squadrant 0.19.3 → 0.19.5

package/dist/squadrantd.js

@@ -887,6 +887,11 @@ function reduce(rec, ev, now) {
       return stampAttempt(base, { resumeRef: ev.resumeRef }, now);
     case "task.turn.started":
       return { ...stampAttempt(base, {}, now), state: "working", pendingTool: void 0, pendingMonitor: void 0 };
+    case "task.turn.failed":
+    // #763: a crew's turn died on an API error (StopFailure hook). Anti-#2576:
+    // NOT task.failed — a hook may never terminalize a task. Falls through to
+    // the exact same turn-boundary transition as task.turn.completed below;
+    // only the notify message (reduce.ts formatMessage) differs.
     case "task.turn.completed":
       if (isStickyAttention(rec.state))
         return { ...rec, lastHeartbeat: now, lastEvent: ev.type };
@@ -1002,6 +1007,9 @@ function formatMessage(rec, event) {
       return `CREW STALLED ${tag}: no heartbeat in ${rec.heartbeatBudgetMs}ms`;
     }
     case "awaiting-input":
+      if (event?.type === "task.turn.failed") {
+        return `CREW TURN FAILED ${tag}: API error ended the turn (not a stall) \u2014 ${event.error}`;
+      }
       return `CREW IDLE ${tag}: turn ended, awaiting your reply.`;
     default:
       return null;
@@ -1411,6 +1419,7 @@ var init_reduce = __esm({
       "task.session",
       "task.turn.started",
       "task.turn.completed",
+      "task.turn.failed",
       "task.delta",
       "task.input.requested",
       "task.approval.requested",
@@ -2322,6 +2331,26 @@ function parseProgramArgs(plistXml) {
     return null;
   return { nodeBin: xmlUnescape(m[1]), daemonEntry: xmlUnescape(m[2]) };
 }
+function isVersionUpgrade(registeredEntry, thisEntry) {
+  const a = registeredEntry.split("/");
+  const b = thisEntry.split("/");
+  if (a.length !== b.length)
+    return false;
+  let diffIndex = -1;
+  for (let i = 0; i < a.length; i++) {
+    if (a[i] !== b[i]) {
+      if (diffIndex !== -1)
+        return false;
+      diffIndex = i;
+    }
+  }
+  if (diffIndex === -1)
+    return false;
+  const versionedSegment = /^(.+)@([^@]+)$/;
+  const ma = a[diffIndex].match(versionedSegment);
+  const mb = b[diffIndex].match(versionedSegment);
+  return ma !== null && mb !== null && ma[1] === mb[1] && ma[2] !== mb[2];
+}
 function detectForeignInstall(parsed, thisEntry, registeredEntryExists) {
   if (!parsed)
     return null;
@@ -2329,7 +2358,11 @@ function detectForeignInstall(parsed, thisEntry, registeredEntryExists) {
     return null;
   if (!registeredEntryExists)
     return null;
-  return { registeredEntry: parsed.daemonEntry, thisEntry };
+  return {
+    registeredEntry: parsed.daemonEntry,
+    thisEntry,
+    isUpgrade: isVersionUpgrade(parsed.daemonEntry, thisEntry)
+  };
 }
 function sanitizePathForPlist(path21) {
   const seen = /* @__PURE__ */ new Set();
@@ -2496,7 +2529,7 @@ function forceKickstartAndVerify(target, opts = {}) {
   const pollDelayMs = opts.pollDelayMs ?? 300;
   const kickstartRetries = opts.kickstartRetries ?? 10;
   const kickstartRetryDelayMs = opts.kickstartRetryDelayMs ?? 300;
-  const pidBefore = getDaemonPid(target);
+  const pidBefore = opts.pidBefore !== void 0 ? opts.pidBefore : getDaemonPid(target);
   let kickstartError = null;
   for (let i = 0; i < kickstartRetries; i++) {
     try {
@@ -2533,6 +2566,9 @@ function forceKickstartAndVerify(target, opts = {}) {
 function isOperatorInitiatedCommand(topLevelArg) {
   return topLevelArg !== void 0 && OPERATOR_INITIATED_COMMANDS.has(topLevelArg);
 }
+function isReadOnlyCrewCommand(argv) {
