fix(chunk): keep surrogate pairs whole when hard-splitting an over-long line (#96951)

Bartok9 · ly-wang19 · claude · web-flow · commit 29680046801d · 2026-06-27T05:21:45.000+08:00
chunkByNewline length-splits a single over-long line that has no usable break
point. The first head cut used a raw UTF-16 slice (lineValue.slice(0,
firstLimit)), so when firstLimit landed inside a surrogate pair it emitted a
chunk ending in a lone high surrogate and a next chunk starting with a lone low
surrogate, which render as U+FFFD on delivery. The recursive chunkText that
handles the remainder is already surrogate-safe; only this first cut was raw.

Route the head cut through avoidTrailingHighSurrogateBreak (the same helper
chunkText and chunkMarkdownText already use) so the cut backs off to a
code-point boundary. ASCII and non-surrogate cuts are unaffected. Reproduces at
the production-default 4096 limit for emoji-dense lines; chunkByNewline is the
published plugin-SDK channel.text helper, reachable with arbitrary outbound text.

Co-authored-by: ly-wang19 &lt;ly-wang19@users.noreply.github.com&gt;
Co-authored-by: Claude Opus 4.8 (1M context) &lt;noreply@anthropic.com&gt;
diff --git a/src/auto-reply/chunk.test.ts b/src/auto-reply/chunk.test.ts
@@ -513,6 +513,19 @@ describe("chunkByNewline", () => {
   it.each(["", "   \n\n   "] as const)("returns empty array for input %j", (text) => {
     expect(chunkByNewline(text, 100)).toStrictEqual([]);
   });
+
+  it("does not split surrogate pairs when hard-splitting an over-long line", () => {
+    // An emoji-dense line with no break point forces the raw head cut at an odd code-unit offset;
+    // it must back off to a code-point boundary so no chunk ends in a high (or starts with a low)
+    // surrogate — the same contract the recursive chunkText path already honors.
+    const text = "😀".repeat(30);
+    const chunks = chunkByNewline(text, 11);
+
+    expect(chunks.join("")).toBe(text);
+    expect(chunks.length).toBeGreaterThan(1);
+    expect(chunks.every((chunk) => !/[\uD800-\uDBFF]$/u.test(chunk))).toBe(true);
+    expect(chunks.every((chunk) => !/^[\uDC00-\uDFFF]/u.test(chunk))).toBe(true);
+  });
 });
 
 describe("chunkTextWithMode", () => {
diff --git a/src/auto-reply/chunk.ts b/src/auto-reply/chunk.ts
@@ -163,7 +163,10 @@ export function chunkByNewline(
       continue;
     }
 
-    const firstLimit = Math.max(1, maxLineLength - prefix.length);
+    // Back the head cut off to a code-point boundary so an over-long line never splits a surrogate
+    // pair; the recursive chunkText below is already surrogate-safe, only this first cut was raw.
+    const rawLimit = Math.max(1, maxLineLength - prefix.length);
+    const firstLimit = avoidTrailingHighSurrogateBreak(lineValue, 0, rawLimit);
     const first = lineValue.slice(0, firstLimit);
     chunks.push(prefix + first);
     const remaining = lineValue.slice(firstLimit);
