feat(socialMeta): settings.socialMeta.description override (#7599 follow-up) (#7691)

Issue #7599 follow-up from @stffen: the OG description has no obvious
settings.json knob and the i18n catalog default is English, but most
preview crawlers (WhatsApp, Signal, Slack, Telegram, Facebook) don't
send Accept-Language and so always hit the English fallback regardless
of how many locale files translate `pad.social.description`.

Keep the i18n catalog as the default source — translatable strings
belong in locale files, per the original Qodo review on PR #7635 — but
add an explicit `socialMeta.description` setting that wins when set as
a non-empty string, regardless of negotiated language. This is the
lever that fixes the crawler case for non-English instances without
re-introducing per-language config in settings.json (operators who
want that still use customLocaleStrings).

- Empty/whitespace overrides are treated as unset (would otherwise
  silently blank the preview).
- Override is HTML-escaped via the same path as every other value.
- og:locale stays language-negotiated; only the description is forced.
- Documented next to publicURL in settings.json.template and
  settings.json.docker (env var SOCIAL_META_DESCRIPTION). The
  customLocaleStrings example now spells out pad.social.description so
  operators discover both routes.

5 new unit specs + 4 new integration specs cover override-wins,
null/missing fallback, blank-treated-as-unset, and HTML-escaping.

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

docs/superpowers/specs/2026-04-30-issue-7599-open-graph-metadata-design.md

@@ -144,3 +144,38 @@ The XSS escape test is the security-relevant one: pad IDs are user-controlled
 - A `padSocialMetadata` hook that lets plugins override the values.
 - Per-pad description (e.g. ep_pad_title integration).
 - Generated preview images (would require a rendering service).
+
+## Follow-up (2026-05-07): operator description override
+
+Issue #7599 follow-up comment from @stffen flagged two gaps in the shipped
+behaviour:
+
+1. The default description is in English and there is no obvious place in
+   `settings.json` to change it.
+2. The visitor's language is negotiated from `Accept-Language`, which most
+   link-preview crawlers (WhatsApp, Signal, Slack, Telegram, Facebook) do not
+   send — so non-English instances always serve the English fallback to
+   crawlers regardless of which locale files exist.
+
+Resolution: keep the i18n catalog as the default source (the original Qodo
+review still stands — translatable strings belong in locale files), but add
+an explicit `settings.socialMeta.description` override that wins when set:
+
+- `socialMeta.description: null` (default) → existing behaviour: i18n
+  catalog with `Accept-Language` negotiation, English fallback.
+- `socialMeta.description: "<text>"` → that string is used verbatim for
+  `og:description` / `twitter:description` regardless of the negotiated
+  language. This is the lever that fixes the crawler-no-Accept-Language
+  case.
+- Empty / whitespace-only override is treated as unset (would otherwise
+  blank out previews silently — a footgun).
+- The override is HTML-escaped via the same path as every other
+  interpolated value.
+- `og:locale` is unaffected; it continues to reflect the negotiated render
+  language. Operators who want fully localised descriptions still use
+  `customLocaleStrings` to override `pad.social.description` per-language.
+
+Documentation lives next to `publicURL` in both `settings.json.template`
+and `settings.json.docker` (mirrors how the original feature is
+configured), and the `customLocaleStrings` example now shows the
+`pad.social.description` key explicitly so operators can find both routes.

settings.json.docker

@@ -125,6 +125,19 @@
    */
   "publicURL": "${PUBLIC_URL:null}",
 
+  /*
+   * Open Graph / Twitter Card metadata for link previews.
+   *
+   * SOCIAL_META_DESCRIPTION: when set, this exact text is used as
+   * og:description regardless of negotiated language. Most preview crawlers
+   * (WhatsApp, Signal, Slack, ...) don't send Accept-Language, so without an
+   * override they always hit the English fallback in the i18n catalog.
+   * Leave unset (null) to use the catalog (key `pad.social.description`).
+   */
+  "socialMeta": {
+    "description": "${SOCIAL_META_DESCRIPTION:null}"
+  },
+
   /*
    * Skin name.
    *