+  return argv[2] === "crew" && argv[3] !== void 0 && READ_ONLY_CREW_SUBCOMMANDS.has(argv[3]);
+}
 function ensureDaemon(nodeBin = process.execPath, opts = {}) {
   if (restartInFlight)
     return;
@@ -2553,8 +2589,10 @@ function ensureDaemon(nodeBin = process.execPath, opts = {}) {
   try {
     const drift = computeDaemonDrift(nodeBin);
     if (drift.foreignInstall) {
-      process.stderr.write(printForeignInstallError(drift.foreignInstall));
-      return;
+      const decision = decideForeignInstall(drift.foreignInstall);
+      process.stderr.write(decision.message);
+      if (!decision.proceed)
+        return;
     }
     applyDaemonDrift(drift);
   } catch (e) {
@@ -2564,6 +2602,12 @@ function ensureDaemon(nodeBin = process.execPath, opts = {}) {
     releaseDaemonLock();
   }
 }
+function decideForeignInstall(foreign) {
+  if (foreign.isUpgrade) {
+    return { proceed: true, message: printVersionUpgradeNotice(foreign) };
+  }
+  return { proceed: false, message: printForeignInstallError(foreign) };
+}
 function printForeignInstallError(foreign) {
   return `[squadrant] refusing to restart the daemon: the registered launchd config belongs to a DIFFERENT squadrant install than this one.
   registered install: ${foreign.registeredEntry}
@@ -2571,14 +2615,19 @@ function printForeignInstallError(foreign) {
   Two squadrant installs on this machine will keep fighting over the daemon (#670). Uninstall the one you don't use, then run \`squadrant heal daemon\` to reconcile.
 `;
 }
+function printVersionUpgradeNotice(foreign) {
+  return `[squadrant] upgraded plist: ${foreign.registeredEntry} -> ${foreign.thisEntry}
+`;
+}
 function reregisterDaemon(nodeBin = process.execPath, kickstartOpts = {}) {
   if (!tryAcquireDaemonLock()) {
     throw new Error("could not acquire the daemon lock \u2014 another squadrant process is already restarting the daemon");
   }
   try {
     const drift = computeDaemonDrift(nodeBin);
+    const pidBeforeReconcile = getDaemonPid(drift.target);
     reconcilePlistAndService(drift);
-    const result = forceKickstartAndVerify(drift.target, kickstartOpts);
+    const result = forceKickstartAndVerify(drift.target, { ...kickstartOpts, pidBefore: pidBeforeReconcile });
     if (!result.restarted) {
       throw new Error(`daemon did not restart (pid before=${result.pidBefore ?? "none"}, after=${result.pidAfter ?? "none"}) \u2014 \`launchctl kickstart -k ${drift.target}\` ran but the pid never changed`);
     }
@@ -2587,13 +2636,14 @@ function reregisterDaemon(nodeBin = process.execPath, kickstartOpts = {}) {
     releaseDaemonLock();
   }
 }
-var LABEL, AGENT_BINS, restartInFlight, OPERATOR_INITIATED_COMMANDS;
+var LABEL, AGENT_BINS, restartInFlight, OPERATOR_INITIATED_COMMANDS, READ_ONLY_CREW_SUBCOMMANDS;
 var init_launchd = __esm({
   "packages/core/dist/launchd.js"() {
     LABEL = "com.squadrant.daemon";
     AGENT_BINS = ["cmux", "claude", "opencode", "codex", "gemini", "node"];
     restartInFlight = false;
     OPERATOR_INITIATED_COMMANDS = /* @__PURE__ */ new Set(["launch", "init"]);
+    READ_ONLY_CREW_SUBCOMMANDS = /* @__PURE__ */ new Set(["list", "read", "tasks"]);
   }
 });
 
@@ -3290,6 +3340,51 @@ var init_captain_delivery = __esm({
   }
 });
 
+// packages/core/dist/daemon/down-alert.js
+function pad2(n) {
+  return String(n).padStart(2, "0");
+}
+function formatLocalDateTime(d) {
+  return `${d.getFullYear()}-${pad2(d.getMonth() + 1)}-${pad2(d.getDate())} ${pad2(d.getHours())}:${pad2(d.getMinutes())}`;
+}
+function formatTzOffset(d) {
