fix(agents): keep scp stderr tail UTF-16 safe end-to-end (#104148)

hugenshen · NIO · web-flow · commit 58ef98ec1311 · 2026-07-11T13:07:12.000-04:00
Co-authored-by: NIO &lt;nocodet@mail.com&gt;
diff --git a/src/auto-reply/reply.stage-sandbox-media.scp-remote-path.test.ts b/src/auto-reply/reply.stage-sandbox-media.scp-remote-path.test.ts
@@ -98,6 +98,15 @@ describe("stageSandboxMedia scp remote paths", () => {
     expect(stderr).not.toContain("start-");
   });
 
+  it("keeps scp stderr tail UTF-16 safe when the boundary bisects an emoji", () => {
+    const stderr = appendScpStderrTail("prefix", "🤖tail", 5);
+
+    expect(stderr).toBe("tail");
+    expect(
+      /[\uD800-\uDBFF](?![\uDC00-\uDFFF])|(?<![\uD800-\uDBFF])[\uDC00-\uDFFF]/.test(stderr),
+    ).toBe(false);
+  });
+
   it("rejects remote attachment filenames with shell metacharacters before spawning scp", async () => {
     await withSandboxMediaTempHome("openclaw-triggers-", async (home) => {
       const { cfg, workspaceDir, sessionKey, remoteCacheDir } = createRemoteStageParams(home);
diff --git a/src/auto-reply/reply/stage-sandbox-media.scp.test.ts b/src/auto-reply/reply/stage-sandbox-media.scp.test.ts
@@ -1,7 +1,10 @@
 import type { ChildProcess } from "node:child_process";
 import { EventEmitter } from "node:events";
 import { beforeEach, describe, expect, it, vi } from "vitest";
-import { testing } from "./stage-sandbox-media.js";
+import { SCP_STDERR_TAIL_CHARS, testing } from "./stage-sandbox-media.js";
+
+const hasUnpairedUtf16Surrogate = (text: string): boolean =>
+  /[\uD800-\uDBFF](?![\uDC00-\uDFFF])|(?<![\uD800-\uDBFF])[\uDC00-\uDFFF]/.test(text);
 
 const { spawnMock } = vi.hoisted(() => ({ spawnMock: vi.fn() }));
 
@@ -48,6 +51,31 @@ describe("scpFile", () => {
     await expect(resultPromise).rejects.toThrow("scp failed (1): stderr EPIPE");
   });
 
+  it("surfaces UTF-16 safe scp stderr when transfer fails with emoji at tail boundary", async () => {
+    const { child, stderr } = createChild();
+    // Place the retained tail window on the emoji's low surrogate so raw slicing
+    // would keep a lone surrogate half before the thrown error is built.
+    const lowSurrogateTailStart = 100;
+    const padding = "n".repeat(lowSurrogateTailStart - 1);
+    const recent = "🤖" + "n".repeat(SCP_STDERR_TAIL_CHARS - 5) + "fail";
+
+    const resultPromise = testing.scpFile("host", "/remote/path", "/local/path");
+    stderr.emit("data", padding);
+    stderr.emit("data", recent);
+    child.emit("close", 1);
+
+    let message = "";
+    try {
+      await resultPromise;
+    } catch (error) {
+      message = error instanceof Error ? error.message : String(error);
+    }
+    expect(message).toMatch(/^scp failed \(1\):/);
+    expect(message).toContain("fail");
+    expect(message).not.toContain("🤖");
+    expect(hasUnpairedUtf16Surrogate(message)).toBe(false);
+  });
+
   it("does not terminate scp again when spawning fails", async () => {
     const { child, kill } = createChild();
 
diff --git a/src/auto-reply/reply/stage-sandbox-media.ts b/src/auto-reply/reply/stage-sandbox-media.ts
@@ -6,6 +6,7 @@ import path from "node:path";
 import { fileURLToPath } from "node:url";
 import { isInboundPathAllowed } from "@openclaw/media-core/inbound-path-policy";
 import { normalizeOptionalString } from "@openclaw/normalization-core/string-coerce";
+import { sliceUtf16Safe } from "@openclaw/normalization-core/utf16-slice";
 import { assertSandboxPath } from "../../agents/sandbox-paths.js";
 import { ensureSandboxWorkspaceForSession } from "../../agents/sandbox.js";
 import { slugifySessionKey } from "../../agents/sandbox/shared.js";
@@ -427,7 +428,7 @@ export function appendScpStderrTail(
   if (combined.length <= maxChars) {
     return combined;
   }
-  return combined.slice(-maxChars);
+  return sliceUtf16Safe(combined, Math.max(0, combined.length - maxChars));
 }
 
 export const testing = { scpFile } as const;