settings.json.template

@@ -123,6 +123,26 @@
    */
   "publicURL": null,
 
+  /*
+   * Open Graph / Twitter Card metadata, served on the homepage, pad pages and
+   * timeslider for nicer previews when a pad URL is shared in chat apps
+   * (WhatsApp, Signal, Slack, ...).
+   *
+   * - description: when set to a non-empty string, this exact text is used as
+   *   og:description / twitter:description regardless of the visitor's
+   *   negotiated language. Most link-preview crawlers don't send an
+   *   Accept-Language header, so without an override they always see the
+   *   English fallback. Set this if your instance serves a non-English
+   *   audience and you want a fixed blurb in shared previews.
+   *
+   * Leave description as null to use Etherpad's i18n catalog (key
+   * `pad.social.description`), which honours Accept-Language and can be
+   * overridden per-language via `customLocaleStrings` further down.
+   */
+  "socialMeta": {
+    "description": null
+  },
+
   /*
    * Skin name.
    *
@@ -820,7 +840,19 @@
    */
    "logLayoutType": "colored",
 
-  /* Override any strings found in locale directories */
+  /*
+   * Override any strings found in locale directories.
+   *
+   * Format: { "<lang>": { "<key>": "<text>", ... }, ... }
+   * Example, per-language Open Graph description for link previews:
+   *   "customLocaleStrings": {
+   *     "en": { "pad.social.description": "Our team's collaborative pads." },
+   *     "de": { "pad.social.description": "Kollaborative Notizblöcke." }
+   *   }
+   * For a single description regardless of language, prefer
+   * `socialMeta.description` above — link-preview crawlers usually don't
+   * send Accept-Language and otherwise hit the English fallback.
+   */
   "customLocaleStrings": {},
 
   /* Disable Admin UI tests */

src/node/utils/Settings.ts

@@ -165,6 +165,9 @@ export type SettingsType = {
   showRecentPads: boolean,
   favicon: string | null,
   publicURL: string | null,
+  socialMeta: {
+    description: string | null,
+  },
   ttl: {
     AccessToken: number,
     AuthorizationCode: number,
@@ -360,6 +363,24 @@ const settings: SettingsType = {
    * No trailing slash. Must include scheme.
    */
   publicURL: null,
+
+  /**
+   * Open Graph / Twitter Card metadata, served on the homepage, pad pages and
+   * timeslider for nicer previews when a pad URL is shared in chat apps.
+   *
+   * description: when non-null, this exact string is used as og:description /
+   *   twitter:description regardless of the visitor's negotiated language. Most
+   *   crawlers (WhatsApp, Signal, Telegram, Slack, Facebook) don't send an
+   *   Accept-Language header, so without an override they always see the
+   *   English fallback — set this if your instance serves a non-English
+   *   audience and you want a fixed blurb. Leave null to use Etherpad's
+   *   built-in i18n catalog (key `pad.social.description`), which honours the
+   *   visitor's Accept-Language and can be overridden per-language via the
+   *   standard `customLocaleStrings` mechanism below.
+   */
+  socialMeta: {
+    description: null,
+  },
   ttl: {
     AccessToken: 1 * 60 * 60, // 1 hour in seconds
     AuthorizationCode: 10 * 60, // 10 minutes in seconds

src/node/utils/socialMeta.ts

@@ -9,8 +9,13 @@ import type {Request} from 'express';
  * XSS via crafted pad IDs.
  *
  * The description text is sourced from Etherpad's i18n catalog under the key
- * `pad.social.description`. Operators can override it per-language via the
- * standard `customLocaleStrings` mechanism in settings.json.
+ * `pad.social.description`. Operators have two ways to override it:
+ *   - `settings.socialMeta.description` — flat string used regardless of
+ *     negotiated language. Useful because most link-preview crawlers don't
+ *     send Accept-Language and would otherwise always hit the English fallback.
+ *   - `customLocaleStrings` — per-language override that participates in
+ *     normal Accept-Language negotiation.
+ * The flat setting wins over the i18n catalog when set.
  */
 
 const SOCIAL_DESCRIPTION_KEY = 'pad.social.description';
@@ -96,6 +101,9 @@ type SocialMetaSettings = {
   title?: string,
   favicon?: string | null,
   publicURL?: string | null,
+  socialMeta?: {
+    description?: string | null,
+  },
 };
 
 const negotiateRenderLang = (req: Request, availableLangs: AvailableLangs): string => {
@@ -153,10 +161,25 @@ export type RenderOpts = {
   padName?: string,
 };
 