+  const offMin = -d.getTimezoneOffset();
+  const sign = offMin >= 0 ? "+" : "-";
+  const abs = Math.abs(offMin);
+  const hh = pad2(Math.floor(abs / 60));
+  const mm = abs % 60;
+  return mm === 0 ? `${sign}${hh}` : `${sign}${hh}:${pad2(mm)}`;
+}
+function formatDownAlertText(opts) {
+  const start = new Date(opts.startTs);
+  const end = new Date(opts.endTs);
+  return `\u26A0\uFE0F daemon was down ${formatLocalDateTime(start)} \u2192 ${formatLocalDateTime(end)} (${formatTzOffset(end)}), ${opts.minutes} min (last exit reason=${opts.reasonText})`;
+}
+function formatRelative(ms) {
+  const minutes = Math.round(ms / 6e4);
+  if (minutes < 60)
+    return `${minutes} min`;
+  const hours = Math.floor(minutes / 60);
+  const remMinutes = minutes % 60;
+  if (hours < 24)
+    return remMinutes > 0 ? `${hours}h ${remMinutes}m` : `${hours}h`;
+  const days = Math.floor(hours / 24);
+  const remHours = hours % 24;
+  return remHours > 0 ? `${days}d ${remHours}h` : `${days}d`;
+}
+function stalePrefix(generatedAtMs, deliveredAtMs) {
+  const ageMs = deliveredAtMs - generatedAtMs;
+  if (ageMs <= STALE_ALERT_THRESHOLD_MS)
+    return null;
+  return `[stale \u2014 generated ${formatRelative(ageMs)} ago] `;
+}
+var STALE_ALERT_THRESHOLD_MS;
+var init_down_alert = __esm({
+  "packages/core/dist/daemon/down-alert.js"() {
+    STALE_ALERT_THRESHOLD_MS = 60 * 60 * 1e3;
+  }
+});
+
 // packages/core/dist/control-channel.js
 function fallsBackToPane(o) {
   return o.status === "gone" || o.status === "unsupported";
@@ -3442,12 +3537,20 @@ function ensureSocksDir(dir = CC_SOCKS_DIR) {
   if ((fs9.statSync(dir).mode & 511) !== 448)
     fs9.chmodSync(dir, 448);
 }
-async function pollFirstTurnConfirmedAt(getTaskRecord, project, id) {
-  const deadline = Date.now() + FIRST_TURN_HOOK_CONFIRM_WINDOW_MS;
+async function pollFirstTurnConfirmedAt(getTaskRecord, project, id, scrapeSettled, cancel) {
+  let scrapeSettledAt;
+  const markSettled = () => {
+    scrapeSettledAt ??= Date.now();
+  };
+  scrapeSettled.then(markSettled, markSettled);
+  const hardDeadline = Date.now() + FIRST_TURN_HOOK_CONFIRM_MAX_MS;
   for (; ; ) {
+    if (cancel.stopped)
+      return false;
     const rec = await getTaskRecord(project, id).catch(() => void 0);
     if (rec?.firstTurnConfirmedAt)
       return true;
+    const deadline = scrapeSettledAt !== void 0 ? Math.min(hardDeadline, scrapeSettledAt + FIRST_TURN_HOOK_CONFIRM_GRACE_MS) : hardDeadline;
     if (Date.now() >= deadline)
       return false;
     await new Promise((r) => setTimeout(r, FIRST_TURN_HOOK_POLL_INTERVAL_MS));
@@ -3629,7 +3732,9 @@ async function runCrewSpawn(input, config, deps) {
 
 ${buildCompletionProtocol(rec.id, input.project)}`, preLaunchScreen);
     const scrapeDelivered = sendPromise.then((r) => r.delivered).catch(() => false);
-    const delivered = hooksInstalled && deps.getTaskRecord ? await firstTrueOrBothFalse(scrapeDelivered, pollFirstTurnConfirmedAt(deps.getTaskRecord, input.project, rec.id)) : await scrapeDelivered;
+    const cancelHookPoll = { stopped: false };
+    const delivered = hooksInstalled && deps.getTaskRecord ? await firstTrueOrBothFalse(scrapeDelivered, pollFirstTurnConfirmedAt(deps.getTaskRecord, input.project, rec.id, scrapeDelivered, cancelHookPoll)) : await scrapeDelivered;
+    cancelHookPoll.stopped = true;
     if (!delivered) {
       process.stderr.write(`\u26A0\uFE0F  First turn not delivered for crew '${name}' \u2014 use 'squadrant crew send ${input.project} ${name}' to re-send the task.
