@integrity-labs/agt-cli 0.28.1158 → 0.28.1160

package/dist/lib/manager-worker.js

@@ -77,7 +77,7 @@ import {
   setConfigHash,
   shellGuard,
   tripClass
-} from "../chunk-VYBTCFQQ.js";
+} from "../chunk-Y5W6WEJL.js";
 import {
   CLAUDE_CODE_SECURITY_FLOOR,
   SELF_UPDATE_TIMEOUTS,
@@ -4721,6 +4721,26 @@ var SETTLE_MS = 3e4;
 var TRANSCRIPT_MTIME_WINDOW_MS2 = 2 * 24 * 60 * 60 * 1e3;
 var MAX_RUNS_PER_TICK = 20;
 var lastCheckedAt = /* @__PURE__ */ new Map();
+var sharedWorkList = null;
+var sharedWorkListInFlight = null;
+async function unreconciledRunsForHost(api2, nowMs) {
+  if (sharedWorkList && nowMs - sharedWorkList.fetchedAt < MIN_CHECK_INTERVAL_MS4) {
+    return sharedWorkList.runs;
+  }
+  if (sharedWorkListInFlight) return sharedWorkListInFlight;
+  const flight = (async () => {
+    const res = await api2.get("/host/workflow-runs/unreconciled");
+    const runs = res?.runs ?? [];
+    sharedWorkList = { fetchedAt: nowMs, runs };
+    return runs;
+  })();
+  sharedWorkListInFlight = flight;
+  try {
+    return await flight;
+  } finally {
+    if (sharedWorkListInFlight === flight) sharedWorkListInFlight = null;
+  }
+}
 function emptyTotals() {
   return { inputTokens: 0, outputTokens: 0, cacheCreationTokens: 0, cacheReadTokens: 0 };
 }
@@ -4798,8 +4818,7 @@ async function maybeReconcileWorkflowRunTokens(args) {
   lastCheckedAt.set(codeName, nowMs);
   let runs;
   try {
-    const res = await api2.get("/host/workflow-runs/unreconciled");
-    runs = (res?.runs ?? []).filter(
+    runs = (await unreconciledRunsForHost(api2, nowMs)).filter(
       (r) => r.agent_id === agentId && !!r.finished_at && nowMs - new Date(r.finished_at).getTime() >= SETTLE_MS
     );
   } catch (err) {
@@ -7564,6 +7583,12 @@ function createFetchGate(ttlMs) {
     }
   };
 }
+function forceGates(gates, agentId) {
+  for (const gate of gates) {
+    if (agentId === void 0) gate.forceAll();
+    else gate.forceNext(agentId);
+  }
+}
 
 // src/lib/conditional-fetch.ts
 function stillHeld(quoted, live) {
@@ -10585,6 +10610,47 @@ var BlockedPromptCostLedger = class {
   }
 };
 
+// src/lib/blocked-prompt-report-cadence.ts
+var BLOCKED_PROMPT_HEARTBEAT_MS = 5 * 6e4;
+function blockedPromptReportFingerprint(body) {
+  const agents = [...body.agents].map((a) => [
+    a.agent_id,
+    a.state,
+    a.variant ?? "",
+    a.unknown_reason ?? "",
+    a.recovering === true,
+    a.tasks_deferred ?? 0,
+    a.messages_withheld ?? 0
+  ]).sort((x, y) => String(x[0]) < String(y[0]) ? -1 : String(x[0]) > String(y[0]) ? 1 : 0);
+  return JSON.stringify({ agents, complete: body.sweep_complete, omitted: body.omitted_no_session });
+}
+var BlockedPromptReportCadence = class {
+  constructor(heartbeatMs = BLOCKED_PROMPT_HEARTBEAT_MS) {
+    this.heartbeatMs = heartbeatMs;
+  }
+  lastDelivered = null;
+  /** Should this poll's report go out? `send` false only for `unchanged`. */
+  decide(body, now, opts) {
+    const fingerprint3 = blockedPromptReportFingerprint(body);
+    const last = this.lastDelivered;
+    let reason;
+    if (opts.hasUnreportedRecovery) reason = "recovery-riding";
+    else if (!last) reason = "first";
+    else if (last.fingerprint !== fingerprint3) reason = "changed";
+    else if (now - last.at >= this.heartbeatMs) reason = "heartbeat";
+    else reason = "unchanged";
+    return { send: reason !== "unchanged", reason, fingerprint: fingerprint3 };
+  }
+  /**
+   * Record what happened to a report that went out. Only a confirmed delivery
+   * advances the baseline; anything else clears it, so the next poll sends
+   * whatever it holds — the self-heal for a lost request.
