@integrity-labs/agt-cli 0.28.272 → 0.28.274

package/dist/{claude-pair-runtime-KOEEK5LD.js → claude-pair-runtime-Z726QSAR.js}

@@ -100,7 +100,7 @@ async function spawnPairSession(session) {
     return { ok: true };
   } catch {
   }
-  const { resolveClaudeBinary } = await import("./persistent-session-HOILVE3Q.js");
+  const { resolveClaudeBinary } = await import("./persistent-session-SDDSZN5E.js");
   const claudeBin = resolveClaudeBinary();
   const pairEnv = {
     ...process.env,
@@ -373,4 +373,4 @@ export {
   startClaudePair,
   submitClaudePairCode
 };
-//# sourceMappingURL=claude-pair-runtime-KOEEK5LD.js.map
+//# sourceMappingURL=claude-pair-runtime-Z726QSAR.js.map

package/dist/lib/manager-worker.js

@@ -38,7 +38,7 @@ import {
   requireHost,
   safeWriteJsonAtomic,
   setConfigHash
-} from "../chunk-5HCINBKO.js";
+} from "../chunk-73NA6BKT.js";
 import {
   getProjectDir as getProjectDir2,
   getReadyTasks,
@@ -52,6 +52,7 @@ import {
   FLAGS_SCHEMA_VERSION,
   FLAG_REGISTRY,
   KANBAN_CHECK_COMMAND,
+  KANBAN_NUDGE_ACTIONABLE_STATUSES,
   MAX_AVATAR_ENV_URL_BYTES,
   SUPPRESS_SENTINEL,
   StreamEncoder,
@@ -124,7 +125,7 @@ import {
   takeZombieDetection,
   transcriptActivityAgeSeconds,
   writeEgressAllowlist
-} from "../chunk-5TL5THQE.js";
+} from "../chunk-THMHJITM.js";
 import {
   reapOrphanChannelMcps
 } from "../chunk-XWVM4KPK.js";
@@ -610,8 +611,27 @@ function isQuarantineFile(value) {
 function reaperRestartBreakerReason(activeKeys) {
   return activeKeys.length >= 2 ? "mcp-presence-reaper" : void 0;
 }
+var PROVISIONING_RELOAD_REASONS = /* @__PURE__ */ new Set([
+  "hot-reload-mcp",
+  "managed-mcp-churn",
+  "bind-remediation"
+]);
+function isProvisioningReloadReason(reason) {
+  return PROVISIONING_RELOAD_REASONS.has(reason);
+}
+function countByClass(events) {
+  let crash = 0;
+  let provisioning = 0;
+  for (const e of events) {
+    if (isProvisioningReloadReason(e.reason)) provisioning += 1;
+    else crash += 1;
+  }
+  return { crash, provisioning };
+}
 var DEFAULT_MAX = 2;
 var DEFAULT_WINDOW_MS = 6e5;
+var DEFAULT_PROVISIONING_MAX = 5;
+var DEFAULT_PROVISIONING_WINDOW_MS = 18e5;
 function readEnvNumber(name, fallback) {
   const raw = process.env[name];
   if (!raw) return fallback;
@@ -621,12 +641,18 @@ function readEnvNumber(name, fallback) {
 var RestartBreaker = class {
   max;
   windowMs;
+  provisioningMax;
+  provisioningWindowMs;
+  /** Longest window either tally needs — how long the events log must retain. */
+  retentionMs;
   now;
   events = /* @__PURE__ */ new Map();
   trips = /* @__PURE__ */ new Map();
   constructor(opts = {}) {
     this.max = opts.max ?? readEnvNumber("AGT_RESTART_BREAKER_MAX", DEFAULT_MAX);
     this.windowMs = opts.windowMs ?? readEnvNumber("AGT_RESTART_BREAKER_WINDOW_MS", DEFAULT_WINDOW_MS);
+    this.provisioningMax = opts.provisioningMax ?? readEnvNumber("AGT_RESTART_BREAKER_PROVISIONING_MAX", DEFAULT_PROVISIONING_MAX);
+    this.provisioningWindowMs = opts.provisioningWindowMs ?? readEnvNumber("AGT_RESTART_BREAKER_PROVISIONING_WINDOW_MS", DEFAULT_PROVISIONING_WINDOW_MS);
     this.now = opts.now ?? Date.now;
     if (!Number.isFinite(this.max) || this.max < 1) {
       throw new Error("restart-breaker max must be a finite number >= 1");
@@ -634,6 +660,13 @@ var RestartBreaker = class {
     if (!Number.isFinite(this.windowMs) || this.windowMs < 1e3) {
       throw new Error("restart-breaker windowMs must be a finite number >= 1000");
     }
