Open Graph Preview

Open Graph Preview hands a page address to the OmniKitly processing service, which requests that page and returns the source down to the closing head tag. Ten properties are read out of it and printed in one table — og:title, og:description, og:image, og:url, og:type, og:site_name, og:locale, og:image:alt, og:image:width and og:image:height — each showing the value found, or the words not set where the tag is absent. Above the table sits a schematic of the large-image unfurl: the host in capitals, the title cut at 88 characters, the description cut at 200, and a grey panel naming the og:image address rather than showing it.

Every value is then judged, not merely listed. Errors are reserved for the four faults that break a share outright: no og:image at all, an og:image written as a relative path (the absolute form is worked out and offered), an og:url that is not absolute, and a page with neither og:title nor a title element. Warnings cover the rest — a missing og:title where a title element can stand in, a title past 88 characters, an og:description that is missing or past 200, an og:image served over plain http, declared dimensions under 600 by 315, a missing og:url, an og:type that is neither a standard value nor namespaced with a colon, an og:locale not shaped like en_GB, og:title, og:description, og:url or og:type declared twice, and og values spelled name= where the vocabulary requires property=. Notes carry the softer observations: a second og:image tag, no alt text, no site name, no og:type, no declared dimensions, an aspect ratio wandering from 1.91:1. An og:image:secure_url counts as an image alongside og:image.

The picture is named, never requested. Nothing downloads the file at og:image, so an address that answers 404, a file too small for the layout, or one whose subject falls outside the crop all pass without comment — the size check reads only the width and height you declared. The report is Open Graph and nothing else, too: the canonical link, the robots value, the HTTP status, the address the redirects ended on and the four twitter: values the fetch collected sit in the meta.json download rather than on screen. Nothing on the page is executed and only the head is parsed, so a tag a framework writes after load, or one emitted below the closing head tag, reads as absent — much as it would to a scraper. Markup inside a comment is skipped; markup inside a script string or an unrendered template is counted as though it were live.

This tab never issues the request itself: same-origin rules stop one site's page from reading another site's markup, so the work is handed off. The address travels to the service over an encrypted connection and is kept in the job record for as long as that job exists. It goes out as OmniKitly/1.0 carrying no cookies and no sign-in, follows at most five redirects with every hop resolved and re-checked, reads at most 512 KB, and abandons any single request that has been silent for fifteen seconds. Nothing but a public http or https address will be attempted, and every address a hostname answers with is inspected first, so a name pointing inward is turned away before a socket opens. The job record and the meta.json go together six hours after the run, unless an administrator has changed that window.

Secure server processing. This operation requires temporary server processing.