diff --git a/skills/open-kimi-ppt/scripts/export_pptx.py b/skills/open-kimi-ppt/scripts/export_pptx.py
index 92bd9ae..1bf8c1c 100755
--- a/skills/open-kimi-ppt/scripts/export_pptx.py
+++ b/skills/open-kimi-ppt/scripts/export_pptx.py
@@ -461,18 +461,23 @@ def find_download(
     last_sizes: Dict[Path, int] = {}
     stable: Dict[Path, int] = {}
     while time.monotonic() < deadline:
-        candidates: List[Path] = []
+        # Snapshot stats while collecting and tolerate races everywhere: the
+        # search roots include the live Downloads folder, where Chrome renames
+        # .crdownload files away between directory listing and stat().
+        entries: List[Tuple[Path, float, int]] = []
         for root in search_roots:
             if not root.exists():
                 continue
-            candidates.extend(path for path in root.rglob("*") if path.is_file())
-        for path in sorted(candidates, key=lambda item: item.stat().st_mtime, reverse=True):
-            try:
-                stat = path.stat()
-                size = stat.st_size
-                if since is not None and stat.st_mtime < since:
+            for path in root.rglob("*"):
+                if not path.is_file():
                     continue
-            except OSError:
+                try:
+                    info = path.stat()
+                except OSError:
+                    continue
+                entries.append((path, info.st_mtime, info.st_size))
+        for path, mtime, size in sorted(entries, key=lambda entry: entry[1], reverse=True):
+            if since is not None and mtime < since:
                 continue
             if size == last_sizes.get(path) and size > 0:
                 stable[path] = stable.get(path, 0) + 1
diff --git a/skills/open-kimi-ppt/tests/test_export_pptx.py b/skills/open-kimi-ppt/tests/test_export_pptx.py
index 24d5009..29640bb 100755
--- a/skills/open-kimi-ppt/tests/test_export_pptx.py
+++ b/skills/open-kimi-ppt/tests/test_export_pptx.py
@@ -163,6 +163,37 @@ class ExportPptxTests(unittest.TestCase):
             found = MODULE.find_download([root], timeout=2.0, since=since)
             self.assertEqual(found.resolve(), new.resolve())
 
+    def test_find_download_survives_files_vanishing_mid_scan(self):
+        # Chrome renames "*.crdownload" files away between directory listing
+        # and stat(); a vanished file must be skipped, not crash the export.
+        with tempfile.TemporaryDirectory() as name:
+            root = Path(name)
+            deck = root / "deck.pptx"
+            with zipfile.ZipFile(deck, "w") as archive:
+                archive.writestr(
+                    "[Content_Types].xml",
+                    '<Types xmlns="http://schemas.openxmlformats.org/package/2006/content-types">'
+                    '<Override PartName="/ppt/presentation.xml" '
+                    f'ContentType="{MODULE.PPTX_CONTENT_TYPE}"/></Types>',
+                )
+                archive.writestr("ppt/presentation.xml", "<p:presentation/>")
+            ghost = root / "ghost.crdownload"
+            ghost.write_bytes(b"partial download")
+
+            real_stat = Path.stat
+            seen = {"count": 0}
+
+            def racy_stat(self, **kwargs):
+                if self.name == "ghost.crdownload":
+                    seen["count"] += 1
+                    if seen["count"] > 1:
+                        raise FileNotFoundError(2, "vanished mid-scan", str(self))
+                return real_stat(self, **kwargs)
+
+            with patch.object(Path, "stat", racy_stat):
+                found = MODULE.find_download([root], timeout=2.0)
+            self.assertEqual(found.resolve(), deck.resolve())
+
     def test_browser_open_does_not_pass_download_path(self):
         session = MODULE.BrowserSession(
             "/bin/agent-browser",