add keyboard tracking and app-qualified smart titles

**Keyboard hook (Windows)**
- Extends the existing C# WH_MOUSE_LL process to also install
  WH_KEYBOARD_LL alongside it (keyboard hook is optional — failure
  does not break mouse capture).
- Emits CHAR <code> <ts> for printable keystrokes, KEY <name> <ts>
  for modifier combos (Ctrl+T) and special keys (Backspace, Enter).

**Text accumulation in capture session**
- CaptureService tracks _keyBuffer (typed chars since last step) and
  _lastShortcut (last modifier combo) using the new onKeyboardEvent()
  method.
- snapshotKeyContext() is called at enqueueClickCapture time so each
  step's clickMeta.keyContext carries { recentTyped, recentShortcut }.
- Buffer resets after each snapshot; stale input (>8s gap) is dropped.

**UIAutomation element value**
- collectWindowsWindowContext now reads ValuePattern.Current.Value
  from the clicked element — giving us what's actually typed in a
  search box or text field without needing the keyboard buffer.

**Smart title generation (core/text-intel.js)**
- Priority chain: keyboard shortcut → element value → typed text
  → OCR → element label → page title → app name.
- SHORTCUT_TITLES maps 50+ common shortcuts (Ctrl+T, Ctrl+S, F5 …)
  to natural language descriptions: "Open new tab", "Save", etc.
- qualifyTitleWithApp() appends "in Chrome / VS Code / Terminal / …"
  to every title when the app is known: "Click Save in VS Code",
  "Search for oracle in Chrome", "Open new tab in Chrome".
- APP_DISPLAY_NAMES covers browsers, editors, terminals, office apps.

Six new unit tests cover shortcuts, typed-text search, element value,
and app-qualified OCR titles. Capture test updated for keyContext.

Co-Authored-By: Claude Sonnet 4.6 <[email protected]>

app/capture.js

@@ -136,6 +136,9 @@ class CaptureService {
     this.linuxEvent = null; // event block currently being parsed
     this.pendingRawClick = null; // raw press waiting for its coordinate twin
     this.pendingClickOsPoint = null;
+    this._keyBuffer = '';     // printable chars typed since last capture
+    this._lastShortcut = '';  // last Ctrl+X / Alt+X combination
+    this._keyLastAt = 0;      // timestamp of last key event
     this.clickQueue = Promise.resolve();
     this.frameLoopInFlight = false;
     this.frameLoopGrabStartedAt = null;
@@ -799,12 +802,15 @@ using System.Collections.Concurrent;
 using System.Runtime.InteropServices;
 using System.Threading;
 
-public static class SFMouseHook {
+public static class SFHook {
   private const int WH_MOUSE_LL = 14;
+  private const int WH_KEYBOARD_LL = 13;
   private const int WM_LBUTTONDOWN = 0x0201;
   private const int WM_RBUTTONDOWN = 0x0204;
   private const int WM_MBUTTONDOWN = 0x0207;
   private const int WM_XBUTTONDOWN = 0x020B;
+  private const int WM_KEYDOWN = 0x0100;
+  private const int WM_SYSKEYDOWN = 0x0104;
   private const long UnixEpochMilliseconds = 62135596800000L;
   // Opting this process out of Windows Power Throttling (EcoQoS). In a
   // power-saving plan the OS CPU-starves background processes; a starved
@@ -818,7 +824,9 @@ public static class SFMouseHook {
   private const uint HIGH_PRIORITY_CLASS = 0x00000080;
 
   private static IntPtr hook = IntPtr.Zero;
-  private static LowLevelMouseProc proc = HookCallback;
+  private static IntPtr keyHook = IntPtr.Zero;
+  private static LowLevelMouseProc proc = MouseHookCallback;
+  private static LowLevelKeyboardProc keyProc = KeyboardHookCallback;
   private static readonly ConcurrentQueue<string> queue = new ConcurrentQueue<string>();
   private static readonly AutoResetEvent signal = new AutoResetEvent(false);
 
@@ -854,11 +862,34 @@ public static class SFMouseHook {
     public uint StateMask;
   }
 
