fix(browser): let the newest input or extension gesture decide who set an extension call off

src/main/browser/extensions/permissions/browser-extension-user-gesture.ts

@@ -22,19 +22,21 @@ export function noteBrowserExtensionUserGesture(
   recentGestures.set(signal, Date.now())
 }
 
-export function hasBrowserExtensionUserGesture(caller: BrowserExtensionCaller): boolean {
-  if (caller.userGesture === true) {
-    return true
-  }
+/** When Orca last handed the caller's extension a gesture, if still within the gesture window. */
+export function getBrowserExtensionUserGestureAt(caller: BrowserExtensionCaller): number | null {
   const state = getHostSessionState(caller.session)
   if (!state) {
-    return false
+    return null
   }
   const signal = getBrowserExtensionContextSignal(state, caller.extension.id)
   const at = recentGestures.get(signal)
-  return (
-    at !== undefined &&
+  return at !== undefined &&
     !signal.aborted &&
     Date.now() - at <= BROWSER_EXTENSION_USER_GESTURE_WINDOW_MS
-  )
+    ? at
+    : null
+}
+
+export function hasBrowserExtensionUserGesture(caller: BrowserExtensionCaller): boolean {
+  return caller.userGesture === true || getBrowserExtensionUserGestureAt(caller) !== null
 }

src/main/browser/extensions/tabs-and-windows/browser-extension-agent-caller.ts

@@ -3,7 +3,7 @@ import { getLastBrowserGuestInputInSession } from '../../guest-lifecycle/input/b
 import type { BrowserExtensionCaller } from '../api-runtime/browser-extension-namespace-registry'
 import {
   BROWSER_EXTENSION_USER_GESTURE_WINDOW_MS,
-  hasBrowserExtensionUserGesture
+  getBrowserExtensionUserGestureAt
 } from '../permissions/browser-extension-user-gesture'
 import type { openBrowserExtensionTab } from './browser-extension-tab-model'
 
@@ -19,24 +19,28 @@ function pageOfGuest(guestWebContentsId: number): AgentPage {
 /**
  * The agent's page this call acts for, or null when the user set it off. An extension reacts to
  * its profile's pages (a wallet popup, an OAuth tab) without saying which input started it, so
- * the latest click or key in the profile decides, then whether an agent is driving a page there.
+ * the newest recent click, key or Orca-delivered gesture decides, then whether an agent leads.
  */
 export function findBrowserExtensionCallerAgentPage(
   caller: BrowserExtensionCaller
 ): AgentPage | null {
-  // Why first: an agent's click on an extension page also counts as that page's user gesture.
-  if (caller.webContents && browserManager.isAgentLeadingGuest(caller.webContents.id)) {
-    return pageOfGuest(caller.webContents.id)
-  }
-  if (hasBrowserExtensionUserGesture(caller)) {
+  const now = Date.now()
+  const lastInput = getLastBrowserGuestInputInSession(caller.session)
+  const input =
+    lastInput && now - lastInput.at <= BROWSER_EXTENSION_USER_GESTURE_WINDOW_MS ? lastInput : null
+  const gestureAt = getBrowserExtensionUserGestureAt(caller)
+  if (gestureAt !== null && (!input || gestureAt >= input.at)) {
     return null
   }
-  const input = getLastBrowserGuestInputInSession(caller.session)
-  if (input && Date.now() - input.at <= BROWSER_EXTENSION_USER_GESTURE_WINDOW_MS) {
+  if (input) {
     // Agents and remote viewers send injected input; only the desktop user's own may surface.
     return input.byAgent ? pageOfGuest(input.guestWebContentsId) : null
   }
-  return browserManager.findAgentPageInSession(caller.session)
+  // Why before userGesture: an agent's click on an extension page also activates that page.
+  if (caller.webContents && browserManager.isAgentLeadingGuest(caller.webContents.id)) {
+    return pageOfGuest(caller.webContents.id)
+  }
+  return caller.userGesture ? null : browserManager.findAgentPageInSession(caller.session)
 }
 
 /** An agent's extension tab opens in the background beside its page, never in the user's view. */