fix(desktop): accept coordinate input without a window, one-axis scrolls and targeted keys

Mirrors the worker contract. Coordinate clicks, scrolls and drags without a windowId
target the window the snapshot selected; scroll deltas default to 0; press_key with an
elementId focuses that element first. Each shape was rejected before reaching the app
in benchmark runs (about one call per task).

Author: waleedlatif1

apps/desktop/native/computer-use/ComputerUse.swift

@@ -224,6 +224,8 @@ final class Snapshot {
     /// The Accessibility window behind each observed windowId, so input checks compare identity, not
     /// just bounds that two windows can share.
     var windowElements: [String: AXUIElement] = [:]
+    /// The window this observation selected; coordinate input without a windowId targets it.
+    var selectedWindowId: String?
     var nodes: [[String: Any]] = []
     var truncated = false
     init(app: NSRunningApplication) { pid = app.processIdentifier; launchDate = app.launchDate }
@@ -502,7 +504,7 @@ final class Driver {
             guard let frame = elementRect(try snapshot.element(elementId)) else { throw ComputerError("element_unavailable", "Element has no usable frame.") }
             return CGPoint(x: frame.midX, y: frame.midY)
         }
-        let windowId = try required(p.windowId, "windowId")
+        let windowId = try required(p.windowId ?? snapshot.selectedWindowId, "windowId")
         let frame = try windowRect(windowId, pid: app.processIdentifier)
         guard snapshot.windowFrames[windowId] == frame else { throw ComputerError("stale_window", "Window moved or changed; observe it again.") }
         let x = try required(end ? p.toX : p.x, end ? "toX" : "x"); let y = try required(end ? p.toY : p.y, end ? "toY" : "y")
@@ -579,6 +581,7 @@ final class Driver {
             }
             guard let selectedWindowId = p.windowId ?? focusedID ?? windows.first?["windowId"] as? String else { throw ComputerError("app_window_unavailable", "No accessible app window is available. Open a window, then observe the app again.") }
             guard snapshot.windowFrames[selectedWindowId] != nil else { throw ComputerError("window_unavailable", "Selected window does not belong to the app.") }
+            snapshot.selectedWindowId = selectedWindowId
             var result: [String: Any] = ["kind": "state", "windowId": selectedWindowId, "snapshotId": snapshot.id, "bundleId": app.bundleIdentifier!, "nodes": snapshot.nodes, "windows": windows, "truncated": snapshot.truncated]
             if p.includeScreenshot == true {
                 do {
@@ -730,6 +733,14 @@ final class Driver {
                 return frame
             }
             let root = AXUIElementCreateApplication(pid)
+            // A targeted key first moves keyboard focus to that element, as clicking into it would.
+            if let elementId = p.elementId {
+                let element = try snapshot.element(elementId)
+                if attribute(root, kAXFocusedUIElementAttribute).map({ CFEqual($0, element) }) != true {
+                    try axCheck(AXUIElementSetAttributeValue(element, kAXFocusedAttribute as CFString, kCFBooleanTrue))
+                    try await Task.sleep(for: .milliseconds(50))
+                }
+            }
             // Keys go to the app's key window: the observed window, or a dialog, panel or popover above it,
             // but never another document window the snapshot did not observe.
             guard let rawWindow = attribute(root, kAXFocusedWindowAttribute), CFGetTypeID(rawWindow) == AXUIElementGetTypeID() else { throw ComputerError("focus_failed", "The app has no focused window to receive keys; click a window or call activate_app, then observe again.") }
@@ -763,7 +774,7 @@ final class Driver {
         case "drag":
             try await coordinateAction { dispatch in
                 let start = try point(p, app: app, snapshot: snapshot); let end = try point(p, app: app, snapshot: snapshot, end: true)
-                let windowID = try required(p.windowId, "windowId"); let frame = try required(snapshot.windowFrames[windowID], "window frame")
+                let windowID = try required(p.windowId ?? snapshot.selectedWindowId, "windowId"); let frame = try required(snapshot.windowFrames[windowID], "window frame")
                 try postMouse(pid: pid, type: .leftMouseDown, point: start, button: .left, windowID: windowID, window: snapshot.windowElements[windowID], frame: frame, dispatch: dispatch)
                 var lastPoint = start
                 defer { try? postMouse(pid: pid, type: .leftMouseUp, point: lastPoint, button: .left, windowID: windowID, window: snapshot.windowElements[windowID], frame: frame, release: true, dispatch: dispatch) }

apps/desktop/native/computer-use/README.md

@@ -10,7 +10,7 @@ Persistent newline-delimited JSON over stdin/stdout: `{id,method,params}` return
 