+  [StructLayout(LayoutKind.Sequential)]
+  private struct KBDLLHOOKSTRUCT {
+    public uint vkCode;
+    public uint scanCode;
+    public uint flags;
+    public uint time;
+    public UIntPtr dwExtraInfo;
+  }
+
   private delegate IntPtr LowLevelMouseProc(int nCode, IntPtr wParam, IntPtr lParam);
+  private delegate IntPtr LowLevelKeyboardProc(int nCode, IntPtr wParam, IntPtr lParam);
 
   [DllImport("user32.dll", SetLastError = true)]
   private static extern IntPtr SetWindowsHookEx(int idHook, LowLevelMouseProc lpfn, IntPtr hMod, uint dwThreadId);
 
+  [DllImport("user32.dll", SetLastError = true)]
+  private static extern IntPtr SetWindowsHookEx(int idHook, LowLevelKeyboardProc lpfn, IntPtr hMod, uint dwThreadId);
+
+  [DllImport("user32.dll")]
+  private static extern short GetKeyState(int nVirtKey);
+
+  [DllImport("user32.dll")]
+  private static extern bool GetKeyboardState([Out] byte[] lpKeyState);
+
+  [DllImport("user32.dll", CharSet = CharSet.Unicode)]
+  private static extern int ToUnicode(uint wVirtKey, uint wScanCode, byte[] lpKeyState,
+      System.Text.StringBuilder pwszBuff, int cchBuff, uint wFlags);
+
   [DllImport("user32.dll", SetLastError = true)]
   private static extern bool UnhookWindowsHookEx(IntPtr hhk);
 
@@ -917,6 +948,8 @@ public static class SFMouseHook {
     if (hook == IntPtr.Zero) {
       throw new System.ComponentModel.Win32Exception(Marshal.GetLastWin32Error());
     }
+    // Keyboard hook is optional — if it fails we still get click titles.
+    try { keyHook = SetWindowsHookEx(WH_KEYBOARD_LL, keyProc, GetModuleHandle(null), 0); } catch { }
 
     Console.Out.WriteLine("READY");
     Console.Out.Flush();
@@ -928,6 +961,7 @@ public static class SFMouseHook {
     }
 
     UnhookWindowsHookEx(hook);
+    if (keyHook != IntPtr.Zero) UnhookWindowsHookEx(keyHook);
   }
 
   private static void WriterLoop() {
@@ -941,7 +975,7 @@ public static class SFMouseHook {
     }
   }
 
-  private static IntPtr HookCallback(int nCode, IntPtr wParam, IntPtr lParam) {
+  private static IntPtr MouseHookCallback(int nCode, IntPtr wParam, IntPtr lParam) {
     if (nCode >= 0) {
       int message = wParam.ToInt32();
       string button = ButtonName(message, lParam);
@@ -955,6 +989,75 @@ public static class SFMouseHook {
     return CallNextHookEx(hook, nCode, wParam, lParam);
   }
 
