@integrity-labs/agt-cli 0.28.384 → 0.28.386

package/dist/{claude-pair-runtime-OIK6HKPS.js → claude-pair-runtime-NVMVUAGS.js}

@@ -100,7 +100,7 @@ async function spawnPairSession(session) {
     return { ok: true };
   } catch {
   }
-  const { resolveClaudeBinary } = await import("./persistent-session-MDQVHXKX.js");
+  const { resolveClaudeBinary } = await import("./persistent-session-KQHWYEQF.js");
   const claudeBin = resolveClaudeBinary();
   const pairEnv = {
     ...process.env,
@@ -373,4 +373,4 @@ export {
   startClaudePair,
   submitClaudePairCode
 };
-//# sourceMappingURL=claude-pair-runtime-OIK6HKPS.js.map
+//# sourceMappingURL=claude-pair-runtime-NVMVUAGS.js.map

package/dist/lib/manager-worker.js

@@ -43,7 +43,7 @@ import {
   resolveEffectivePinRaw,
   safeWriteJsonAtomic,
   setConfigHash
-} from "../chunk-IXN2AFOX.js";
+} from "../chunk-O37EMHVE.js";
 import {
   getProjectDir as getProjectDir2,
   getReadyTasks,
@@ -145,7 +145,7 @@ import {
   toOpencodeModel,
   transcriptActivityAgeSeconds,
   writeEgressAllowlist
-} from "../chunk-MQHDPJZD.js";
+} from "../chunk-LAO2CUZU.js";
 import {
   reapOrphanChannelMcps
 } from "../chunk-XWVM4KPK.js";
@@ -725,6 +725,24 @@ var ChannelQuarantineStore = class {
     const forAgent = this.load()[codeName];
     return new Set(forAgent ? Object.keys(forAgent) : []);
   }
+  /**
+   * ENG-8018: the full quarantine entries for `codeName` (serverKey → {reason,
+   * quarantinedAt}), empty array if none. Unlike {@link getQuarantinedKeys} this
+   * carries the reason + timestamp needed to surface the quarantine to the
+   * control plane (the manager folds it into the agent heartbeat's diagnostics)
+   * and to the agent itself, so an operator sees WHY/WHEN on the webapp badge
+   * and the agent reports the real cause instead of confabulating one (ENG-8003).
+   * Returns a defensive copy so callers can't mutate the cached store.
+   */
+  getQuarantineEntries(codeName) {
+    const forAgent = this.load()[codeName];
+    if (!forAgent) return [];
+    return Object.entries(forAgent).map(([serverKey, entry]) => ({
+      serverKey,
+      reason: entry.reason,
+      quarantinedAt: entry.quarantinedAt
+    }));
+  }
   /**
    * Mark `serverKey` quarantined for `codeName`. Idempotent — re-quarantining
    * an already-quarantined key preserves the original `quarantinedAt` (so the
@@ -8207,6 +8225,7 @@ function hasRevokedResiduals(state7) {
   return state7.gatewayRunning || state7.portAllocated || state7.provisionDirExists;
 }
 var pendingSessionRestarts = /* @__PURE__ */ new Map();
+var lastRestartWasExternal = /* @__PURE__ */ new Map();
 var pendingRestartVerifications = /* @__PURE__ */ new Map();
 var restartBreaker = new RestartBreaker();
 var reportedTrips = /* @__PURE__ */ new Map();
@@ -8907,6 +8926,7 @@ function stopPersistentSessionAndForgetMcpBaseline(codeName, breakerReason, gate
   }
   cancelPendingSessionRestart(codeName);
   stopPersistentSession(codeName, log);
+  lastRestartWasExternal.set(codeName, gateReason !== "mcp-presence-reaper");
   runningMcpHashes.delete(codeName);
   runningMcpServerKeys.delete(codeName);
   runningChannelSecretHashes.delete(codeName);
@@ -9093,7 +9113,7 @@ var agentRestartTimezoneInputs = /* @__PURE__ */ new Map();
 var lastVersionCheckAt = 0;
 var VERSION_CHECK_INTERVAL_MS = 5 * 60 * 1e3;
 var lastResponsivenessProbeAt = 0;
-var agtCliVersion = true ? "0.28.384" : "dev";
+var agtCliVersion = true ? "0.28.386" : "dev";
 function resolveBrewPath(execFileSync2) {
   try {
     const out = execFileSync2("which", ["brew"], { timeout: 5e3 }).toString().trim();
@@ -10083,6 +10103,9 @@ function channelQuarantineStore() {
   }
   return _channelQuarantineStore;
 }
+function quarantineEntriesFor(codeName) {
+  return channelQuarantineStore().getQuarantineEntries(codeName);
+}
 var _hostFlagStore = null;
 function hostFlagStore() {
   if (!_hostFlagStore) {
@@ -10222,10 +10245,10 @@ function flushRestartedAgentDiagnostics(hostId, codeNames) {
   if (codeNames.length === 0) return;
   void (async () => {
     try {
-      const { collectDiagnostics } = await import("../persistent-session-MDQVHXKX.js");
+      const { collectDiagnostics } = await import("../persistent-session-KQHWYEQF.js");
       await api.post("/host/heartbeat", {
         host_id: hostId,
-        agent_diagnostics: collectDiagnostics(codeNames)
+        agent_diagnostics: collectDiagnostics(codeNames, quarantineEntriesFor)
       });
     } catch (err) {
       log(`[restart] post-respawn diagnostics flush failed: ${err.message}`);
@@ -10321,9 +10344,9 @@ async function pollCycle() {
     }
     try {
       const { detectHostSecurity } = await import("../host-security-6PDFG7F5.js");
-      const { collectDiagnostics } = await import("../persistent-session-MDQVHXKX.js");
+      const { collectDiagnostics } = await import("../persistent-session-KQHWYEQF.js");
       const diagCodeNames = [...agentState.persistentSessionAgents];
-      const agentDiagnostics = diagCodeNames.length > 0 ? collectDiagnostics(diagCodeNames) : void 0;
+      const agentDiagnostics = diagCodeNames.length > 0 ? collectDiagnostics(diagCodeNames, quarantineEntriesFor) : void 0;