 `);
@@ -3911,7 +4016,7 @@ async function runCrewList(project, runtime, workspaceId) {
     surfaceId: c.surfaceId
   }));
 }
-var CC_SOCKS_DIR, FIRST_TURN_INLINE_MAX_BYTES, TEMPLATES_DIR, STATE_ROOT, FIRST_TURN_HOOK_CONFIRM_WINDOW_MS, FIRST_TURN_HOOK_POLL_INTERVAL_MS, CLOSE_LOOKUP_RETRIES, CLOSE_LOOKUP_RETRY_DELAY_MS;
+var CC_SOCKS_DIR, FIRST_TURN_INLINE_MAX_BYTES, TEMPLATES_DIR, STATE_ROOT, FIRST_TURN_HOOK_CONFIRM_GRACE_MS, FIRST_TURN_HOOK_CONFIRM_MAX_MS, FIRST_TURN_HOOK_POLL_INTERVAL_MS, CLOSE_LOOKUP_RETRIES, CLOSE_LOOKUP_RETRY_DELAY_MS;
 var init_crew_spawn = __esm({
   "packages/core/dist/crew-spawn.js"() {
     init_control_channel();
@@ -3923,7 +4028,8 @@ var init_crew_spawn = __esm({
     FIRST_TURN_INLINE_MAX_BYTES = 1200;
     TEMPLATES_DIR = path9.join(os4.homedir(), ".config", "squadrant", "templates");
     STATE_ROOT = path9.join(os4.homedir(), ".config", "squadrant", "state");
-    FIRST_TURN_HOOK_CONFIRM_WINDOW_MS = 1e5;
+    FIRST_TURN_HOOK_CONFIRM_GRACE_MS = 15e3;
+    FIRST_TURN_HOOK_CONFIRM_MAX_MS = 18e4;
     FIRST_TURN_HOOK_POLL_INTERVAL_MS = 2e3;
     CLOSE_LOOKUP_RETRIES = 3;
     CLOSE_LOOKUP_RETRY_DELAY_MS = 150;
@@ -4169,7 +4275,13 @@ function createDelivery(ctx, daemonCmux, isSurfaceAlive) {
               log(`delivery seq=${entry.seq} kind=${entry.kind} outcome=stale-terminal-deliver`);
             }
           }
-          const result = await d.deliver(entry, async (text, sendOpts) => {
+          let deliverEntry = entry;
+          if (entry.kind === "captain.message" && entry.payload?.source === "daemon" && entry.message) {
+            const prefix = stalePrefix(new Date(entry.ts).getTime(), Date.now());
+            if (prefix)
+              deliverEntry = { ...entry, message: `${prefix}${entry.message}` };
+          }
+          const result = await d.deliver(deliverEntry, async (text, sendOpts) => {
             let handledByChannel = false;
             try {
               const mode = ctx.captainChannelMode?.() ?? "off";
@@ -4261,6 +4373,7 @@ var init_delivery_loop = __esm({
     init_defer_delivery();
     init_dist();
     init_interactive_probe();
+    init_down_alert();
     init_liveness2();
     init_captain_channel();
     CURSOR_SUBSCRIBER = "captain";
@@ -4695,8 +4808,8 @@ function startDaemon(ctx, opts, pkgVersion) {
   log(`boot pid=${process.pid} version=${pkgVersion} socket=${ctx.sockPath} stateRoot=${stateRoot}`);
   const bootTs = (/* @__PURE__ */ new Date()).toISOString();
   {
-    const sendDownAlert = (minutes, reasonText) => {
-      const text = `\u26A0\uFE0F daemon was down for ${minutes} min (last exit reason=${reasonText})`;
+    const sendDownAlert = (startTs, endTs, minutes, reasonText) => {
+      const text = formatDownAlertText({ startTs, endTs, minutes, reasonText });