 The helper observes up to 2000 accessibility nodes (depth 64, queue 4000, traversal deadline 8 seconds, each AX call timeout 250ms). Selected-window and focused-element ancestry are prioritized; menu traversal is deferred and capped at 80 nodes. Electron accessibility is enabled using its documented `AXManualAccessibility` attribute. The helper observes 100 windows and 1000 apps. Requests are bounded to 128 KiB during ingestion. Text/value input is capped at 32000 UTF-16 units. State references are per-process, per-app, expire after 60 seconds, and are consumed by every action that reaches the app; a request that fails before sending anything reports `dispatchState: "not_started"` and keeps them valid. Restarts invalidate all references. Screen captures exclude window shadows, preserve the selected window’s coordinate geometry, and are capped to 1600 pixels on the longest side and 8 MiB encoded. Complete JSON replies are capped at 16 MiB: oversized observations omit screenshots first, then truncate accessibility nodes, while preserving action outcomes. Screenshots are optional and failure is reported in `screenshotError` without discarding accessibility state. macOS 26 uses the screenshot-specific capture API; the macOS 14/15 API path includes only the selected window in a display filter and crops its exact frame. That older path requires the window to fit within one display and reports `capture_geometry_unavailable` otherwise.
 
-Coordinates in action inputs are window-local points; node frames are global screen points. Window bounds must still match the observed snapshot. App PID and launch time must match. Secure fields and descendants are unavailable as action targets; secure values and labels are suppressed. Sim's own apps and System Settings are protected targets. `get_app_state` can open an installed bundle ID without activating it. `list_apps` discovers installed apps under `/Applications`, `/System/Applications`, and `~/Applications` (bounded to 5000 filesystem entries and depth 3), merging running regular apps; installed apps omit `pid`.
+Coordinates in action inputs are window-local points; node frames are global screen points. Coordinate input without a `windowId` targets the window the snapshot selected; `press_key` with an `elementId` focuses that element first. Window bounds must still match the observed snapshot. App PID and launch time must match. Secure fields and descendants are unavailable as action targets; secure values and labels are suppressed. Sim's own apps and System Settings are protected targets. `get_app_state` can open an installed bundle ID without activating it. `list_apps` discovers installed apps under `/Applications`, `/System/Applications`, and `~/Applications` (bounded to 5000 filesystem entries and depth 3), merging running regular apps; installed apps omit `pid`.
 
 AXPress and AXSetValue are direct semantic operations. Keyboard events target the app PID. Cmd+A uses the focused editor's writable accessibility selection range where supported. Coordinate mouse input requires the target app in front; `activate_app` makes that focus change explicit and confirms it. Click/drag/scroll validate the foreground process, unchanged window geometry and the app that receives the point; the topmost window-list entry is not the click target because Notification Center and the Dock keep full-screen click-through windows. A press that times out while the app opens a sheet or modal (`cannotComplete`) counts as dispatched. A menu-item press that leaves its menu open with the item highlighted (Calendar on macOS 26) is completed with Return. Open menus and their submenus are observed first, and menu item labels carry their shortcuts (for example ⇧⌘N). Some apps ignore background keyboard events. `dispatched:true,verified:false` means only dispatch succeeded; the agent must observe again to establish the outcome. No foreground activation fallback is hidden inside the helper. Native key codes currently assume the US layout for shortcuts; literal text uses Unicode events. Unsupported shortcuts fail explicitly.
 

