wire scroll modifier through to hold_keys

n1.5's scroll action accepts an optional `modifier` (e.g., shift) that
on browsers translates a vertical wheel into a horizontal scroll. Plumb
it into Kernel's `ComputerScrollParams.hold_keys` so the OS-level
modifier+wheel event is dispatched correctly.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Author: dprevoznik

pkg/templates/python/yutori/tools/computer.py

@@ -199,13 +199,17 @@ async def _handle_scroll(self, action: N15Action) -> ToolResult:
         elif direction == "right":
             delta_x = notches
 
-        self.kernel.browsers.computer.scroll(
-            self.session_id,
-            x=coords["x"],
-            y=coords["y"],
-            delta_x=delta_x,
-            delta_y=delta_y,
-        )
+        modifier = action.get("modifier")
+        scroll_kwargs: dict[str, Any] = {
+            "x": coords["x"],
+            "y": coords["y"],
+            "delta_x": delta_x,
+            "delta_y": delta_y,
+        }
+        if modifier:
+            scroll_kwargs["hold_keys"] = [self._map_key(modifier)]
+
+        self.kernel.browsers.computer.scroll(self.session_id, **scroll_kwargs)
 
         await asyncio.sleep(SCREENSHOT_DELAY_S)
         screenshot_result = await self.screenshot()

pkg/templates/typescript/yutori/tools/computer.ts

@@ -229,11 +229,14 @@ export class ComputerTool {
         break;
     }
 
+    const holdKeys = action.modifier ? [this.mapKey(action.modifier)] : undefined;
+
     await this.kernel.browsers.computer.scroll(this.sessionId, {
       x: coords.x,
       y: coords.y,
       delta_x,
       delta_y,
+      ...(holdKeys ? { hold_keys: holdKeys } : {}),
     });
 
     await this.sleep(SCREENSHOT_DELAY_MS);