+   */
+  recordOutcome(fingerprint3, now, delivered) {
+    this.lastDelivered = delivered ? { fingerprint: fingerprint3, at: now } : null;
+  }
+};
+
 // src/lib/occupancy-gate.ts
 import { closeSync as closeSync3, openSync as openSync3, readdirSync as readdirSync10, readSync as readSync3, statSync as statSync12 } from "fs";
 import { join as join39 } from "path";
@@ -14215,6 +14281,7 @@ var pendingSessionRestarts = /* @__PURE__ */ new Map();
 var lastRestartWasExternal = /* @__PURE__ */ new Map();
 var blockedPromptCost = new BlockedPromptCostLedger();
 var pendingRecoveryCost = /* @__PURE__ */ new Map();
+var blockedPromptReportCadence = new BlockedPromptReportCadence();
 var computerUseTunnels = new TunnelWatcher();
 var pendingRestartVerifications = /* @__PURE__ */ new Map();
 var agentCreatedAtMs = /* @__PURE__ */ new Map();
@@ -15807,7 +15874,7 @@ var agentRestartTimezoneInputs = /* @__PURE__ */ new Map();
 var lastVersionCheckAt = 0;
 var VERSION_CHECK_INTERVAL_MS = 5 * 60 * 1e3;
 var lastResponsivenessProbeAt = 0;
-var agtCliVersion = true ? "0.28.1158" : "dev";
+var agtCliVersion = true ? "0.28.1160" : "dev";
 function resolveBrewPath(execFileSync3) {
   try {
     const out = execFileSync3("which", ["brew"], { timeout: 5e3 }).toString().trim();
@@ -17568,14 +17635,17 @@ var pollCycleQuarantinedFetches = 0;
 var pollCycleQuarantinedSkips = 0;
 var pollCycleMemoriesUnchanged = 0;
 var pollCycleMemoriesFull = 0;
-function forceRefreshFetch() {
-  refreshFetchGate.forceAll();
-  memoriesFetchGate.forceAll();
-  secretsFetchGate.forceAll();
-  defaultSchedulesFetchGate.forceAll();
-  integrationsFetchGate.forceAll();
-  quarantinedFetchGate.forceAll();
-  forceNextExchange();
+var DESIRED_STATE_FETCH_GATES = [
+  refreshFetchGate,
+  memoriesFetchGate,
+  secretsFetchGate,
+  defaultSchedulesFetchGate,
+  integrationsFetchGate,
+  quarantinedFetchGate
+];
+function forceRefreshFetch(agentId) {
+  forceGates(DESIRED_STATE_FETCH_GATES, agentId);
+  if (agentId === void 0) forceNextExchange();
 }
 var consecutivePollFailures = 0;
 var consecutiveEmptyAgentPolls = 0;
@@ -18912,20 +18982,28 @@ async function pollCycleInner() {
           `[blocked-prompt-monitor] NOTHING REPORTED: all ${omittedNoSession} active claude-code agent(s) had no live session this poll, so no batch was sent and this host is unassessed. A wedged tmux now reads as 'unknown' rather than 'dead', so reaching here means either a genuine host-wide teardown or no tmux server at all (ENG-10825)`
         );
       }
-      if (sweeps.length > 0) {
+      const reportBody = {
+        agents: sweeps,
+        sweep_complete: sweepComplete,
+        // WHY the sweep was partial, not just that it was (obi-wan
+        // `82a0dd7ef253`). `sweep_complete: false` tells the route it may
+        // not close an alert; it does not tell an operator that four agents
+        // on this host have no session and that is why their critical will
+        // not clear. The route writes this onto the alert payload.