+  private static IntPtr KeyboardHookCallback(int nCode, IntPtr wParam, IntPtr lParam) {
+    if (nCode >= 0 && (wParam.ToInt32() == WM_KEYDOWN || wParam.ToInt32() == WM_SYSKEYDOWN)) {
+      try {
+        KBDLLHOOKSTRUCT kb = (KBDLLHOOKSTRUCT)Marshal.PtrToStructure(lParam, typeof(KBDLLHOOKSTRUCT));
+        int vk = (int)kb.vkCode;
+        bool ctrl = (GetKeyState(0x11) & 0x8000) != 0;
+        bool alt  = (GetKeyState(0x12) & 0x8000) != 0;
+        bool shift = (GetKeyState(0x10) & 0x8000) != 0;
+        // Skip standalone modifier keys.
+        bool isMod = vk == 0x10 || vk == 0x11 || vk == 0x12 || vk == 0x5B || vk == 0x5C;
+        if (!isMod) {
+          long unixMs = DateTime.UtcNow.Ticks / TimeSpan.TicksPerMillisecond - UnixEpochMilliseconds;
+          if (ctrl || alt) {
+            // Emit shortcut: "KEY Ctrl+T 123456"
+            string name = VkName(vk);
+            if (name != null) {
+              string mods = (ctrl ? "Ctrl" : "") + (ctrl && (alt || shift) ? "+" : "") +
+                            (alt ? "Alt" : "") + ((alt || ctrl) && shift ? "+" : "") +
+                            (shift ? "Shift" : "");
+              queue.Enqueue("KEY " + mods + "+" + name + " " + unixMs);
+              signal.Set();
+            }
+          } else if (vk == 0x08) {
+            queue.Enqueue("KEY Backspace " + unixMs); signal.Set();
+          } else if (vk == 0x1B) {
+            queue.Enqueue("KEY Escape " + unixMs); signal.Set();
+          } else if (vk == 0x0D) {
+            queue.Enqueue("KEY Enter " + unixMs); signal.Set();
+          } else {
+            // Map to Unicode character using current keyboard layout + shift state.
+            byte[] ks = new byte[256];
+            GetKeyboardState(ks);
+            var sb = new System.Text.StringBuilder(4);
+            int res = ToUnicode(kb.vkCode, kb.scanCode, ks, sb, 4, 0);
+            if (res > 0) {
+              char ch = sb[0];
+              if (ch >= 0x20 && ch < 0x7F) {
+                queue.Enqueue("CHAR " + (int)ch + " " + unixMs);
+                signal.Set();
+              }
+            }
+          }
+        }
+      } catch { }
+    }
+    return CallNextHookEx(keyHook, nCode, wParam, lParam);
+  }
+
+  private static string VkName(int vk) {
+    if (vk >= 0x41 && vk <= 0x5A) return ((char)vk).ToString(); // A-Z
+    if (vk >= 0x30 && vk <= 0x39) return ((char)vk).ToString(); // 0-9
+    if (vk >= 0x70 && vk <= 0x87) return "F" + (vk - 0x6F);    // F1-F24
+    switch (vk) {
+      case 0x09: return "Tab";
+      case 0x0D: return "Enter";
+      case 0x1B: return "Escape";
+      case 0x20: return "Space";
+      case 0x21: return "PageUp";    case 0x22: return "PageDown";
+      case 0x23: return "End";       case 0x24: return "Home";
+      case 0x25: return "Left";      case 0x26: return "Up";
+      case 0x27: return "Right";     case 0x28: return "Down";
+      case 0x2E: return "Delete";
+      case 0x6B: case 0xBB: return "Plus";
+      case 0x6D: case 0xBD: return "Minus";
+      case 0xBE: return "Period";    case 0xBF: return "Slash";
+      default: return null;
+    }
+  }
+
   private static string ButtonName(int message, IntPtr lParam) {
     if (message == WM_LBUTTONDOWN) return "left";
     if (message == WM_RBUTTONDOWN) return "right";
@@ -968,7 +1071,7 @@ public static class SFMouseHook {
   }
 }
 '@
-[SFMouseHook]::Run()
+[SFHook]::Run()
 `;
         this.clickWatcher = spawn('powershell.exe', ['-NoProfile', '-NonInteractive', '-ExecutionPolicy', 'Bypass', '-Command', ps], {
           stdio: ['ignore', 'pipe', 'pipe'],
@@ -1103,11 +1206,22 @@ public static class SFMouseHook {
     }
     if (platform === 'win32') {
       for (const line of lines) {
-        const m = /^CLICK(?:\s+(-?\d+)\s+(-?\d+)(?:\s+([A-Za-z0-9_-]+))?(?:\s+(\d+))?)?\s*$/.exec(line.trim());
-        if (m) {
-          const osPoint = m[1] === undefined ? null : { x: Number(m[1]), y: Number(m[2]) };
-          const eventAt = m[4] === undefined ? Date.now() : Number(m[4]);
-          this.onOsClick(Number.isFinite(eventAt) ? eventAt : Date.now(), osPoint, m[3] || 'mouse');
+        const trimmed = line.trim();
+        const clickM = /^CLICK(?:\s+(-?\d+)\s+(-?\d+)(?:\s+([A-Za-z0-9_-]+))?(?:\s+(\d+))?)?\s*$/.exec(trimmed);
+        if (clickM) {
+          const osPoint = clickM[1] === undefined ? null : { x: Number(clickM[1]), y: Number(clickM[2]) };
+          const eventAt = clickM[4] === undefined ? Date.now() : Number(clickM[4]);
+          this.onOsClick(Number.isFinite(eventAt) ? eventAt : Date.now(), osPoint, clickM[3] || 'mouse');
+          continue;
+        }
+        const keyM = /^KEY\s+(\S+)\s+(\d+)\s*$/.exec(trimmed);
+        if (keyM) {
+          this.onKeyboardEvent('KEY', keyM[1], Number(keyM[2]));
+          continue;
+        }
+        const charM = /^CHAR\s+(\d+)\s+(\d+)\s*$/.exec(trimmed);
+        if (charM) {
+          this.onKeyboardEvent('CHAR', Number(charM[1]), Number(charM[2]));
         }
       }
     }
@@ -1285,6 +1399,8 @@ public static class SFMouseHook {
     if (osPoint) {
       clickMeta.osPoint = { x: osPoint.x, y: osPoint.y };
     }
+    // Snapshot keyboard context accumulated since the last capture.
+    clickMeta.keyContext = this.snapshotKeyContext();
     if (this.session && !this.session.paused && !this.userIsInApp()) {
       // The guide id pins the click to its recording so it can still be
       // stored if the session stops while this click waits in the queue.
@@ -1298,6 +1414,46 @@ public static class SFMouseHook {
     return this.clickQueue;
   }
 