+// Operator override wins, but only when it's a non-empty string. An empty
+// string from settings would silently blank out og:description / twitter:
+// description and break previews, so we treat empty/whitespace-only as unset
+// and fall back to the i18n catalog.
+const resolveDescriptionWithOverride = (
+  override: string | null | undefined,
+  locales: {[lang: string]: {[key: string]: string}} | undefined,
+  renderLang: string,
+): string => {
+  if (typeof override === 'string' && override.trim() !== '') return override;
+  return resolveDescription(locales, renderLang);
+};
+
 export const renderSocialMeta = (o: RenderOpts): string => {
   const renderLang = negotiateRenderLang(o.req, o.availableLangs);
   const siteName = o.settings.title || 'Etherpad';
-  const description = resolveDescription(o.locales, renderLang);
+  const description = resolveDescriptionWithOverride(
+      o.settings.socialMeta && o.settings.socialMeta.description,
+      o.locales, renderLang);
   const imageUrl = resolveImageUrl(o.req, o.settings.favicon, o.settings.publicURL);
   const imageAlt = `${siteName} logo`;
 

src/tests/backend/specs/socialMeta-unit.ts

@@ -168,6 +168,89 @@ describe(__filename, function () {
     });
   });
 
+  describe('renderSocialMeta — settings.socialMeta.description override', function () {
+    it('overrides i18n catalog regardless of negotiated language', function () {
+      // Crawler sends de, catalog has both en and de entries — operator
+      // override wins anyway. This is the crawler-no-Accept-Language case.
+      const html = renderSocialMeta({
+        req: fakeReq({acceptsLanguages: () => 'de'}),
+        settings: {
+          title: 'Etherpad', favicon: null,
+          socialMeta: {description: 'Operator-set blurb'},
+        },
+        availableLangs: {en: {}, de: {}},
+        locales: {
+          en: {'pad.social.description': 'En catalog'},
+          de: {'pad.social.description': 'De catalog'},
+        },
+        kind: 'pad', padName: 'P',
+      });
+      assert.equal(ogTag(html, 'og:description'), 'Operator-set blurb');
+      assert.equal(ogTag(html, 'twitter:description'), 'Operator-set blurb');
+    });
+
+    it('null override falls back to i18n catalog', function () {
+      const html = renderSocialMeta({
+        req: fakeReq({acceptsLanguages: () => 'de'}),
+        settings: {
+          title: 'Etherpad', favicon: null,
+          socialMeta: {description: null},
+        },
+        availableLangs: {en: {}, de: {}},
+        locales: {
+          en: {'pad.social.description': 'En'},
+          de: {'pad.social.description': 'De'},
+        },
+        kind: 'pad', padName: 'P',
+      });
+      assert.equal(ogTag(html, 'og:description'), 'De');
+    });
+
+    it('empty / whitespace override does NOT silence the description', function () {
+      // An accidental empty string in settings.json must not blank out the tag —
+      // we'd lose previews entirely. Treat it as unset.
+      for (const blank of ['', '   ', '\t\n']) {
+        const html = renderSocialMeta({
+          req: fakeReq({acceptsLanguages: () => 'en'}),
+          settings: {
+            title: 'Etherpad', favicon: null,
+            socialMeta: {description: blank},
+          },
+          availableLangs: {en: {}},
+          locales: {en: {'pad.social.description': 'Catalog wins'}},
+          kind: 'pad', padName: 'P',
+        });
+        assert.equal(ogTag(html, 'og:description'), 'Catalog wins',
+            `blank override (${JSON.stringify(blank)}) should fall back`);
+      }
+    });
+
+    it('HTML-escapes the override (it is operator-controlled but renders into HTML)', function () {
+      const html = renderSocialMeta({
+        req: fakeReq(),
+        settings: {
+          title: 'Etherpad', favicon: null,
+          socialMeta: {description: 'A & B "<C>"'},
+        },
+        availableLangs: {en: {}}, locales: enLocales,
+        kind: 'pad', padName: 'P',
+      });
+      assert.equal(ogTag(html, 'og:description'), 'A &amp; B &quot;&lt;C&gt;&quot;');
+    });
+
+    it('missing socialMeta block is treated as unset', function () {
+      // Older settings.json files won't have the socialMeta block at all.
+      const html = renderSocialMeta({
+        req: fakeReq({acceptsLanguages: () => 'en'}),
+        settings: {title: 'Etherpad', favicon: null},
+        availableLangs: {en: {}},
+        locales: {en: {'pad.social.description': 'Catalog'}},
+        kind: 'pad', padName: 'P',
+      });
+      assert.equal(ogTag(html, 'og:description'), 'Catalog');
+    });
+  });
+
   describe('renderSocialMeta — image URL', function () {
     it('builds absolute URL to /favicon.ico when settings.favicon is null', function () {
       const html = renderSocialMeta({