apps/sim/lib/mothership/generated/computer-use.ts

@@ -33,7 +33,15 @@ const InputSequenceStep = z.discriminatedUnion("action", [
 ]);
 const ElementTarget = { bundleId: BundleId, snapshotId: SnapshotId, elementId: ElementId };
 const WindowTarget = { bundleId: BundleId, snapshotId: SnapshotId, windowId: WindowId };
-const PointTarget = { ...WindowTarget, x: Coordinate, y: Coordinate };
+const PointTarget = {
+  bundleId: BundleId,
+  snapshotId: SnapshotId,
+  windowId: WindowId.optional().describe(
+    "Window reference from that snapshot. Defaults to the window the snapshot selected.",
+  ),
+  x: Coordinate,
+  y: Coordinate,
+};
 const ClickOptions = {
   button: z.enum(["left", "right"]).optional(),
   clickCount: z.number().int().min(1).max(3).optional(),
@@ -43,12 +51,14 @@ const ScrollOffsets = {
     .number()
     .min(-10_000)
     .max(10_000)
-    .describe("Horizontal scroll distance in points; positive scrolls right, negative left."),
+    .optional()
+    .describe("Horizontal scroll distance in points; positive scrolls right, negative left. Defaults to 0."),
   deltaY: z
     .number()
     .min(-10_000)
     .max(10_000)
-    .describe("Vertical scroll distance in points; positive scrolls down, negative up."),
+    .optional()
+    .describe("Vertical scroll distance in points; positive scrolls down, negative up. Defaults to 0."),
 };
 
 /** Strict branches require an observed, unambiguous target before native input. */
@@ -81,6 +91,7 @@ export const ComputerUseSchema = z.union([
     windowId: WindowId.optional().describe(
       "Window reference from that snapshot. Omit to send the key to the app's focused window.",
     ),
+    elementId: ElementId.optional().describe("Focus this element from the snapshot before pressing the key."),
     key: z.string().min(1).max(128).describe("One key or chord, such as Enter, Tab, Escape, or Cmd+A."),
   }),
   z.strictObject({ action: z.literal("scroll"), ...ElementTarget, ...ObserveAfter, ...ScrollOffsets }),

packages/desktop-bridge/src/computer-use.generated.ts

@@ -33,7 +33,15 @@ const InputSequenceStep = z.discriminatedUnion("action", [
 ]);
 const ElementTarget = { bundleId: BundleId, snapshotId: SnapshotId, elementId: ElementId };
 const WindowTarget = { bundleId: BundleId, snapshotId: SnapshotId, windowId: WindowId };
-const PointTarget = { ...WindowTarget, x: Coordinate, y: Coordinate };
+const PointTarget = {
+  bundleId: BundleId,
+  snapshotId: SnapshotId,
+  windowId: WindowId.optional().describe(
+    "Window reference from that snapshot. Defaults to the window the snapshot selected.",
+  ),
+  x: Coordinate,
+  y: Coordinate,
+};
 const ClickOptions = {
   button: z.enum(["left", "right"]).optional(),
   clickCount: z.number().int().min(1).max(3).optional(),
@@ -43,12 +51,14 @@ const ScrollOffsets = {
     .number()
     .min(-10_000)
     .max(10_000)
-    .describe("Horizontal scroll distance in points; positive scrolls right, negative left."),
+    .optional()
+    .describe("Horizontal scroll distance in points; positive scrolls right, negative left. Defaults to 0."),
   deltaY: z
     .number()
     .min(-10_000)
     .max(10_000)
-    .describe("Vertical scroll distance in points; positive scrolls down, negative up."),
+    .optional()
+    .describe("Vertical scroll distance in points; positive scrolls down, negative up. Defaults to 0."),
 };
 
 /** Strict branches require an observed, unambiguous target before native input. */
@@ -81,6 +91,7 @@ export const ComputerUseSchema = z.union([
     windowId: WindowId.optional().describe(
       "Window reference from that snapshot. Omit to send the key to the app's focused window.",
     ),
+    elementId: ElementId.optional().describe("Focus this element from the snapshot before pressing the key."),
     key: z.string().min(1).max(128).describe("One key or chord, such as Enter, Tab, Escape, or Cmd+A."),
   }),
   z.strictObject({ action: z.literal("scroll"), ...ElementTarget, ...ObserveAfter, ...ScrollOffsets }),