From b3428667d48fb68fa52f72963194aad5e9cb61e9 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E7=81=B5=E8=B6=8A=E7=BE=BD=E6=AF=9B?=
 <97326386+Icather@users.noreply.github.com>
Date: Mon, 15 Jun 2026 17:00:34 +0800
Subject: [PATCH] fix(computer_use): address Copilot review feedback

- Validate direction before key dispatch (unknown values return False)
- Log exceptions instead of silently swallowing them
- Add 150ms delay before overlay restart to avoid DWM race
- Route switch_desktop through _maybe_follow_capture for consistency
- Add JSON Schema if/then to require direction for switch_desktop
---
 tools/computer_use/schema.py          | 11 +++++++++++
 tools/computer_use/tool.py            |  2 +-
 tools/computer_use/windows_backend.py | 13 ++++++++++---
 3 files changed, 22 insertions(+), 4 deletions(-)

diff --git a/tools/computer_use/schema.py b/tools/computer_use/schema.py
index b83ab60f6047..22259377ca7c 100644
--- a/tools/computer_use/schema.py
+++ b/tools/computer_use/schema.py
@@ -230,6 +230,17 @@ COMPUTER_USE_SCHEMA: Dict[str, Any] = {
             },
         },
         "required": ["action"],
+        "allOf": [
+            {
+                "if": {
+                    "properties": {"action": {"const": "switch_desktop"}},
+                    "required": ["action"],
+                },
+                "then": {
+                    "required": ["direction"],
+                },
+            },
+        ],
     },
 }
 
diff --git a/tools/computer_use/tool.py b/tools/computer_use/tool.py
index e4be71b8f0c1..c83183b62230 100644
--- a/tools/computer_use/tool.py
+++ b/tools/computer_use/tool.py
@@ -431,7 +431,7 @@ def _dispatch(backend: ComputerUseBackend, action: str, args: Dict[str, Any]) ->
         if not hasattr(backend, "switch_desktop"):
             return json.dumps({"error": "switch_desktop not supported by current backend"})
         res = backend.switch_desktop(str(direction))
-        return json.dumps({"ok": res.ok, "message": res.message})
+        return _maybe_follow_capture(backend, res, capture_after)
 
     return json.dumps({"error": f"unknown action {action!r}"})
 
diff --git a/tools/computer_use/windows_backend.py b/tools/computer_use/windows_backend.py
index 62db4bb4c10c..f9e878f29182 100644
--- a/tools/computer_use/windows_backend.py
+++ b/tools/computer_use/windows_backend.py
@@ -302,6 +302,9 @@ def _switch_desktop_via_keybd(direction: str, overlay_client) -> bool:
     virtual-desktop transition, so we stop it before switching and restart
     it on the new desktop.
     """
+    if direction not in ("left", "right"):
+        return False
+
     VK_CONTROL = 0x11
     VK_LWIN = 0x5B
     VK_LEFT = 0x25
@@ -322,15 +325,19 @@ def _switch_desktop_via_keybd(direction: str, overlay_client) -> bool:
     try:
         overlay_client.stop()
         _send_inputs(seq)
+        # The virtual-desktop transition is asynchronous — a brief delay
+        # before restarting the overlay avoids racing the DWM compositor.
+        time.sleep(0.15)
         overlay_client._dead = False
         overlay_client.start()
         return True
-    except Exception:
+    except Exception as e:
+        logger.warning("switch_desktop SendInput failed: %s", e)
         try:
             overlay_client._dead = False
             overlay_client.start()
-        except Exception:
-            pass
+        except Exception as e2:
+            logger.warning("switch_desktop overlay restart also failed: %s", e2)
         return False