+        omitted_no_session: omittedNoSession
+      };
+      const reportDecision = sweeps.length > 0 ? blockedPromptReportCadence.decide(reportBody, Date.now(), {
+        hasUnreportedRecovery: unreportedRecovery.size > 0
+      }) : null;
+      if (reportDecision?.send) {
+        let reportResolved = false;
         const disposition = await api.post(
           "/host/blocked-prompt-report",
-          {
-            agents: sweeps,
-            sweep_complete: sweepComplete,
-            // WHY the sweep was partial, not just that it was (obi-wan
-            // `82a0dd7ef253`). `sweep_complete: false` tells the route it may
-            // not close an alert; it does not tell an operator that four agents
-            // on this host have no session and that is why their critical will
-            // not clear. The route writes this onto the alert payload.
-            omitted_no_session: omittedNoSession
-          }
+          // ENG-11720: the SAME object the cadence fingerprinted, so what was
+          // compared and what was sent cannot drift apart.
+          reportBody
         ).then((res) => {
+          reportResolved = true;
           const dropped = res?.dropped_agent_ids ?? [];
           if (dropped.length > 0) {
             log(
@@ -18940,6 +19018,11 @@ async function pollCycleInner() {
           return "withheld";
         });
         settlePendingRecovery(pendingRecoveryCost, unreportedRecovery, disposition);
+        blockedPromptReportCadence.recordOutcome(
+          reportDecision.fingerprint,
+          Date.now(),
+          reportResolved && disposition !== "withheld"
+        );
         if (disposition === "withheld" && unreportedRecovery.size > 0) {
           log(
             `[blocked-prompt-monitor] holding ${unreportedRecovery.size} unreported recovery total(s) for the next poll \u2014 the route did not persist them (ENG-10825)`
@@ -23216,7 +23299,7 @@ async function handleRestartDoorbell(agentId, requestedAt, restartReason) {
     if (restartReasonStampsTiming(restartReason)) {
       noteRestartRequested(codeName, requestedAt);
     }
-    triggerEarlyPoll(`config-change reload for '${codeName}' (reason=${restartReason})`);
+    triggerEarlyPoll(`config-change reload for '${codeName}' (reason=${restartReason})`, agentId);
     return;
   }
   if (!hostFlagStore().getBoolean("restart-doorbell")) return;
@@ -23371,7 +23454,7 @@ function ensureRealtimeAssignStarted(agentStates) {
         refreshFetchGate.forceNext(payload.agent_id);
         memoriesFetchGate.forceNext(payload.agent_id);
         log(`[realtime] Agent ${payload.agent_id} assigned`);
-        triggerEarlyPoll(`agent ${payload.agent_id} assigned to this host`);
+        triggerEarlyPoll(`agent ${payload.agent_id} assigned to this host`, payload.agent_id);
         markAgentForFreshMemorySync(payload.agent_id);
       },
       onUnassign: (payload) => {
@@ -23503,7 +23586,7 @@ function ensureRealtimeIntegrationContextStarted(agentStates) {
             }
           }
         }
-        triggerEarlyPoll(`integration context changed for agent ${payload.agent_id}`);
+        triggerEarlyPoll(`integration context changed for agent ${payload.agent_id}`, payload.agent_id);
       },
       log
     });
@@ -23514,7 +23597,8 @@ function ensureRealtimeIntegrationContextStarted(agentStates) {
       agentIds: [...targetSet],
       onTokenRotated: (agentId, event) => {
         triggerEarlyPoll(
-          `integration token rotated for agent ${agentId} (integration ${event.integration_id}, ${event.definition_id})`
+          `integration token rotated for agent ${agentId} (integration ${event.integration_id}, ${event.definition_id})`,
+          agentId
         );
       },
       onSubscribed: (agentId) => {
@@ -23556,7 +23640,7 @@ function ensureRealtimeIntegrationContextStarted(agentStates) {
     log(`[realtime] Integration context subscription failed: ${err.message}`);
   });
 }
-function triggerEarlyPoll(reason) {
+function triggerEarlyPoll(reason, agentId) {
   if (!running) return;
   if (pollTimer) {
     clearTimeout(pollTimer);
@@ -23564,7 +23648,7 @@ function triggerEarlyPoll(reason) {
   }
   log(`[realtime] Triggering early poll: ${reason}`);
   consecutiveEmptyAgentPolls = 0;
-  forceRefreshFetch();
+  forceRefreshFetch(agentId);
   pollTimer = setTimeout(() => {
     void pollCycleAndReschedule(pollCycle, {
       onError: (err) => log(`[poll-loop] early-poll cycle rejected: ${err?.message ?? String(err)}`),