Pup - the AI's own browser

@@ -1642,32 +1642,16 @@ function osBackgroundWindowByPid(pid, cdpPort, restoreHwnd, respectUser) {       // Resolve the ONE real browser window (largest visible Chrome_WidgetWin_1), not a helper.       '$pup = [WinBgP]::BestWindow([int[]]@($pidList.ToArray()))',       'if ($pup -eq [IntPtr]::Zero) { Write-Output "no-window"; exit }',-      // v1.8.44: if the USER deliberately raised this window (clicked its taskbar button to-      // WATCH a page load), do NOT fight them. A physical input in the last ~1.5s WITH the pup-      // window already foreground means the user took it — not Chrome self-raising (which is-      // programmatic, no input). Skip the re-park and signal the caller to stop retrying.-      // respectUser is 0 for the initial off-screen park (user can't have clicked it yet).+      // respectUser (v1.8.49): used ONLY by LATE re-parks (post-resize etc.), never the launch+      // passes. Rationale: after startup, Chrome does NOT self-raise — so if the pup window is+      // foreground later, the USER took it (clicked its taskbar button). Leave it exactly as they+      // have it. NO input-timing heuristics: GetLastInputInfo measures input ANYWHERE on the+      // system, so "recent input" false-fires whenever the user is typing while a window opens —+      // that bug shipped in 1.8.44-48 and mis-shoved fresh windows into a small centered state.       `$respectUser = ${respectUser ? 1 : 0}`,       'if ($respectUser -eq 1) {',       '  $fg0 = [WinBgP]::GetForegroundWindow()',-      '  if ($fg0 -eq $pup) {',-      '    $lii = New-Object LASTINPUTINFO',-      '    $lii.cbSize = [uint32][System.Runtime.InteropServices.Marshal]::SizeOf($lii)',-      '    [WinBgP]::GetLastInputInfo([ref]$lii) | Out-Null',-      '    $idle = [WinBgP]::GetTickCount() - $lii.dwTime',-      '    if ($idle -lt 1500) {',-      // The USER deliberately raised this window (clicked its taskbar button) to SEE it. It may-      // still be sitting at its off-screen LAUNCH coords, so just "leaving it alone" would show-      // them NOTHING. Transition it into a normal, visible, on-screen window (centered ~80% of-      // the monitor), SHOW it, and bring it to the TOP — and DON'T hand focus back, because the-      // user wants to use it. This is the background↔foreground handoff: AI runs it in the bg,-      // the moment the human takes it, it becomes a real usable window.-      '      $vw = [int]($scrW * 0.8); $vh = [int]($scrH * 0.8)',-      '      $vx = [int](($scrW - $vw) / 2); $vy = [int](($scrH - $vh) / 2)',-      '      [WinBgP]::SetWindowPos($pup, [IntPtr]0, $vx, $vy, $vw, $vh, 0x40) | Out-Null', // HWND_TOP, SWP_SHOWWINDOW-      '      Write-Output "user-foreground"; exit',-      '    }',-      '  }',+      '  if ($fg0 -eq $pup) { Write-Output "user-foreground"; exit }',       '}',       // ORDER MATTERS (v1.8.46): the browser grabbed the foreground on launch while it's still       // OFF-SCREEN (invisible). Hand focus back to the user FIRST — deactivating the pup window@@ -2282,31 +2266,24 @@ async function getOrLaunchBrowser(profileName, { freshProfile = false, strictPer       // can re-raise itself during startup (first paint, restore prompts). Resolve the       // window via the CDP-port owner (reliable across launcher→browser pid relaunches).       if (!foreground && process.platform === 'win32') {-        // v1.8.47b: background = ON-SCREEN at z-bottom (occluded, still renders → screenshots-        // work). Minimize was tried and REJECTED — a minimized window reports visibilityState-        // 'hidden' and Page.captureScreenshot hangs (~27s) even with occlusion disabled. The-        // park deactivates FIRST then moves on-screen behind everything (no pop), using the-        // EnumWindows handle (the real window, not a phantom). The re-park window is SHORT-        // (only ~1.2s, just long enough to counter Chrome's startup foreground-grab) so a user-        // click after that is NEVER fought; within it, GetLastInputInfo distinguishes a real-        // click (→ bring on-screen + latch, stop) from Chrome self-raising (→ re-park).+        // v1.8.49: background = ON-SCREEN at z-bottom (occluded, still renders → screenshots+        // work; minimize was tried and REJECTED — visibilityState 'hidden' hangs+        // Page.captureScreenshot ~27s even with occlusion disabled). The park deactivates FIRST+        // then moves on-screen behind everything (no pop), using the EnumWindows handle (the+        // real window, not a phantom). The launch passes ALWAYS park, no user detection: within+        // ~1.2s of launch the taskbar button barely exists, so a foreground pup window is+        // Chrome's startup self-raise, not the user — and input-timing heuristics false-fire+        // whenever the user is typing anywhere (the 1.8.44-48 bug that mis-sized fresh windows).+        // After this window ends, pup NEVER auto-re-parks, so a user click is never fought.         b._bgResult = osBackgroundWindowByPid(child.pid, debugPort, priorForeground);         console.log(`[bg] initial park for "${profileName}": ${b._bgResult}`);-        const _bgTimers = [];         for (const delay of [300, 700, 1200]) {-          const t = setTimeout(() => {+          setTimeout(() => {             try {-              if (b._userForeground) return; // user has it — never re-park-              const r = osBackgroundWindowByPid(child.pid, debugPort, priorForeground, true);+              const r = osBackgroundWindowByPid(child.pid, debugPort, priorForeground);               if (r && r !== 'no-window') b._bgResult = r;-              if (r === 'user-foreground') {-                b._userForeground = true;-                for (const tt of _bgTimers) clearTimeout(tt);-                console.log(`[bg] user raised "${profileName}" — respecting it, halting re-park`);-              }             } catch (e) {}           }, delay);-          _bgTimers.push(t);         }       }       return b;