fix(sync): restore UI state and task sync after reconnect or sleep (#107)

public/index.html

@@ -7382,10 +7382,10 @@ <h2 id="dlgModalTitle" data-i18n="dlg.title">Delegate to Agent</h2>
     loadFiles();
     loadHist();
     loadActivity();
-    // On page refresh: currentSessionId is null (restoreUIState doesn't call loadSess),
-    // but activeTabId may point to a real session — load it now so history appears.
-    if (!currentSessionId && activeTabId && openTabs.find(t => t.id === activeTabId && !t.isNew)) {
-      loadSess(activeTabId); // loadSess sets currentSessionId and sends start_session at the end
+    // On page refresh or reconnect after sleep/drop:
+    // loadSess syncs any tasks or final messages that finished while away (#107)
+    if (activeTabId && openTabs.find(t => t.id === activeTabId && !t.isNew)) {
+      loadSess(activeTabId);
     } else if (currentSessionId && ws.readyState === 1) {
       // Re-attach to active session after a reconnect
       ws.send(JSON.stringify({ type: 'start_session', sessionId: currentSessionId, mode: curMode, agentMode: curAgent, model: curModel}));
@@ -7420,6 +7420,12 @@ <h2 id="dlgModalTitle" data-i18n="dlg.title">Delegate to Agent</h2>
 
 // Stop reconnect loop on page unload to avoid pointless connections
 window.addEventListener('beforeunload', () => { _intentionalClose = true; if (_reconnectTimer) clearTimeout(_reconnectTimer); });
+window.addEventListener('online', () => {
+  if (!ws || ws.readyState !== 1) {
+    if (_reconnectTimer) { clearTimeout(_reconnectTimer); _reconnectTimer = null; }
+    if (!_intentionalClose) connect();
+  }
+});
 
 // Self-heal on refocus. While a tab/window is hidden, the browser throttles or
 // suspends it: a queued task that dequeues and starts streaming in the background
@@ -7447,8 +7453,16 @@ <h2 id="dlgModalTitle" data-i18n="dlg.title">Delegate to Agent</h2>
     if (!_intentionalClose) connect();
     return;
   }
-  // Socket alive — replay any buffered stream we may have missed while throttled.
-  if (currentSessionId) ws.send(JSON.stringify({ type: 'subscribe_session', sessionId: currentSessionId }));
+  // Socket alive — replay any buffered stream we may have missed while throttled,
+  // and resync session state if a background task completed while hidden (#107).
+  if (currentSessionId) {
+    ws.send(JSON.stringify({ type: 'subscribe_session', sessionId: currentSessionId }));
+    fetch(`/api/sessions/${currentSessionId}`).then(r => r.json()).then(d => {
+      if (!d.isChatRunning && !d.hasRunningTask && openTabs.find(t => t.id === currentSessionId)?.generating) {
+        loadSess(currentSessionId);
+      }
+    }).catch(() => {});
+  }
   subscribeAllTabs();
   // Same recovery for the visible terminal tab, if any — its own WebSocket is
   // independent of the chat socket above and dies the same way while backgrounded.

public/kanban.html

@@ -1071,6 +1071,22 @@
   refreshTimer=setTimeout(async()=>{await refresh(false);startAutoRefresh();},delay);
 }
 
+// Self-heal on tab focus, visibility change, and network reconnection (#107)
+document.addEventListener('visibilitychange', () => {
+  if (document.visibilityState === 'visible') {
+    refresh(true);
+    startAutoRefresh();
+  }
+});
+window.addEventListener('focus', () => {
+  refresh(true);
+  startAutoRefresh();
+});
+window.addEventListener('online', () => {
+  refresh(true);
+  startAutoRefresh();
+});
+
 // ─── Stats ─────────────────────────────────────────────────────────────────
 function updateStats(){
   const el=$i('kbStats');

server.js

@@ -445,6 +445,18 @@ function killByPid(pid) {
   } catch {} // Process may already be dead (ESRCH)
 }
 