src/tests/backend/specs/socialMeta.ts

@@ -24,11 +24,15 @@ describe(__filename, function () {
   beforeEach(async function () {
     backup.title = settings.title;
     backup.favicon = settings.favicon;
+    backup.socialMeta = settings.socialMeta;
+    // Default shape — every test starts with no override.
+    settings.socialMeta = {description: null};
   });
 
   afterEach(async function () {
     settings.title = backup.title;
     settings.favicon = backup.favicon;
+    settings.socialMeta = backup.socialMeta;
   });
 
   describe('pad page', function () {
@@ -121,4 +125,45 @@ describe(__filename, function () {
       assert.equal(ogTag(res.text, 'og:title'), settings.title);
     });
   });
+
+  describe('settings.socialMeta.description override', function () {
+    it('overrides og:description and twitter:description', async function () {
+      settings.socialMeta = {description: 'Custom blurb for issue 7599'};
+      const res = await agent.get('/p/TestPad7599').expect(200);
+      assert.equal(ogTag(res.text, 'og:description'), 'Custom blurb for issue 7599');
+      assert.equal(ogTag(res.text, 'twitter:description'), 'Custom blurb for issue 7599');
+    });
+
+    it('override beats Accept-Language negotiation', async function () {
+      // Crawlers (WhatsApp/Signal/etc.) typically send no Accept-Language and
+      // would otherwise always hit the English fallback. Operator override
+      // wins regardless of the negotiated locale.
+      settings.socialMeta = {description: 'Operator wins'};
+      const res = await agent.get('/p/TestPad7599')
+          .set('Accept-Language', 'de').expect(200);
+      assert.equal(ogTag(res.text, 'og:description'), 'Operator wins');
+    });
+
+    it('blank override falls back to i18n catalog (does not silence preview)', async function () {
+      settings.socialMeta = {description: '   '};
+      const res = await agent.get('/p/TestPad7599')
+          .set('Accept-Language', 'en').expect(200);
+      const desc = ogTag(res.text, 'og:description');
+      assert.ok(desc && desc.length > 0,
+          'blank override should not blank out og:description');
+      assert.match(desc!, /collaborative/i);
+    });
+
+    it('HTML-escapes the override', async function () {
+      settings.socialMeta = {description: 'A & B <c> "d"'};
+      const res = await agent.get('/p/TestPad7599').expect(200);
+      // The HTML body contains the *escaped* form; the parsed attribute value
+      // (what ogTag returns) is the unescaped logical string the meta tag
+      // exposes — assert both: no raw <c> in the served HTML, and the logical
+      // value round-trips correctly.
+      assert.ok(!/content="[^"]*<c>/.test(res.text),
+          'raw "<c>" must not appear inside content="..."');
+      assert.equal(ogTag(res.text, 'og:description'), 'A &amp; B &lt;c&gt; &quot;d&quot;');
+    });
+  });
 });