+    if (!Number.isFinite(this.provisioningMax) || this.provisioningMax < 1) {
+      throw new Error("restart-breaker provisioningMax must be a finite number >= 1");
+    }
+    if (!Number.isFinite(this.provisioningWindowMs) || this.provisioningWindowMs < 1e3) {
+      throw new Error("restart-breaker provisioningWindowMs must be a finite number >= 1000");
+    }
+    this.retentionMs = Math.max(this.windowMs, this.provisioningWindowMs);
   }
   /** True if this agent's breaker is currently tripped (manager must skip spawn). */
   isTripped(codeName) {
@@ -654,24 +687,54 @@ var RestartBreaker = class {
   record(codeName, reason) {
     const existing = this.trips.get(codeName);
     if (existing) {
-      return { tripped: false, trip: existing, windowCount: existing.eventsAtTrip.length };
+      return {
+        tripped: false,
+        trip: existing,
+        windowCount: existing.eventsAtTrip.length,
+        classCounts: countByClass(existing.eventsAtTrip)
+      };
     }
     const at = this.now();
-    const cutoff = at - this.windowMs;
-    const prior = (this.events.get(codeName) ?? []).filter((e) => e.at >= cutoff);
+    const retentionCutoff = at - this.retentionMs;
+    const prior = (this.events.get(codeName) ?? []).filter((e) => e.at >= retentionCutoff);
     prior.push({ reason, at });
     this.events.set(codeName, prior);
-    if (prior.length > this.max) {
+    const crashCount = prior.filter(
+      (e) => e.at >= at - this.windowMs && !isProvisioningReloadReason(e.reason)
+    ).length;
+    const provEvents = prior.filter(
+      (e) => e.at >= at - this.provisioningWindowMs && isProvisioningReloadReason(e.reason)
+    );
+    const windowCount = prior.filter((e) => e.at >= at - this.windowMs).length;
+    const classCounts = {
+      crash: crashCount,
+      provisioning: provEvents.length
+    };
+    let trippedClass;
+    let tripEvents;
+    let tripWindowMs = this.windowMs;
+    if (crashCount > this.max) {
+      trippedClass = "crash";
+      tripEvents = prior.filter(
+        (e) => e.at >= at - this.windowMs && !isProvisioningReloadReason(e.reason)
+      );
+      tripWindowMs = this.windowMs;
+    } else if (provEvents.length > this.provisioningMax) {
+      trippedClass = "provisioning";
+      tripEvents = provEvents;
+      tripWindowMs = this.provisioningWindowMs;
+    }
+    if (trippedClass && tripEvents) {
       const trip = {
         trippedAt: at,
-        eventsAtTrip: [...prior],
-        statusMessage: formatStatusMessage(prior, this.windowMs)
+        eventsAtTrip: [...tripEvents],
+        statusMessage: formatStatusMessage(tripEvents, tripWindowMs, trippedClass)
       };
       this.trips.set(codeName, trip);
       this.events.delete(codeName);
-      return { tripped: true, trip, windowCount: prior.length };
+      return { tripped: true, trip, windowCount, classCounts, trippedClass };
     }
-    return { tripped: false, windowCount: prior.length };
+    return { tripped: false, windowCount, classCounts };
   }
   /** Operator-initiated reset: drops the trip + the events log for this agent. */
   clear(codeName) {
@@ -702,13 +765,14 @@ var RestartBreaker = class {
     return (this.events.get(codeName) ?? []).filter((e) => e.at >= cutoff).length;
   }
 };
-function formatStatusMessage(events, windowMs) {
+function formatStatusMessage(events, windowMs, klass = "crash") {
   const last = events[events.length - 1];
   const windowLabel = windowMs < 6e4 ? `${Math.round(windowMs / 1e3)}s` : `${(windowMs / 6e4).toFixed(1).replace(/\.0$/, "")}min`;
