fix(bookmarks): avoid splitting surrogate pairs in automatic labels (#7966)

* fix(bookmarks): avoid splitting surrogate pairs in automatic labels

* fix(bookmarks): sanitize unpaired surrogates before save

examples/deerflow-extension-bookmarks/README.md

@@ -73,6 +73,10 @@ file above stores user bookmarks, not deployment settings.
   most 10 excerpts (1,000 characters each) and a total count. Narrow the query for
   more matches; the page can load the full saved answer. Duplicate saves are
   idempotent and do not overwrite an existing label.
+  Automatic labels keep up to 80 UTF-16 code units after whitespace normalization,
+  stopping before a character if the cut would split its surrogate pair.
+  Unpaired UTF-16 surrogates are replaced with U+FFFD in the saved copy and its
+  label.
 - Storage uses SQLite and blocking I/O is offloaded from the event loop. This is a
   single-host example, not a multi-node storage contract or tenant administration UI.
 - Plugin browser/Python code is trusted deployment code. Shadow DOM scopes CSS,

examples/deerflow-extension-bookmarks/deerflow_extension_bookmarks/static/dist/index.mjs

@@ -44,11 +44,18 @@ export default {
               );
               return;
             }
+            // Unicode mode matches lone surrogates without touching intact pairs.
+            const text = answer.text.replace(/[\uD800-\uDFFF]/gu, "\uFFFD");
             await services.callBackend("save", {
               thread_id: context.thread.thread_id,
               message_id: answer.id,
-              label: answer.text.replace(/\s+/g, " ").trim().slice(0, 80),
-              text: answer.text,
+              // Keep the UTF-16 bound without leaving half a surrogate pair.
+              label: text
+                .replace(/\s+/g, " ")
+                .trim()
+                .slice(0, 80)
+                .replace(/[\uD800-\uDBFF]$/u, ""),
+              text,
             });
             services.showMessage(
               zh

frontend/tests/unit/core/extensions/bookmarks.test.ts

@@ -0,0 +1,40 @@
+import { expect, rs, test } from "@rstest/core";
+
+import bookmarks from "../../../../../examples/deerflow-extension-bookmarks/deerflow_extension_bookmarks/static/dist/index.mjs";
+
+const prefix = "a".repeat(77);
+
+test.each([
+  ["split surrogate pair", `${prefix}\t b😀 remaining answer`, `${prefix} b`],
+  ["complete surrogate pair", `${prefix}\t😀`, `${prefix} 😀`],
+  ["consecutive emoji", "😀".repeat(80), "😀".repeat(40)],
+  [
+    "unpaired surrogates",
+    "a\uD800b\uDC00c",
+    "a\uFFFDb\uFFFDc",
+    "a\uFFFDb\uFFFDc",
+  ],
+])(
+  "keeps bookmark labels within 80 UTF-16 units: %s",
+  async (_name, answer, label, savedAnswer = answer) => {
+    const text = `  ${answer}\n`;
+    const callBackend = rs.fn();
+    const action = bookmarks.conversationActions(undefined, "en").actions[0]!;
+
+    await action.execute(
+      { thread: { thread_id: "thread-1" } },
+      {
+        latestVisibleAnswer: async () => ({ id: "answer-1", text }),
+        callBackend,
+        showMessage: rs.fn(),
+      },
+    );
+
+    expect(callBackend).toHaveBeenCalledExactlyOnceWith("save", {
+      thread_id: "thread-1",
+      message_id: "answer-1",
+      label,
+      text: `  ${savedAnswer}\n`,
+    });
+  },
+);