       const alertProjects = opts.registeredProjects ?? Object.keys(loadConfig().projects);
       for (const project of alertProjects) {
         appendCaptainMessage({ stateRoot, project, text, source: "daemon" }).catch((e) => log(`boot-gap alert failed project=${project}: ${e.message}`));
@@ -4707,13 +4820,13 @@ function startDaemon(ctx, opts, pkgVersion) {
     if (marker) {
       log(`previous exit ts=${marker.ts} reason=${marker.reason} gap=${((gapMs ?? 0) / 1e3).toFixed(1)}s`);
       if ((gapMs ?? 0) > 6e4)
-        sendDownAlert(Math.round((gapMs ?? 0) / 6e4), marker.reason);
+        sendDownAlert(marker.ts, bootTs, Math.round((gapMs ?? 0) / 6e4), marker.reason);
     } else if (prevRunning) {
       const lastHeartbeatMs = new Date(prevRunning.lastHeartbeatTs).getTime();
       const uncleanGapMs = Math.max(0, Date.now() - lastHeartbeatMs);
       log(`previous exit: UNCLEAN (no marker; last heartbeat ${prevRunning.lastHeartbeatTs}, gap=${(uncleanGapMs / 1e3).toFixed(1)}s)`);
       if (uncleanGapMs > 6e4) {
-        sendDownAlert(Math.round(uncleanGapMs / 6e4), "unclean/unknown \u2014 no exit marker, likely SIGKILL/OOM/power-loss");
+        sendDownAlert(prevRunning.lastHeartbeatTs, bootTs, Math.round(uncleanGapMs / 6e4), "unclean/unknown \u2014 no exit marker, likely SIGKILL/OOM/power-loss");
       }
     } else {
       log("previous exit: none (clean or first boot)");
@@ -4835,6 +4948,7 @@ var init_start = __esm({
     init_server();
     init_mailbox();
     init_exit_marker();
+    init_down_alert();
     init_liveness2();
     init_dist();
     init_snapshot_gather();
@@ -6804,6 +6918,7 @@ __export(dist_exports, {
   OPERATOR_INITIATED_COMMANDS: () => OPERATOR_INITIATED_COMMANDS,
   PROBE_QUIET_MS: () => PROBE_QUIET_MS,
   PROTOCOL_VERSION: () => PROTOCOL_VERSION,
+  READ_ONLY_CREW_SUBCOMMANDS: () => READ_ONLY_CREW_SUBCOMMANDS,
   STALE_THRESHOLD_MS: () => STALE_THRESHOLD_MS,
   TERMINAL_RECORD_KEEP_PER_PROJECT: () => TERMINAL_RECORD_KEEP_PER_PROJECT,
   TERMINAL_RECORD_TTL_MS: () => TERMINAL_RECORD_TTL_MS,
@@ -6852,6 +6967,7 @@ __export(dist_exports, {
   crewTag: () => crewTag,
   daemonEntryPath: () => daemonEntryPath,
   daemonLockPath: () => daemonLockPath,
+  decideForeignInstall: () => decideForeignInstall,
   decodeFrames: () => decodeFrames,
   defaultIsPidAlive: () => defaultIsPidAlive,
   defaultListenError: () => defaultListenError,
@@ -6893,9 +7009,11 @@ __export(dist_exports, {
   isDaemonSocketLive: () => isDaemonSocketLive,
   isNotifyActive: () => isNotifyActive,
   isOperatorInitiatedCommand: () => isOperatorInitiatedCommand,
+  isReadOnlyCrewCommand: () => isReadOnlyCrewCommand,
   isSideTitle: () => isSideTitle,
   isStickyAttention: () => isStickyAttention,
   isTurnAccepted: () => isTurnAccepted,
+  isVersionUpgrade: () => isVersionUpgrade,
   kickstartArgv: () => kickstartArgv,