       let tailscaleHostname;
       try {
         const { execSync: es } = await import("child_process");
@@ -10476,7 +10499,7 @@ async function pollCycle() {
       const {
         collectResponsivenessProbes,
         getResponsivenessIntervalMs
-      } = await import("../responsiveness-probe-CG7XM7GL.js");
+      } = await import("../responsiveness-probe-5X6WVKUZ.js");
       const probeIntervalMs = getResponsivenessIntervalMs();
       if (now - lastResponsivenessProbeAt > probeIntervalMs) {
         const probeCodeNames = [...agentState.persistentSessionAgents];
@@ -10508,7 +10531,7 @@ async function pollCycle() {
           collectResponsivenessProbes,
           livePendingInboundOldestAgeSeconds,
           parkPendingInbound
-        } = await import("../responsiveness-probe-CG7XM7GL.js");
+        } = await import("../responsiveness-probe-5X6WVKUZ.js");
         const { getProjectDir: wedgeProjectDir } = await import("../scheduler-engine-HYOOQV4A.js");
         const wedgeNow = /* @__PURE__ */ new Date();
         const liveAgents = agentState.persistentSessionAgents;
@@ -12723,6 +12746,14 @@ async function processAgent(agent, agentStates) {
           // but-not-yet-live MCP, so don't let the reaper re-fire + inflate the
           // McpReaperRestartsHourly metric into a false flapping alarm.
           restartPending: pendingSessionRestarts.has(agent.code_name),
+          // ENG-8013 (storm-awareness): tell the reaper whether THIS session
+          // generation was born from an external respawn (integration-change /
+          // channel-set-change / day-rollover / wedge / operator restart, or a
+          // storm) vs the reaper's own recovery restart. When external, a missing
+          // channel MCP is churn collateral and must not accrue quarantine budget,
+          // the durable fix for the ENG-8003 storm-quarantine incident. Absent map
+          // entry (first boot) → false → count normally.
+          currentGenerationExternallyStarted: lastRestartWasExternal.get(agent.code_name) ?? false,
           // ENG-5441 / ENG-5547: a mcp-presence-reaper restart counts against
           // the agent-wide breaker ONLY when several distinct MCP servers are
           // missing at once — a thrashing collection that rotates restarts of
@@ -12844,6 +12875,14 @@ async function processAgent(agent, agentStates) {
                 "quarantine",
                 `optional channel '${serverKey}' dead past restart budget + dwell \u2014 quarantined`
               );
+              void injectMessage(
+                codeName,
+                "system",
+                `Your "${serverKey}" channel has been quarantined by the host runtime: it repeatedly failed to start (no live MCP process) past the recovery budget, so it has been pulled from your active channel set. You cannot send or receive on ${serverKey} until an operator re-enables it. This is an infrastructure fault, not a policy or allowlist decision. If anyone asks why you are quiet on ${serverKey}, tell them the channel is quarantined and ask them to have an operator re-enable it. Do not guess at other causes.`,
+                { task_name: "channel-quarantine" },
+                log
+              ).catch(() => {
+              });
             }
           }
         });
@@ -13703,10 +13742,10 @@ async function handleRestartDoorbell(agentId, requestedAt, restartReason) {
     void api.post("/host/restart-ack", { host_id: hostId, agent_id: agentId, restart_requested_at: requestedAt }).catch((err) => log(`[restart-lane] ack failed for '${codeName}': ${err.message}`));
     void (async () => {
       try {
-        const { collectDiagnostics } = await import("../persistent-session-MDQVHXKX.js");
+        const { collectDiagnostics } = await import("../persistent-session-KQHWYEQF.js");
         await api.post("/host/heartbeat", {
           host_id: hostId,
-          agent_diagnostics: collectDiagnostics([codeName])
+          agent_diagnostics: collectDiagnostics([codeName], quarantineEntriesFor)
         });
       } catch (err) {
         log(`[restart-lane] post-respawn diagnostics flush failed for '${codeName}': ${err.message}`);
@@ -13753,10 +13792,10 @@ async function respawnAgentAfterMcpStop(codeName, reason) {
     }
     try {
       const hostId = await getHostId();
-      const { collectDiagnostics } = await import("../persistent-session-MDQVHXKX.js");
+      const { collectDiagnostics } = await import("../persistent-session-KQHWYEQF.js");
       await api.post("/host/heartbeat", {
         host_id: hostId,
-        agent_diagnostics: collectDiagnostics([codeName])
+        agent_diagnostics: collectDiagnostics([codeName], quarantineEntriesFor)
       });
     } catch (err) {
       log(`[fast-mcp-respawn] post-respawn diagnostics flush failed for '${codeName}': ${err.message}`);
@@ -14206,7 +14245,7 @@ async function processClaudePairSessions(agents) {
     killPairSession,
     pairTmuxSession,
     finalizeClaudePairOnboarding
-  } = await import("../claude-pair-runtime-OIK6HKPS.js");
+  } = await import("../claude-pair-runtime-NVMVUAGS.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));