@integrity-labs/agt-cli 0.27.124 → 0.27.126

package/dist/lib/manager-worker.js

@@ -17,7 +17,7 @@ import {
   provisionStopHook,
   requireHost,
   safeWriteJsonAtomic
-} from "../chunk-MQHZBG7J.js";
+} from "../chunk-LTZS7SHV.js";
 import {
   getProjectDir as getProjectDir2,
   getReadyTasks,
@@ -56,7 +56,7 @@ import {
   stopPersistentSession,
   takeWatchdogGiveUpCount,
   takeZombieDetection
-} from "../chunk-AZEYTJ4L.js";
+} from "../chunk-QON5CU3L.js";
 import {
   KANBAN_CHECK_COMMAND,
   SUPPRESS_SENTINEL,
@@ -83,7 +83,7 @@ import {
   resolveDmTarget,
   worseConnectivityOutcome,
   wrapScheduledTaskPrompt
-} from "../chunk-Z6JJRLHH.js";
+} from "../chunk-MH7HA6QV.js";
 import {
   parsePsRows,
   reapOrphanChannelMcps
@@ -127,6 +127,22 @@ function channelSecretValueHash(envEntries, channelSecretKeys) {
   return createHash("sha256").update(basis).digest("hex").slice(0, 16);
 }
 
