fix(agent-core): avoid splitting surrogate pairs in grep line truncation (#97559)

truncateLine could cut a surrogate pair when the maxChars boundary
fell between a high surrogate and its paired low surrogate, producing
a broken unpaired surrogate in grep tool output.

packages/agent-core/src/harness/utils/truncate.test.ts

@@ -1,6 +1,6 @@
 // Agent Core tests cover truncate behavior.
 import { describe, expect, it } from "vitest";
-import { truncateHead, truncateTail } from "./truncate.js";
+import { truncateHead, truncateLine, truncateTail } from "./truncate.js";
 
 describe("truncate utilities", () => {
   it("does not count a trailing newline as an extra display line", () => {
@@ -20,4 +20,44 @@ describe("truncate utilities", () => {
     expect(result.lastLinePartial).toBe(true);
     expect(result.outputBytes).toBe(4);
   });
+
+  describe("truncateLine", () => {
+    it("returns text unchanged when within limit", () => {
+      expect(truncateLine("short", 10)).toEqual({ text: "short", wasTruncated: false });
+    });
+
+    it("truncates and appends suffix when over limit", () => {
+      const result = truncateLine("this is a very long line", 10);
+      expect(result.wasTruncated).toBe(true);
+      expect(result.text).toBe("this is a ... [truncated]");
+    });
+
+    it("uses GREP_MAX_LINE_LENGTH as the default limit", () => {
+      const result = truncateLine("x");
+      expect(result.wasTruncated).toBe(false);
+      expect(result.text).toBe("x");
+    });
+
+    it("does not split a surrogate pair at the cut point", () => {
+      // Emoji at boundary: "AB" + 🤖(surrogate pair) + "CD" — cut at 3 splits the emoji.
+      expect(truncateLine("AB🤖CD", 3).text).toBe("AB... [truncated]");
+      // Three emoji, cut in the middle of the second emoji.
+      expect(truncateLine("🤖🤖🤖", 5).text).toBe("🤖🤖... [truncated]");
+      // CJK Extension B (surrogate pair) at boundary stays intact.
+      expect(truncateLine("AB𠮷CD", 5).text).toBe("AB𠮷C... [truncated]");
+    });
+
+    it("never produces unpaired surrogates in output", () => {
+      const results = [
+        truncateLine("AB🤖CD", 3).text,
+        truncateLine("🤖🤖🤖", 5).text,
+        truncateLine("AB𠮷CD", 5).text,
+      ];
+      for (const text of results) {
+        expect(text).not.toMatch(
+          /[\uD800-\uDBFF](?![\uDC00-\uDFFF])|(?<![\uD800-\uDBFF])[\uDC00-\uDFFF]/,
+        );
+      }
+    });
+  });
 });

packages/agent-core/src/harness/utils/truncate.ts

@@ -359,6 +359,10 @@ function truncateStringToBytesFromEnd(str: string, maxBytes: number): string {
 
 /**
  * Trim a single display line and mark it with the grep-style truncation suffix.
+ *
+ * The cut point is backed off by one code unit when it would otherwise split a
+ * surrogate pair, so emoji / CJK Extension B characters crossing the boundary
+ * stay intact instead of rendering as replacement characters.
  */
 export function truncateLine(
   line: string,
@@ -367,5 +371,16 @@ export function truncateLine(
   if (line.length <= maxChars) {
     return { text: line, wasTruncated: false };
   }
-  return { text: `${line.slice(0, maxChars)}... [truncated]`, wasTruncated: true };
+  let cut = maxChars;
+  // Avoid splitting a surrogate pair at the truncation boundary.
+  if (cut < line.length) {
+    const lastCode = line.charCodeAt(cut - 1);
+    if (lastCode >= 0xd800 && lastCode <= 0xdbff) {
+      const nextCode = line.charCodeAt(cut);
+      if (nextCode >= 0xdc00 && nextCode <= 0xdfff) {
+        cut -= 1;
+      }
+    }
+  }
+  return { text: `${line.slice(0, cut)}... [truncated]`, wasTruncated: true };
 }