+  // --- Keyboard context tracking -------------------------------------------
+
+  onKeyboardEvent(type, data, eventAt) {
+    const now = Number.isFinite(eventAt) ? eventAt : Date.now();
+    // Discard stale typing that happened more than 8 seconds ago (user moved on).
+    if (now - this._keyLastAt > 8000) {
+      this._keyBuffer = '';
+    }
+    this._keyLastAt = now;
+
+    if (type === 'CHAR') {
+      const ch = typeof data === 'number' ? String.fromCharCode(data) : String(data);
+      this._keyBuffer = (this._keyBuffer + ch).slice(-200);
+    } else if (type === 'KEY') {
+      const key = String(data);
+      if (key === 'Backspace') {
+        this._keyBuffer = this._keyBuffer.slice(0, -1);
+      } else if (key === 'Escape') {
+        this._keyBuffer = '';
+      } else if (key === 'Enter') {
+        // Keep typed text — Enter often submits what was typed.
+      } else {
+        // It's a modifier+key shortcut (Ctrl+T etc.).
+        this._lastShortcut = key;
+        this._keyBuffer = ''; // a shortcut resets the typed-text buffer
+      }
+    }
+  }
+
+  snapshotKeyContext() {
+    const ctx = {
+      recentTyped: this._keyBuffer.trim(),
+      recentShortcut: this._lastShortcut,
+    };
+    // Reset after snapshot so each step gets its own context.
+    this._keyBuffer = '';
+    this._lastShortcut = '';
+    return ctx;
+  }
+
   async captureCurrentFrame(mode, capturePoint = null, startedAt = Date.now()) {
     const grabbed = await this.grab(mode, capturePoint);
     return {

app/text-intel.js

@@ -181,6 +181,7 @@ class TextIntelService {
       $elementRole = '';
       $elementClass = '';
       $elementProcessId = 0;
+      $elementValue = '';
       if (${hasPoint ? '$true' : '$false'}) {
         try {
           Add-Type -AssemblyName UIAutomationClient,UIAutomationTypes,WindowsBase | Out-Null
@@ -192,6 +193,12 @@ class TextIntelService {
             $elementRole = $current.LocalizedControlType;
             $elementClass = $current.ClassName;
             $elementProcessId = $current.ProcessId;
+            try {
+              $valPattern = [System.Windows.Automation.ValuePattern]::Pattern;
+              if ($element.GetSupportedPatterns() -contains $valPattern) {
+                $elementValue = $element.GetCurrentPattern($valPattern).Current.Value;
+              }
+            } catch { }
           }
         } catch { }
       }
@@ -218,6 +225,7 @@ public static class Win32 {
         elementLabel = $elementLabel;
         elementRole = $elementRole;
         elementClass = $elementClass;
+        elementValue = $elementValue;
         elementProcessId = $elementProcessId;
         pid = $pid;
       };
@@ -284,11 +292,14 @@ public static class Win32 {
   }
 