   const reasonCounts = /* @__PURE__ */ new Map();
   for (const e of events) reasonCounts.set(e.reason, (reasonCounts.get(e.reason) ?? 0) + 1);
   const breakdown = Array.from(reasonCounts.entries()).map(([r, n]) => `${r}=${n}`).join(", ");
-  return `Circuit breaker tripped: ${events.length} restarts in ${windowLabel} (${breakdown}); most recent=${last.reason} at ${new Date(last.at).toISOString()}`;
+  const classLabel = klass === "provisioning" ? "provisioning-reload " : "";
+  return `Circuit breaker tripped: ${events.length} ${classLabel}restarts in ${windowLabel} (${breakdown}); most recent=${last.reason} at ${new Date(last.at).toISOString()}`;
 }
 
 // src/lib/mcp-flap-dampener.ts
@@ -3745,7 +3809,7 @@ function isKanbanHybridDryRun() {
   const v = process.env["AGT_KANBAN_HYBRID_DRY_RUN"];
   return v === "1" || v?.toLowerCase() === "true";
 }
-var HYBRID_ACTIONABLE_STATUSES = /* @__PURE__ */ new Set(["todo", "in_progress"]);
+var HYBRID_ACTIONABLE_STATUSES = new Set(KANBAN_NUDGE_ACTIONABLE_STATUSES);
 function isHybridActionable(item) {
   return HYBRID_ACTIONABLE_STATUSES.has(item.status) && item.source_type !== "scheduled_task";
 }
@@ -4016,6 +4080,21 @@ async function enqueueKanbanNotice(opts) {
     return false;
   }
 }
+async function cancelKanbanNotice(agentId) {
+  try {
+    const res = await api.post(
+      "/host/kanban/notify/cancel",
+      { agent_id: agentId }
+    );
+    if (res.ok !== true) return null;
+    return typeof res.cancelled === "number" ? res.cancelled : 0;
+  } catch (err) {
+    const errText = err instanceof Error ? err.message : String(err);
+    const errId = createHash7("sha256").update(errText).digest("hex").slice(0, 12);
+    log(`[kanban] notice cancel failed for agent_id=${agentId} error_id=${errId}`);
+    return null;
+  }
+}
 
 // src/lib/manager/channels/state.ts
 var agentChannelTokens = /* @__PURE__ */ new Map();
@@ -5937,6 +6016,7 @@ var pendingSessionRestarts = /* @__PURE__ */ new Map();
 var pendingRestartVerifications = /* @__PURE__ */ new Map();
 var restartBreaker = new RestartBreaker();
 var reportedTrips = /* @__PURE__ */ new Map();
+var RESTART_BREAKER_PROVISIONING_MAX = readEnvNumber("AGT_RESTART_BREAKER_PROVISIONING_MAX", 5);
 var mcpFlapDampener = new McpFlapDampener();
 var FLAP_BREAKER_COOLDOWN_MS = readEnvNumber2(
   "AGT_MCP_FLAP_BREAKER_COOLDOWN_MS",
@@ -5976,7 +6056,14 @@ function recordRestartForBreaker(codeName, reason) {
     return;
   }
   const result = restartBreaker.record(codeName, reason);
-  if (!result.tripped || !result.trip) return;
+  if (!result.tripped || !result.trip) {
+    if (isProvisioningReloadReason(reason) && result.classCounts.provisioning >= RESTART_BREAKER_PROVISIONING_MAX) {
+      log(
+        `[restart-breaker] agent=${codeName} provisioning-reload churn at the trip bar: ${result.classCounts.provisioning} provisioning restarts in window (reason=${reason}); one more will auto-pause the agent \u2014 a freshly-added integration is likely churning (ENG-7560)`
+      );
+    }
+    return;
+  }
   const trip = result.trip;
   log(
     `[persistent-session-decision] agent=${codeName} decision=circuit-tripped detail=${JSON.stringify(trip.statusMessage)} (ENG-5441)`
@@ -6701,7 +6788,7 @@ var agentRestartTimezoneInputs = /* @__PURE__ */ new Map();
 var lastVersionCheckAt = 0;
 var VERSION_CHECK_INTERVAL_MS = 5 * 60 * 1e3;
 var lastResponsivenessProbeAt = 0;
-var agtCliVersion = true ? "0.28.272" : "dev";