   launchOneWorkspace: () => launchOneWorkspace,
   loadSessions: () => loadSessions,
@@ -6912,6 +7030,7 @@ __export(dist_exports, {
   parseProgramArgs: () => parseProgramArgs,
   plistPath: () => plistPath,
   printForeignInstallError: () => printForeignInstallError,
+  printVersionUpgradeNotice: () => printVersionUpgradeNotice,
   programArgsBlock: () => programArgsBlock,
   projectHealth: () => projectHealth,
   purgeExpiredWorkItems: () => purgeExpiredWorkItems,
@@ -7887,7 +8006,7 @@ import { execSync as execSync6 } from "child_process";
 import { readFileSync as readFileSync10 } from "fs";
 import { homedir as homedir8 } from "os";
 import { join as join15 } from "path";
-var nextAskUserQuestionRequestId = Date.now();
+var nextHookRequestId = Date.now();
 
 // packages/agents/dist/headless/types.js
 var HEADLESS_ERROR_TAIL = 2e3;
@@ -9593,7 +9712,13 @@ var CLAUDE_HOOK_EVENTS = [
   ["Stop", "stop"],
   ["Notification", "notification"],
   ["PreToolUse", "ask-question", "AskUserQuestion"],
-  ["SessionEnd", "session-end"]
+  ["SessionEnd", "session-end"],
+  // #760: fires ~6s before the matching Notification, carrying tool_name +
+  // tool_input directly — a strictly better task.blocked source.
+  ["PermissionRequest", "permission-request"],
+  // #763: the turn died on an API error — today squadrant has no source for
+  // this at all; without it the watchdog reports a plain stall (wrong story).
+  ["StopFailure", "stop-failure"]
 ];
 var DEFAULT_HOOK_CMD = "squadrant hooks";
 function installClaudeHooks(opts = {}) {
@@ -9671,6 +9796,10 @@ function mapSubToLifecycle(sub) {
       return "needsInput";
     case "ask-question":
       return "needsInput";
+    case "permission-request":
+      return "needsInput";
+    case "stop-failure":
+      return "idle";
     case "session-end":
       return "session-end";
     default:
@@ -9809,9 +9938,12 @@ async function confirmedSendToPane(runtime, pane, message) {
   let sawDraft = await settleInputBox(runtime, pane);
   await runtime.sendKeyToPane(pane, "Enter");
   let repasted = false;
+  let screenMoved = false;
   for (let attempt = 0; attempt < SUBMIT_RETRY_LIMIT; attempt++) {
     await new Promise((r) => setTimeout(r, POST_SEND_CHECK_MS));
     const afterScreen = await runtime.readPaneScreen(pane) ?? "";
+    if (afterScreen !== preSendScreen)
+      screenMoved = true;
     const draft = parseDraftFromScreen(afterScreen);
     if (draft !== "" && draft !== null)
       sawDraft = true;
@@ -9822,7 +9954,7 @@ async function confirmedSendToPane(runtime, pane, message) {
     const settled = await settleInputBox(runtime, pane);
     if (settled)
       sawDraft = true;
-    if (!sawDraft && !repasted) {
+    if (!sawDraft && !repasted && !screenMoved) {
       repasted = true;
       await runtime.pasteToPane(pane, message);
     }