   async buildCaptureContext({ mode, frame, clickPos, clickMeta = null }) {
+    const keyContext = clickMeta?.keyContext || {};
+    const recentTyped = keyContext.recentTyped || '';
+    const recentShortcut = keyContext.recentShortcut || '';
     const [metadata, ocr] = await Promise.all([
       this.collectForegroundWindowContext(clickMeta?.osPoint || null),
       this.ocrAroundClick(frame, clickPos),
     ]);
-    const title = buildCaptureTitle({ mode, metadata, ocrText: ocr.text });
+    const title = buildCaptureTitle({ mode, metadata, ocrText: ocr.text, recentTyped, recentShortcut });
     return {
       title,
       captureMetadata: {
@@ -297,6 +308,9 @@ public static class Win32 {
         appName: metadata.appName || '',
         elementLabel: metadata.elementLabel || '',
         elementRole: metadata.elementRole || '',
+        elementValue: metadata.elementValue || '',
+        recentTyped,
+        recentShortcut,
         mode,
       },
     };
@@ -440,8 +454,11 @@ public static class Win32 {
             appName: step.captureMetadata.appName,
             elementLabel: step.captureMetadata.elementLabel,
             elementRole: step.captureMetadata.elementRole,
+            elementValue: step.captureMetadata.elementValue,
           },
           ocrText: step.captureMetadata.ocrText,
+          recentTyped: step.captureMetadata.recentTyped,
+          recentShortcut: step.captureMetadata.recentShortcut,
         });
         captureContext = {
           ...step.captureMetadata,

core/text-intel.js

@@ -70,6 +70,121 @@ const SEARCH_ENGINE_PAGE_NAMES = new Set([
   'brave search',
 ]);
 