+// Test whether a process with the given PID is currently running.
+function isPidAlive(pid) {
+  const n = Number(pid);
+  if (!Number.isInteger(n) || n <= 0) return false;
+  try {
+    process.kill(n, 0);
+    return true;
+  } catch (e) {
+    return e.code === 'EPERM'; // Process exists but owned by different user
+  }
+}
+
 [WORKDIR, SKILLS_DIR].forEach(d => {
   if (!fs.existsSync(d)) fs.mkdirSync(d, { recursive: true });
 });
@@ -2869,11 +2881,18 @@ setTimeout(() => {
 
 // Watchdog: detect tasks stuck in 'in_progress' with no live worker process.
 // Runs every 60s. If a task is in_progress in DB but not in taskRunning (memory),
-// the worker died without cleanup — recover the task.
+// or if its worker_pid died, recover the task.
 setInterval(() => {
   const inProg = stmts.getInProgressTasks.all();
   for (const task of inProg) {
-    if (taskRunning.has(task.id)) continue; // worker is alive
+    if (taskRunning.has(task.id)) {
+      if (task.worker_pid && !isPidAlive(task.worker_pid)) {
+        log.warn(`[watchdog] task "${task.title}" (${task.id}) worker_pid ${task.worker_pid} died, evicting from taskRunning`);
+        taskRunning.delete(task.id);
+      } else {
+        continue; // worker is alive
+      }
+    }
     // Worker is dead — recover
     log.warn(`[watchdog] task "${task.title}" (${task.id}) stuck in_progress with no live worker, recovering`);
     if (task.worker_pid) killByPid(task.worker_pid);

test/interrupted-recovery.test.js

@@ -356,5 +356,35 @@ check('setLastUserMsg is armed from processChat', /stmts\.setLastUserMsg\.run\(u
 try { db.close(); } catch {}
 for (const suffix of ['', '-wal', '-shm']) { try { fs.unlinkSync(dbPath + suffix); } catch {} }
 
+// ─── 6. Watchdog and Reconnect State Synchronization (#107) ─────────────────
+console.log('\nwatchdog and reconnect state synchronization (#107):');
+{
+  const srvPath = path.join(__dirname, '..', 'server.js');
+  const srvSrc = fs.readFileSync(srvPath, 'utf8');
+  check('server.js defines isPidAlive helper', /function isPidAlive\(pid\)/.test(srvSrc), true);
+  check('watchdog checks isPidAlive before treating in_progress task as alive',
+    /taskRunning\.has\(task\.id\)[\s\S]{0,200}isPidAlive\(task\.worker_pid\)/.test(srvSrc), true);
+  check('watchdog evicts dead worker_pid from taskRunning',
+    /taskRunning\.delete\(task\.id\)/.test(srvSrc), true);
+
+  const kbPath = path.join(__dirname, '..', 'public', 'kanban.html');
+  const kbSrc = fs.readFileSync(kbPath, 'utf8');
+  check('kanban.html listens to visibilitychange for self-healing after sleep',
+    /document\.addEventListener\('visibilitychange'/.test(kbSrc), true);
+  check('kanban.html listens to window focus',
+    /window\.addEventListener\('focus'/.test(kbSrc), true);
+  check('kanban.html listens to window online',
+    /window\.addEventListener\('online'/.test(kbSrc), true);
+
+  const idxPath = path.join(__dirname, '..', 'public', 'index.html');
+  const idxSrc = fs.readFileSync(idxPath, 'utf8');
+  check('index.html calls loadSess on activeTabId on reconnect',
+    /loadSess\(activeTabId\)/.test(idxSrc), true);
+  check('index.html listens to window online for reconnect',
+    /window\.addEventListener\('online'/.test(idxSrc), true);
+  check('index.html resyncs session if task finished while hidden',
+    /!d\.isChatRunning && !d\.hasRunningTask[\s\S]{0,100}loadSess\(currentSessionId\)/.test(idxSrc), true);
+}
+
 console.log(`\n${pass} passed, ${fail} failed`);
 process.exit(fail ? 1 : 0);