fix(browser): cancel drags during their settle and keep parked agent tabs in the main window

- An already-aborted drag sends no input, and an abort during the settle hold
  takes the cancellation cleanup instead of dropping.
- A coordinate hover re-checks the active tab right before it moves.
- Parking an agent tab hands keyboard focus back to whichever of the Sim
  renderer or the visible page held it.
- An agent tab detached from a secondary window is parked in the main window,
  which outlives it.
- A framed target covered by a parent-page overlay lists that overlay's controls.

Author: waleedlatif1

apps/desktop/src/main/browser-agent/cdp.test.ts

@@ -294,6 +294,41 @@ describe('browser-agent CDP instrumentation', () => {
     }
   })
 
+  it('sends nothing for an already-aborted drag and cancels one aborted while it settles', async () => {
+    const contents = new WebContentsView().webContents
+    const mouse = () =>
+      vi
+        .mocked(contents.debugger.sendCommand)
+        .mock.calls.filter(([method]) => method === 'Input.dispatchMouseEvent')
+        .map(([, params]) => toRecord(params).type)
+    const aborted = new AbortController()
+    aborted.abort()
+    await expect(
+      dragPointer(contents, { x: 0, y: 0 }, { x: 50, y: 0 }, undefined, aborted.signal)
+    ).rejects.toMatchObject({ name: 'AbortError' })
+    expect(contents.debugger.sendCommand).not.toHaveBeenCalled()
+
+    vi.useFakeTimers()
+    try {
+      const controller = new AbortController()
+      const drag = dragPointer(
+        contents,
+        { x: 0, y: 0 },
+        { x: 50, y: 0 },
+        undefined,
+        controller.signal
+      )
+      const settled = expect(drag).rejects.toMatchObject({ name: 'AbortError' })
+      // The default route takes 13 moves 20 ms apart, then a 120 ms settle hold.
+      await vi.advanceTimersByTimeAsync(300)
+      controller.abort()
+      await settled
+      expect(mouse().at(-1)).toBe('mouseReleased')
+    } finally {
+      vi.useRealTimers()
+    }
+  })
+
   it('releases the button when a timed drag is aborted mid-route', async () => {
     const contents = new WebContentsView().webContents
     const types = () =>

apps/desktop/src/main/browser-agent/cdp.ts

@@ -11,7 +11,7 @@
 import type { BrowserTheme } from '@sim/browser-protocol'
 import { createLogger } from '@sim/logger'
 import { getErrorMessage } from '@sim/utils/errors'
-import { interruptibleSleep, sleep } from '@sim/utils/helpers'
+import { interruptibleSleep } from '@sim/utils/helpers'
 import { isRecordLike } from '@sim/utils/object'
 import type { NativeImage, WebContents, WebFrameMain } from 'electron'
 
@@ -1165,6 +1165,7 @@ export async function dragPointer(
   path: PointerPath = DIRECT_PATH,
   signal?: AbortSignal
 ): Promise<{ nativeDragIntercepted: boolean }> {
+  signal?.throwIfAborted()
   const { points, stepDelayMs } = pointerPathSteps(from, path, to)
   const interception: DragInterception = { intercepted: false, data: null }
   dragInterceptionsByContents.set(contents, interception)
@@ -1221,7 +1222,8 @@ export async function dragPointer(
     signal?.throwIfAborted()
     // Hold over the target so drop zones running enter/over animations settle
     // before the release lands.
-    await sleep(120)
+    await interruptibleSleep(120, signal)
+    signal?.throwIfAborted()
     if (interception.intercepted && interception.data) {
       await sendInput(contents, 'Input.dispatchDragEvent', {
         type: 'drop',

apps/desktop/src/main/browser-agent/driver.ts

@@ -4528,6 +4528,7 @@ async function executeToolInner(
         const beforePage = await pageActionState(contents, true)
         const beforeElement = await activeElementState(contents)
         assertCurrentExecution()
+        assertActiveContents(contents)
         await cdp.movePointer(contents, path, { x, y }, signal)
         await sleep(150)
         const afterElement = await activeElementState(contents)