+// Common keyboard shortcuts → short action descriptions used as step titles.
+const SHORTCUT_TITLES = {
+  'Ctrl+T':          'Open new tab',
+  'Ctrl+N':          'Open new window',
+  'Ctrl+W':          'Close tab',
+  'Ctrl+Shift+T':    'Reopen closed tab',
+  'Ctrl+Shift+N':    'Open incognito window',
+  'Ctrl+S':          'Save',
+  'Ctrl+Shift+S':    'Save as',
+  'Ctrl+Z':          'Undo',
+  'Ctrl+Y':          'Redo',
+  'Ctrl+Shift+Z':    'Redo',
+  'Ctrl+C':          'Copy selection',
+  'Ctrl+V':          'Paste',
+  'Ctrl+X':          'Cut selection',
+  'Ctrl+A':          'Select all',
+  'Ctrl+F':          'Open Find',
+  'Ctrl+H':          'Open Find and Replace',
+  'Ctrl+R':          'Reload page',
+  'Ctrl+Shift+R':    'Hard reload page',
+  'Ctrl+L':          'Focus address bar',
+  'Ctrl+D':          'Bookmark page',
+  'Ctrl+Tab':        'Switch to next tab',
+  'Ctrl+Shift+Tab':  'Switch to previous tab',
+  'Ctrl+Plus':       'Zoom in',
+  'Ctrl+Minus':      'Zoom out',
+  'Ctrl+0':          'Reset zoom',
+  'Ctrl+P':          'Print',
+  'Ctrl+O':          'Open file',
+  'Ctrl+E':          'Focus search bar',
+  'Ctrl+K':          'Focus search bar',
+  'Ctrl+G':          'Go to line',
+  'Ctrl+B':          'Toggle sidebar',
+  'Ctrl+Shift+P':    'Open command palette',
+  'Ctrl+Shift+E':    'Show file explorer',
+  'Ctrl+Shift+G':    'Show source control',
+  'Ctrl+Shift+D':    'Show debug panel',
+  'Ctrl+Shift+X':    'Show extensions',
+  'Alt+F4':          'Close window',
+  'Alt+Left':        'Go back',
+  'Alt+Right':       'Go forward',
+  'Alt+Tab':         'Switch application',
+  'F2':              'Rename',
+  'F3':              'Find next',
+  'F4':              'Open address bar',
+  'F5':              'Reload page',
+  'F11':             'Toggle fullscreen',
+  'F12':             'Open developer tools',
+};
+
+// Process name → human-readable display name (used to append "in Chrome" etc. to titles).
+const APP_DISPLAY_NAMES = {
+  chrome:           'Chrome',
+  msedge:           'Edge',
+  firefox:          'Firefox',
+  safari:           'Safari',
+  opera:            'Opera',
+  brave:            'Brave',
+  vivaldi:          'Vivaldi',
+  code:             'VS Code',
+  cursor:           'Cursor',
+  'sublime_text':   'Sublime Text',
+  atom:             'Atom',
+  notepad:          'Notepad',
+  'notepad++':      'Notepad++',
+  winword:          'Word',
+  excel:            'Excel',
+  powerpnt:         'PowerPoint',
+  outlook:          'Outlook',
+  teams:            'Teams',
+  slack:            'Slack',
+  discord:          'Discord',
+  zoom:             'Zoom',
+  figma:            'Figma',
+  postman:          'Postman',
+  insomnia:         'Insomnia',
+  notion:           'Notion',
+  obsidian:         'Obsidian',
+  spotify:          'Spotify',
+  terminal:         'Terminal',
+  cmd:              'Command Prompt',
+  powershell:       'PowerShell',
+  windowsterminal:  'Windows Terminal',
+  wt:               'Windows Terminal',
+  iterm2:           'iTerm',
+  wezterm:          'WezTerm',
+  alacritty:        'Alacritty',
+  kitty:            'Kitty',
+  'gnome-terminal': 'Terminal',
+  konsole:          'Konsole',
+  xterm:            'Terminal',
+  xfce4terminal:    'Terminal',
+  bash:             'Terminal',
+  zsh:              'Terminal',
+  fish:             'Terminal',
+  finder:           'Finder',
+  explorer:         'File Explorer',
+  'files-uwp':      'File Explorer',
+  steam:            'Steam',
+  'steamwebhelper': 'Steam',
+};
+
+function cleanAppName(rawName) {
+  if (!rawName) return '';
+  const key = normalizeWhitespace(rawName).toLowerCase().replace(/\.exe$/i, '');
+  return APP_DISPLAY_NAMES[key] || sentenceCase(rawName.replace(/\.exe$/i, ''));
+}
+
+function qualifyTitleWithApp(title, appName) {
+  const app = cleanAppName(appName);
+  if (!app) return title;
+  if (new RegExp(`\\b${app.replace(/[.*+?^${}()|[\]\\]/g, '\\$&')}\\b`, 'i').test(title)) return title;
+  return `${title} in ${app}`;
+}
+
 const ACTION_PREFIXES = [
   'click',
   'select',
@@ -310,27 +425,70 @@ function pickBestTitleFragment(text, { source = 'window', metadata = {} } = {})
   return best ? formatCaptureTitle(best, { source, metadata }) : '';
 }
 