+var agtCliVersion = true ? "0.28.274" : "dev";
 function resolveBrewPath(execFileSync2) {
   try {
     const out = execFileSync2("which", ["brew"], { timeout: 5e3 }).toString().trim();
@@ -7551,7 +7638,7 @@ function flushRestartedAgentDiagnostics(hostId, codeNames) {
   if (codeNames.length === 0) return;
   void (async () => {
     try {
-      const { collectDiagnostics } = await import("../persistent-session-HOILVE3Q.js");
+      const { collectDiagnostics } = await import("../persistent-session-SDDSZN5E.js");
       await api.post("/host/heartbeat", {
         host_id: hostId,
         agent_diagnostics: collectDiagnostics(codeNames)
@@ -7649,7 +7736,7 @@ async function pollCycle() {
     }
     try {
       const { detectHostSecurity } = await import("../host-security-6PDFG7F5.js");
-      const { collectDiagnostics } = await import("../persistent-session-HOILVE3Q.js");
+      const { collectDiagnostics } = await import("../persistent-session-SDDSZN5E.js");
       const diagCodeNames = [...agentState.persistentSessionAgents];
       const agentDiagnostics = diagCodeNames.length > 0 ? collectDiagnostics(diagCodeNames) : void 0;
       let tailscaleHostname;
@@ -7797,7 +7884,7 @@ async function pollCycle() {
       const {
         collectResponsivenessProbes,
         getResponsivenessIntervalMs
-      } = await import("../responsiveness-probe-3OIDWR3T.js");
+      } = await import("../responsiveness-probe-ODKT4WUR.js");
       const probeIntervalMs = getResponsivenessIntervalMs();
       if (now - lastResponsivenessProbeAt > probeIntervalMs) {
         const probeCodeNames = [...agentState.persistentSessionAgents];
@@ -7829,7 +7916,7 @@ async function pollCycle() {
           collectResponsivenessProbes,
           livePendingInboundOldestAgeSeconds,
           parkPendingInbound
-        } = await import("../responsiveness-probe-3OIDWR3T.js");
+        } = await import("../responsiveness-probe-ODKT4WUR.js");
         const { getProjectDir: wedgeProjectDir } = await import("../claude-scheduler-FATCLHDM.js");
         const wedgeNow = /* @__PURE__ */ new Date();
         const liveAgents = agentState.persistentSessionAgents;
@@ -9744,6 +9831,7 @@ async function processAgent(agent, agentStates) {
   const wantsHybridInject = isKanbanHybridEnabled();
   if (agentFw === "claude-code" && sessionMode === "persistent") {
     let hybridBoard = hasBoardTemplates ? boardItems : void 0;
+    let hybridBoardStale = false;
     if (!hybridBoard) {
       try {
         const boardData = await api.post(
@@ -9754,6 +9842,7 @@ async function processAgent(agent, agentStates) {
         kanbanBoardCache.set(agent.code_name, hybridBoard);
       } catch {
         hybridBoard = kanbanBoardCache.get(agent.code_name) ?? [];
+        hybridBoardStale = true;
       }
     }
     void reconcileScheduledRuns(agent.code_name, agent.agent_id, hybridBoard);
@@ -9769,11 +9858,24 @@ async function processAgent(agent, agentStates) {
       const cardStates = wedgeRestartsByCard.get(agent.code_name);
       const actionable = hybridBoard.filter(isHybridActionable);
       const { allowed, suppressed } = cardStates ? partitionActionableByPoison(actionable, cardStates, poisonCfg) : { allowed: actionable, suppressed: [] };
-      if (allowed.length > 0) {
+      if (hybridBoardStale) {
+        if (allowed.length > 0) {
+          log(
+            `[manager-worker] kanban inject skipped for '${agent.code_name}' - board fetch failed this tick (stale cache)`
