v4: link previews are read once, in one place, and no longer take an error page, a relative image or a mask icon for the page's — each provider fetched its page and scanned it with regular expressions of its own, the generic one some fifteen times, and Bandcamp and SoundCloud read only OG tags. fetchPage (remote/fetch.ts) now fetches every page, follows redirects to the address it was served from, and refuses an error status, so a 404 or a bot challenge is unreachable rather than a preview titled "Just a moment...", and a refresh no longer overwrites good metadata with it. parsePage (remote/html.ts) reads it in one linear pass with a real attribute reader: unquoted values, a > inside quotes, og: tags under name as well as property, and nothing from comments or scripts. Every URL a page names is resolved against where it was served from, so a relative og:image works; entities are decoded in one pass, &eacute; and &rsquo; included, and one that names no character no longer fails the preview (bandcamp-tralbum.ts shares the decoder). The mask icon is no longer the favicon, JSON-LD prefers the page's node over the site's, tracking pixels and lazy placeholders aren't thumbnails, an og:url with a slug keeps the OG tags, and a <title> loses its site suffix. genericPreview is the base of the Bandcamp, SoundCloud and YouTube channel previews; SoundCloud takes www., m. and on.soundcloud.com short links, embedding the page they lead to, and Bandcamp keeps only its own player. The service tidies every preview: text on one line, titles and descriptions capped, only http(s) images and players. RichContent.type is now RichContentType, and the unused videoId is gone.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

api/services/rich-content-service.ts

@@ -84,6 +84,65 @@ export function tryFetchRichContent(
   return result;
 }
 
+/** The longest title a preview keeps, in characters. */
+const MAX_TITLE_LENGTH = 300;
+
+/**
+ * The longest description a preview keeps, in characters: it is stored with
+ * the dump, searched and sent with every feed, and some pages put a whole
+ * article in theirs.
+ */
+const MAX_DESCRIPTION_LENGTH = 1000;
+
+/**
+ * Returns `text` with its whitespace collapsed, cut to `max` characters with
+ * an ellipsis; undefined when nothing is left.
+ */
+function tidyText(
+  text: string | undefined,
+  max: number,
+): string | undefined {
+  const collapsed = text?.replace(/\s+/g, " ").trim();
+  if (!collapsed) return undefined;
+  const chars = [...collapsed];
+  return chars.length > max
+    ? chars.slice(0, max - 1).join("").trimEnd() + "…"
+    : collapsed;
+}
+
+/** Returns `url` if it is an absolute http(s) URL, and undefined otherwise. */
+function httpUrl(url: string | undefined): string | undefined {
+  if (!url) return undefined;
+  try {
+    const { protocol } = new URL(url);
+    return protocol === "http:" || protocol === "https:" ? url : undefined;
+  } catch {
+    return undefined;
+  }
+}
+
+/**
+ * Applies the rules every preview follows, whichever provider made it: text
+ * on one line and of bounded length, and only http(s) URLs for the images and
+ * the player, which the frontend loads.
+ */
+function tidy(content: RichContent): RichContent {
+  const tidied: RichContent = {
+    ...content,
+    title: tidyText(content.title, MAX_TITLE_LENGTH),
+    description: tidyText(content.description, MAX_DESCRIPTION_LENGTH),
+    siteName: tidyText(content.siteName, MAX_TITLE_LENGTH),
+    thumbnailUrl: httpUrl(content.thumbnailUrl),
+    faviconUrl: httpUrl(content.faviconUrl),
+    embedUrl: httpUrl(content.embedUrl),
+  };
+  // Stored as JSON: a field left undefined would only be dropped there.
+  for (const key of Object.keys(tidied) as (keyof RichContent)[]) {
+    if (tidied[key] === undefined) delete tidied[key];
+  }
+  return tidied;
+}
+
 /**
  * Fetches the preview of `url` through the first provider that matches, or
  * makes the failure stub when the provider rejects.
@@ -93,7 +152,7 @@ async function fetchRichContentUncached(
 ): Promise<FetchRichContentResult> {
   try {
     const provider = providers.find((p) => p.matches(url))!;
-    return { ok: true, content: await provider.fetch(url) };
+    return { ok: true, content: tidy(await provider.fetch(url)) };
   } catch (err) {
     console.error(`[rich-content] Failed to fetch metadata for ${url}:`, err);
     const siteName = siteHost(url);