-function buildCaptureTitle({ mode = 'fullscreen', metadata = {}, ocrText = '' } = {}) {
+function buildCaptureTitle({ mode = 'fullscreen', metadata = {}, ocrText = '', recentTyped = '', recentShortcut = '' } = {}) {
+  const app = cleanAppName(metadata.appName);
+
+  // 1. Keyboard shortcut → most reliable signal for "what action did the user take".
+  if (recentShortcut && SHORTCUT_TITLES[recentShortcut]) {
+    const base = SHORTCUT_TITLES[recentShortcut];
+    return app ? qualifyTitleWithApp(base, metadata.appName) : base;
+  }
+
+  // 2. UIAutomation element value — what's actually typed inside the clicked field.
+  const elementValue = normalizeWhitespace(metadata.elementValue || '');
+  if (elementValue) {
+    const roleLower = normalizeWhitespace(metadata.elementRole || '').toLowerCase();
+    const labelLower = normalizeWhitespace(metadata.elementLabel || '').toLowerCase();
+    const looksLikeSearch = /(search|find|query|omnibox|address bar)/.test(roleLower + ' ' + labelLower);
+    const action = looksLikeSearch ? 'Search for' : 'Type';
+    const base = `${action} "${elementValue}"`;
+    return app ? qualifyTitleWithApp(base, metadata.appName) : base;
+  }
+
+  // 3. Keyboard-buffer text (typed between captures) + input role context.
+  const typed = normalizeWhitespace(recentTyped || '');
+  if (typed) {
+    const roleLower = normalizeWhitespace(metadata.elementRole || '').toLowerCase();
+    const labelLower = normalizeWhitespace(metadata.elementLabel || '').toLowerCase();
+    const isSearchRole = /(search box|searchbox|search field|search bar|search input)/.test(roleLower);
+    const looksLikeSearch = isSearchRole || /(search|find|query|omnibox|address bar)/.test(roleLower + ' ' + labelLower);
+    const isAnyInput = /(text field|edit|input|field|combo box|textbox|text box)/.test(roleLower);
+    if (looksLikeSearch) {
+      const base = `Search for "${typed}"`;
+      return app ? qualifyTitleWithApp(base, metadata.appName) : base;
+    }
+    if (isAnyInput) {
+      const base = `Type "${typed}"`;
+      return app ? qualifyTitleWithApp(base, metadata.appName) : base;
+    }
+  }
+
+  // 4. OCR text around the click — link text, button labels, menu items.
   const ocrPhrase = pickBestOcrPhrase(ocrText);
-  if (ocrPhrase) return formatCaptureTitle(ocrPhrase, { source: 'ocr', metadata });
+  if (ocrPhrase) {
+    const title = formatCaptureTitle(ocrPhrase, { source: 'ocr', metadata });
+    return app ? qualifyTitleWithApp(title, metadata.appName) : title;
+  }
 
+  // 5. UIAutomation element label.
   const elementPhrase = pickBestTitleFragment(metadata.elementLabel, { source: 'element', metadata });
-  if (elementPhrase) return elementPhrase;
+  if (elementPhrase) {
+    return app ? qualifyTitleWithApp(elementPhrase, metadata.appName) : elementPhrase;
+  }
 
-  // Strip the browser name suffix before processing window titles so that
-  // "oracle - Google Search - Google Chrome" becomes "oracle - Google Search"
-  // and "Oracle | Cloud Applications - Google Chrome" becomes just the page title.
-  const rawWindowTitle = metadata.windowTitle || '';
-  const strippedWindowTitle = stripBrowserNameSuffix(rawWindowTitle);
+  // 6. Browser window title (suffix stripped) → page title or search query.
+  const strippedWindowTitle = stripBrowserNameSuffix(metadata.windowTitle || '');
   if (strippedWindowTitle) {
-    // Detect "[query] - Google Search" → "Search for oracle"
     const searchQuery = extractSearchQuery(strippedWindowTitle);
-    if (searchQuery) return `Search for ${sentenceCase(searchQuery)}`;
-    // Use the page title (now free of browser name noise).
+    if (searchQuery) {
+      const base = `Search for ${sentenceCase(searchQuery)}`;
+      return app ? qualifyTitleWithApp(base, metadata.appName) : base;
+    }
     const windowPhrase = pickBestTitleFragment(strippedWindowTitle, { source: 'window', metadata });
     if (windowPhrase) return windowPhrase;
   }
 
+  // 7. App name alone as last resort.
   const appPhrase = pickBestTitleFragment(metadata.appName, { source: 'app', metadata });
   if (appPhrase) return appPhrase;
 