apps/desktop/src/main/browser-agent/page-functions.ts

@@ -1482,7 +1482,11 @@ export function clickElement(
         if (!scrollToTarget) {
           return clickElement(id, dispatchSynthetic, focusForKeyboard, allowDisabled, true)
         }
-        return { error: 'obstructed', blocker: blockerLabel(parentHit) }
+        return {
+          error: 'obstructed',
+          blocker: blockerLabel(parentHit),
+          blockerControls: window.__simAgentOverlayControls(parentHit),
+        }
       }
     }
     ownerView = frame.ownerDocument.defaultView

apps/desktop/src/main/browser-agent/panel.ts

@@ -315,7 +315,8 @@ function detachAttachedView(): void {
   if (!view || !win) return
   try {
     if (win.isDestroyed() || view.webContents.isDestroyed()) return
-    if (isAgentView(view)) {
+    // An agent view stays in the main window, which outlives any secondary window.
+    if (isAgentView(view) && win === host.getMainWindow()) {
       view.setVisible(false)
       parkedViews.set(view, win)
       return
@@ -326,6 +327,7 @@ function detachAttachedView(): void {
       error: getErrorMessage(error, 'unknown'),
     })
   }
+  if (isAgentView(view)) parkAgentViews()
 }
 
 /** Reveals the native view and invalidates every frame captured for its old state. */
@@ -399,7 +401,11 @@ function parkAgentViews(): void {
     if (view === attachedView || view.webContents.isDestroyed()) continue
     const parkedIn = parkedViews.get(view)
     if (parkedIn && !parkedIn.isDestroyed()) continue
-    const rendererHadFocus = !win.webContents.isDestroyed() && win.webContents.isFocused()
+    // addChildView hands keyboard focus to the parked view; give it back to whichever of the
+    // Sim renderer or the visible browser page held it.
+    const focused = [win.webContents, attachedView?.webContents].find(
+      (contents) => contents && !contents.isDestroyed() && contents.isFocused()
+    )
     try {
       view.setVisible(false)
       win.contentView.addChildView(view)
@@ -409,7 +415,7 @@ function parkAgentViews(): void {
         error: getErrorMessage(error, 'unknown'),
       })
     }
-    if (rendererHadFocus) win.webContents.focus()
+    focused?.focus()
   }
 }
 

apps/desktop/src/main/browser-agent/session.test.ts

@@ -2503,6 +2503,39 @@ describe('browser-agent session', () => {
     expect(content.removeChildView).toHaveBeenCalledWith(second.view)
   })
 
+  it('gives keyboard focus back to the visible page when it parks the agent tab', () => {
+    const visibleTab = session.ensureTab()
+    panel.setPanelBounds({ x: 0, y: 0, width: 800, height: 600 })
+    const visibleContents = visibleTab.view.webContents as unknown as MockView['webContents']
+    visibleContents.isFocused.mockReturnValue(true)
+    vi.mocked(visibleContents.focus).mockClear()
+
+    session.addAutomationTab()
+
+    expect(visibleContents.focus).toHaveBeenCalled()
+  })
+
+  it('parks an agent tab in the main window when a secondary window stops showing it', () => {
+    const otherWindow = mainWindowMock() as unknown as {
+      contentView: {
+        addChildView: ReturnType<typeof vi.fn>
+        removeChildView: ReturnType<typeof vi.fn>
+      }
+    }
+    const agentTab = session.ensureTab()
+    panel.setPanelBounds(
+      { x: 0, y: 0, width: 800, height: 600 },
+      otherWindow as unknown as BrowserWindow
+    )
+    expect(otherWindow.contentView.addChildView).toHaveBeenCalledWith(agentTab.view)
+    vi.mocked(win.contentView.addChildView).mockClear()
+
+    session.addTab()
+
+    expect(otherWindow.contentView.removeChildView).toHaveBeenCalledWith(agentTab.view)
+    expect(win.contentView.addChildView).toHaveBeenCalledWith(agentTab.view)
+  })
+
   it('keeps the agent tab composited while the user views another tab, and releases it on close', () => {
     const agentTab = session.ensureTab()
     panel.setPanelBounds({ x: 0, y: 0, width: 800, height: 600 })