you can see. Edit and revise covers the same rule from the editor
side.