+          );
+        }
+      } else if (allowed.length > 0) {
         void maybeInjectKanbanCheck(agent.code_name, agent.agent_id, allowed.length);
-      } else {
-        const reason = suppressed.length > 0 ? "kanban inject suppressed \u2014 suspected poison card(s) (ENG-6171)" : "kanban board drained";
-        closeInjectedRunIfOpen(agent.code_name, "completed", reason);
+      } else if (suppressed.length > 0) {
+        closeInjectedRunIfOpen(
+          agent.code_name,
+          "completed",
+          "kanban inject suppressed \u2014 suspected poison card(s) (ENG-6171)"
+        );
+      } else if (openInjectedRunByCode.has(agent.code_name)) {
+        const openRunId = openInjectedRunByCode.get(agent.code_name);
+        openInjectedRunByCode.delete(agent.code_name);
+        void cancelKanbanNoticeOnDrain(agent.code_name, agent.agent_id, openRunId);
       }
     }
   }
@@ -10700,7 +10802,7 @@ async function handleRestartDoorbell(agentId, requestedAt, restartReason) {
     void api.post("/host/restart-ack", { host_id: hostId, agent_id: agentId, restart_requested_at: requestedAt }).catch((err) => log(`[restart-lane] ack failed for '${codeName}': ${err.message}`));
     void (async () => {
       try {
-        const { collectDiagnostics } = await import("../persistent-session-HOILVE3Q.js");
+        const { collectDiagnostics } = await import("../persistent-session-SDDSZN5E.js");
         await api.post("/host/heartbeat", {
           host_id: hostId,
           agent_diagnostics: collectDiagnostics([codeName])
@@ -10750,7 +10852,7 @@ async function respawnAgentAfterMcpStop(codeName, reason) {
     }
     try {
       const hostId = await getHostId();
-      const { collectDiagnostics } = await import("../persistent-session-HOILVE3Q.js");
+      const { collectDiagnostics } = await import("../persistent-session-SDDSZN5E.js");
       await api.post("/host/heartbeat", {
         host_id: hostId,
         agent_diagnostics: collectDiagnostics([codeName])
@@ -11031,6 +11133,19 @@ function closeInjectedRunIfOpen(codeName, outcome, outcomeMessage) {
   openInjectedRunByCode.delete(codeName);
   void finishRun(runId, outcome, { outcomeMessage });
 }
+async function cancelKanbanNoticeOnDrain(codeName, agentId, openRunId) {
+  const cancelled = await cancelKanbanNotice(agentId);
+  if (cancelled != null && cancelled > 0) {
+    log(
+      `[manager-worker] cancelled ${cancelled} pending kanban nudge(s) for '${codeName}' - board drained before delivery`
+    );
+    void finishRun(openRunId, "cancelled", {
+      outcomeMessage: "kanban nudge cancelled - board drained before delivery"
+    });
+  } else {
+    void finishRun(openRunId, "completed", { outcomeMessage: "kanban board drained" });
+  }
+}
 var KANBAN_HYBRID_DEBOUNCE_MS = (() => {
   const raw = process.env["AGT_KANBAN_HYBRID_DEBOUNCE_SECONDS"];
   const parsed = raw ? parseInt(raw, 10) : NaN;
@@ -11141,7 +11256,7 @@ async function processClaudePairSessions(agents) {
     killPairSession,
     pairTmuxSession,
     finalizeClaudePairOnboarding
-  } = await import("../claude-pair-runtime-KOEEK5LD.js");
+  } = await import("../claude-pair-runtime-Z726QSAR.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));
@@ -12027,6 +12142,7 @@ export {
   BACK_ONLINE_GREETING_GUIDANCE,
   DAY_ROLLOVER_FORCE_GRACE_MIN,
   applyRestartAcks,
+  cancelKanbanNoticeOnDrain,
   claudeCodeUpgradeMarkerPath,
   claudeCodeUpgradeThrottled,
   claudeManagedSettingsPath,