+// src/lib/mcp-restart-verify.ts
+var RESTART_VERIFY_DEADLINE_MS = 9e4;
+var RESTART_VERIFY_MAX_ATTEMPTS = 3;
+function decidePostRestartVerification(pending, sessionStartedAt, sessionHealthy, sessionRespawnId, now, opts = {}) {
+  const deadlineMs = opts.deadlineMs ?? RESTART_VERIFY_DEADLINE_MS;
+  const maxAttempts = opts.maxAttempts ?? RESTART_VERIFY_MAX_ATTEMPTS;
+  if (sessionStartedAt !== null && sessionStartedAt > pending.firedAt && sessionHealthy && sessionRespawnId !== null) {
+    return { kind: "verified" };
+  }
+  if (now - pending.firedAt < deadlineMs) {
+    return { kind: "waiting" };
+  }
+  const attempt = pending.attempts + 1;
+  return { kind: "unverified", attempt, final: attempt >= maxAttempts };
+}
+
 // src/lib/integration-hash.ts
 import { createHash as createHash2 } from "crypto";
 function canonicalize(value) {
@@ -3959,6 +3975,7 @@ function hasRevokedResiduals(state6) {
   return state6.gatewayRunning || state6.portAllocated || state6.provisionDirExists;
 }
 var pendingSessionRestarts = /* @__PURE__ */ new Map();
+var pendingRestartVerifications = /* @__PURE__ */ new Map();
 var restartBreaker = new RestartBreaker();
 var reportedTrips = /* @__PURE__ */ new Map();
 var mcpFlapDampener = new McpFlapDampener();
@@ -3974,6 +3991,20 @@ function recordConfigChurnEvent(agentId, codeName, channel, signature) {
     );
   });
 }
+async function reportReaperEvent(agentId, reaper, action, reason) {
+  try {
+    await api.post("/host/reaper-event", {
+      agent_id: agentId,
+      reaper,
+      action,
+      reason
+    });
+  } catch (err) {
+    log(
+      `[${reaper}] failed to report reaper-event (ENG-6184): ${err.message} \u2014 local reap still proceeded; the alert may be missed`
+    );
+  }
+}
 function recordRestartForBreaker(codeName, reason) {
   const result = restartBreaker.record(codeName, reason);
   if (!result.tripped || !result.trip) return;
@@ -4087,11 +4118,19 @@ function scheduleSessionRestart(codeName, delayMs, reason, breakerReason = "hot-
     runningMcpHashes.delete(codeName);
     recordRestartForBreaker(codeName, breakerReason);
     log(`[hot-reload] Session stopped for '${codeName}' \u2014 will respawn with ${reason}`);
+    if (breakerReason === "hot-reload-mcp") {
+      const prior = pendingRestartVerifications.get(codeName);
+      pendingRestartVerifications.set(codeName, {
+        firedAt: Date.now(),
+        attempts: prior?.attempts ?? 0
+      });
+    }
   }, delayMs);
   timer.unref?.();
   pendingSessionRestarts.set(codeName, timer);
 }
 function cancelPendingSessionRestart(codeName) {
+  pendingRestartVerifications.delete(codeName);
   const existing = pendingSessionRestarts.get(codeName);
   if (!existing) return;
   clearTimeout(existing);
@@ -4099,6 +4138,37 @@ function cancelPendingSessionRestart(codeName) {
   deferLogThrottle.delete(codeName);
   log(`[hot-reload] Cancelled pending restart timer for '${codeName}' (another teardown path is handling it)`);
 }
+function verifyPendingRestarts(now) {
+  if (pendingRestartVerifications.size === 0) return;
+  for (const [codeName, pending] of pendingRestartVerifications) {
+    const healthy = isSessionHealthy(codeName);
+    const session = getSessionState(codeName);
+    const startedAt = session?.startedAt ?? null;
+    const respawnId = session?.currentSessionId ?? null;
+    const outcome = decidePostRestartVerification(pending, startedAt, healthy, respawnId, now);
+    switch (outcome.kind) {
+      case "verified":
+        pendingRestartVerifications.delete(codeName);
+        log(`[restart-verify] '${codeName}' respawned healthy after MCP change \u2014 tools bound (ENG-6174)`);
+        break;
+      case "waiting":
+        break;
+      case "unverified":
+        if (outcome.final) {
+          pendingRestartVerifications.delete(codeName);
+          log(
+            `[restart-verify] ERROR '${codeName}' did NOT respawn healthy after ${outcome.attempt} attempts following an MCP change \u2014 the live session may be stuck on its pre-restart tools. Check session health / manager.log; a manual session restart may be required (ENG-6174)`
+          );
+        } else {
+          pendingRestartVerifications.set(codeName, { firedAt: now, attempts: outcome.attempt });
+          log(
+            `[restart-verify] WARN '${codeName}' not yet respawned healthy after MCP change (attempt ${outcome.attempt}) \u2014 still watching (ENG-6174)`
+          );
+        }
+        break;
+    }
+  }
+}
 var RESTART_DEFER_RECHECK_MS = 6e4;
 var DEFER_LOG_THROTTLE_MS = 6e5;
 var deferLogThrottle = /* @__PURE__ */ new Map();
@@ -4424,7 +4494,7 @@ var cachedMaintenanceWindow = null;
 var lastVersionCheckAt = 0;
 var VERSION_CHECK_INTERVAL_MS = 5 * 60 * 1e3;
 var lastResponsivenessProbeAt = 0;
-var agtCliVersion = true ? "0.27.124" : "dev";
+var agtCliVersion = true ? "0.27.126" : "dev";
 function resolveBrewPath(execFileSync4) {
   try {
     const out = execFileSync4("which", ["brew"], { timeout: 5e3 }).toString().trim();
@@ -5617,7 +5687,7 @@ async function pollCycle() {
     }
     try {
       const { detectHostSecurity } = await import("../host-security-6PDFG7F5.js");
-      const { collectDiagnostics } = await import("../persistent-session-YWGDREIZ.js");
+      const { collectDiagnostics } = await import("../persistent-session-PJQZYG2L.js");
       const diagCodeNames = [...agentState.persistentSessionAgents];
       const agentDiagnostics = diagCodeNames.length > 0 ? collectDiagnostics(diagCodeNames) : void 0;
       let tailscaleHostname;
@@ -5704,12 +5774,12 @@ async function pollCycle() {
       const {
         collectResponsivenessProbes,
         getResponsivenessIntervalMs
-      } = await import("../responsiveness-probe-6YYCQAFI.js");
+      } = await import("../responsiveness-probe-MGMZQSP7.js");
       const probeIntervalMs = getResponsivenessIntervalMs();
       if (now - lastResponsivenessProbeAt > probeIntervalMs) {
         const probeCodeNames = [...agentState.persistentSessionAgents];
         if (probeCodeNames.length > 0) {
-          const { takeAcpxExecFailureCount, creditAcpxExecFailureCount } = await import("../persistent-session-YWGDREIZ.js");
+          const { takeAcpxExecFailureCount, creditAcpxExecFailureCount } = await import("../persistent-session-PJQZYG2L.js");
           const drainedGiveUps = /* @__PURE__ */ new Map();
           const drainedAcpxFailures = /* @__PURE__ */ new Map();
           const probes = collectResponsivenessProbes(probeCodeNames).map((p) => {
@@ -5743,7 +5813,7 @@ async function pollCycle() {
           collectResponsivenessProbes,
           livePendingInboundOldestAgeSeconds,
           deadLetterPendingInbound
-        } = await import("../responsiveness-probe-6YYCQAFI.js");
+        } = await import("../responsiveness-probe-MGMZQSP7.js");
         const wedgeNow = /* @__PURE__ */ new Date();
         const liveAgents = agentState.persistentSessionAgents;
         for (const tracked of consecutiveWedgeCycles.keys()) {
@@ -6092,6 +6162,7 @@ async function pollCycle() {
       log(`[poll-backoff] recovered after ${consecutivePollFailures} failure(s), resuming normal interval`);
       consecutivePollFailures = 0;
     }
+    verifyPendingRestarts(Date.now());
     send({ type: "state-update", state: state5 });
   } catch (err) {
     state5.errorCount++;
@@ -7550,6 +7621,12 @@ async function processAgent(agent, agentStates) {
               `[mcp-presence-reaper] failed to record restart event for '${codeName}' (ENG-5286): ${err.message} \u2014 local restart still proceeded; CloudWatch metric will under-count this event`
             );
           });
+          void reportReaperEvent(
+            agent.agent_id,
+            "mcp-presence-reaper",
+            "restart",
+            givenUpKeys.length > 0 ? `restarted session; MCP server(s) still down after restart budget: ${givenUpKeys.join(", ")}` : `restarted session to recover MCP server(s): ${activeKeys.join(", ")}`
+          );
         },
         // ENG-5932: classify each declared key for the quarantine path. Reads a
         // server-declared criticality field off the .mcp.json entry when
@@ -7580,6 +7657,12 @@ async function processAgent(agent, agentStates) {
             log(
               `[channel-quarantine] persisted quarantine marker for '${codeName}:${serverKey}' \u2014 will be dropped from .mcp.json on next provisioning poll (ENG-5932)`
             );
+            void reportReaperEvent(
+              agent.agent_id,
+              "mcp-presence-reaper",
+              "quarantine",
+              `optional channel '${serverKey}' dead past restart budget + dwell \u2014 quarantined`
+            );
           }
         }
       });
@@ -10227,7 +10310,7 @@ async function processClaudePairSessions(agents) {
     killPairSession,
     pairTmuxSession,
     finalizeClaudePairOnboarding
-  } = await import("../claude-pair-runtime-ERGB26MZ.js");
+  } = await import("../claude-pair-runtime-QNOWFDJ7.js");
   for (const pairId of pendingResp.cancelled_pair_ids ?? []) {
     log(`[claude-pair] sweeping orphan tmux session for pair ${pairId.slice(0, 8)}`);
     const killed = await killPairSession(pairTmuxSession(pairId));