@@ -516,6 +674,9 @@ function buildAiPrompt({
     captureContext.windowTitle ? `Active window: ${captureContext.windowTitle}` : null,
     captureContext.appName ? `App: ${captureContext.appName}` : null,
     captureContext.elementLabel ? `UI element: ${captureContext.elementLabel}${captureContext.elementRole ? ` (${captureContext.elementRole})` : ''}` : null,
+    captureContext.elementValue ? `Element content (what was typed): ${captureContext.elementValue}` : null,
+    captureContext.recentTyped ? `Keyboard input before this step: ${captureContext.recentTyped}` : null,
+    captureContext.recentShortcut ? `Keyboard shortcut used: ${captureContext.recentShortcut}` : null,
     captureContext.ocrText ? `OCR text near click:\n${captureContext.ocrText}` : null,
     captureContext.titleCandidate ? `Suggested title: ${captureContext.titleCandidate}` : null,
   ].filter(Boolean) : [];

tests/unit/capture.test.js

@@ -375,7 +375,7 @@ test('queued click captures preserve the original event time and button', async
   assert.deepEqual(seen, [{
     trigger: 'click',
     clickPos: { x: 7, y: 8 },
-    clickMeta: { at: 1770000000456, button: 'left' },
+    clickMeta: { at: 1770000000456, button: 'left', keyContext: { recentTyped: '', recentShortcut: '' } },
   }]);
 });
 

tests/unit/text-intel.test.js

@@ -61,7 +61,8 @@ test('capture titles ignore browser chrome noise in favor of OCR', () => {
     },
     ocrText: 'New tab',
   });
-  assert.equal(title, 'Click New tab');
+  // OCR wins over the noisy window title; app name is appended for context.
+  assert.equal(title, 'Click New tab in Google Chrome');
 });
 
 test('tab-like roles use select when OCR identifies a tab label', () => {
@@ -110,7 +111,7 @@ test('search query is extracted from browser window title pattern', () => {
     },
     ocrText: '',
   });
-  assert.equal(title, 'Search for Oracle');
+  assert.equal(title, 'Search for Oracle in Chrome');
 });
 
 test('full link text with pipe separator is preserved in OCR phrases', () => {
@@ -140,6 +141,56 @@ test('search box element role uses Search for verb', () => {
   assert.equal(title, 'Search for Oracle');
 });
 
+test('keyboard shortcut produces action title qualified with app', () => {
+  const title = buildCaptureTitle({
+    mode: 'fullscreen',
+    metadata: { appName: 'chrome' },
+    ocrText: '',
+    recentShortcut: 'Ctrl+T',
+  });
+  assert.equal(title, 'Open new tab in Chrome');
+});
+
+test('keyboard shortcut title without app name', () => {
+  const title = buildCaptureTitle({
+    mode: 'fullscreen',
+    metadata: {},
+    ocrText: '',
+    recentShortcut: 'Ctrl+S',
+  });
+  assert.equal(title, 'Save');
+});
+
+test('typed text with search input role produces Search for title', () => {
+  const title = buildCaptureTitle({
+    mode: 'fullscreen',
+    metadata: { elementRole: 'search box', appName: 'chrome' },
+    ocrText: '',
+    recentTyped: 'oracle',
+  });
+  assert.equal(title, 'Search for "oracle" in Chrome');
+});
+
+test('UIAutomation element value takes priority over keyboard buffer', () => {
+  const title = buildCaptureTitle({
+    mode: 'fullscreen',
+    metadata: { elementRole: 'edit', elementValue: 'oracle', appName: 'chrome' },
+    ocrText: '',
+    recentTyped: 'ignored',
+  });
+  // elementValue (from UIAutomation) wins over the keyboard buffer
+  assert.ok(title.includes('oracle'), `expected oracle in title, got: ${title}`);
+});
+
+test('app-qualified OCR title includes app name', () => {
+  const title = buildCaptureTitle({
+    mode: 'fullscreen',
+    metadata: { appName: 'code' },
+    ocrText: 'Save',
+  });
+  assert.equal(title, 'Click Save in VS Code');
+});
+
 test('ai prompts include the deterministic OCR-backed title candidate', () => {
   const { prompt } = buildAiPrompt({
     captureContext: {