@integrity-labs/agt-cli 0.28.409 → 0.28.411

package/dist/{chunk-QEWIZCQC.js → chunk-JIT5ORMO.js}

@@ -25,7 +25,7 @@ import {
   resolveConnectivityProbe,
   worseConnectivityOutcome,
   wrapScheduledTaskPrompt
-} from "./chunk-YLLK4G3B.js";
+} from "./chunk-D4MDIZ3T.js";
 import {
   parsePsRows
 } from "./chunk-XWVM4KPK.js";
@@ -6247,7 +6247,7 @@ function requireHost() {
 }
 
 // src/lib/api-client.ts
-var agtCliVersion = true ? "0.28.409" : "dev";
+var agtCliVersion = true ? "0.28.411" : "dev";
 var lastConfigHash = null;
 function setConfigHash(hash) {
   lastConfigHash = hash && hash.length > 0 ? hash : null;
@@ -8751,4 +8751,4 @@ export {
   managerInstallSystemUnitCommand,
   managerUninstallSystemUnitCommand
 };
-//# sourceMappingURL=chunk-QEWIZCQC.js.map
+//# sourceMappingURL=chunk-JIT5ORMO.js.map

package/dist/{claude-pair-runtime-XGPENQB7.js → claude-pair-runtime-253N4REG.js}

@@ -100,7 +100,7 @@ async function spawnPairSession(session) {
     return { ok: true };
   } catch {
   }
-  const { resolveClaudeBinary } = await import("./persistent-session-W5QSIM7F.js");
+  const { resolveClaudeBinary } = await import("./persistent-session-YE2Q5LSK.js");
   const claudeBin = resolveClaudeBinary();
   const pairEnv = {
     ...process.env,
@@ -373,4 +373,4 @@ export {
   startClaudePair,
   submitClaudePairCode
 };
-//# sourceMappingURL=claude-pair-runtime-XGPENQB7.js.map
+//# sourceMappingURL=claude-pair-runtime-253N4REG.js.map

package/dist/lib/manager-worker.js

@@ -45,7 +45,7 @@ import {
   resolveEffectivePinRaw,
   safeWriteJsonAtomic,
   setConfigHash
-} from "../chunk-QEWIZCQC.js";
+} from "../chunk-JIT5ORMO.js";
 import {
   getProjectDir as getProjectDir2,
   getReadyTasks,
@@ -85,6 +85,7 @@ import {
   getFramework,
   getIntegration,
   getLastFailureContext,
+  getOpencodeSessionState,
   getProjectDir,
   getSessionState,
   hashFile,
@@ -147,7 +148,7 @@ import {
   toOpencodeModel,
   transcriptActivityAgeSeconds,
   writeEgressAllowlist
-} from "../chunk-YLLK4G3B.js";
+} from "../chunk-D4MDIZ3T.js";
 import {
   reapOrphanChannelMcps
 } from "../chunk-XWVM4KPK.js";
@@ -1588,6 +1589,10 @@ function decideMaintenanceWindowGate(opts) {
 function shouldConsumeForcedUpdate(outcome) {
   return outcome !== "failed";
 }
+var FORCED_UPDATE_DEFER_ESCALATE_AFTER = 20;
+function shouldRelaxForcedUpdateIdleGate(deferStreak) {
+  return deferStreak >= FORCED_UPDATE_DEFER_ESCALATE_AFTER;
+}
 function decideForcedUpdate(opts) {
   const { requestedAt, lastProcessedAt, hostBusy } = opts;
   if (!requestedAt) return "none";
@@ -8931,14 +8936,20 @@ function restartGateFor(codeName, reason) {
     now: /* @__PURE__ */ new Date()
   });
 }
-function isHostBusyForForcedUpdate() {
+function isHostBusyForForcedUpdate(opts) {
   const now = /* @__PURE__ */ new Date();
   for (const agent of state6.agents) {
     const decision = decideRestartGate({
       window: null,
       paneLogAgeSeconds: paneLogAgeSecondsFor(agent.codeName),
       transcriptAgeSeconds: transcriptAgeSecondsFor(agent.codeName),
-      inboundAgeSeconds: inboundAgeSecondsFor(agent.codeName),
+      // ENG-8079: past the defer-streak threshold, drop the inbound-quiet axis
+      // and gate on progress alone. Passing null means "no signal" for that axis
+      // (see decideRestartGate), so only transcript/pane progress governs — an
+      // agent mid-turn still defers, an agent that merely received a message in
+      // the last 5 minutes no longer does. On a 14-agent host the inbound axis
+      // is what made the gate un-satisfiable in practice.
+      inboundAgeSeconds: opts?.relaxed ? null : inboundAgeSecondsFor(agent.codeName),
       now,
       env: {}
     });
@@ -8946,6 +8957,66 @@ function isHostBusyForForcedUpdate() {
   }
   return false;
 }
+function runPendingForcedUpdate() {
+  const requestedUpdateAt = cachedUpdateRequestedAt;
+  const relaxed = shouldRelaxForcedUpdateIdleGate(forcedUpdateDeferStreak);
+  const forcedUpdate = decideForcedUpdate({
+    requestedAt: requestedUpdateAt,
+    lastProcessedAt: state6.lastUpdateRequestProcessedAt ?? null,
+    hostBusy: isHostBusyForForcedUpdate({ relaxed })
+  });
+  if (forcedUpdate !== "defer-idle") forcedUpdateDeferStreak = 0;
+  if (forcedUpdate === "none" || forcedUpdate === "already-handled") return false;
+  if (forcedUpdate === "defer-idle") {
+    forcedUpdateDeferStreak += 1;
+    if (forcedUpdateDeferStreak === 1 || forcedUpdateDeferStreak % 10 === 0) {
+      log(
+        `[self-update] "Update CLI now" requested at ${requestedUpdateAt} \u2014 deferring: an agent is mid-turn / recently active (consecutive defers: ${forcedUpdateDeferStreak}). Retrying next poll.`
+      );
+    }
+    if (forcedUpdateDeferStreak === FORCED_UPDATE_DEFER_ESCALATE_AFTER) {
+      log(
+        `[self-update] WARN "Update CLI now" has deferred ${forcedUpdateDeferStreak} consecutive polls waiting for an idle host. Relaxing the idle gate to the progress axis only (an agent actively producing tokens still defers; "received a message in the last 5 minutes" no longer does).`
+      );
+    }
+    return false;
+  }
+  if (selfUpdateInFlight) {
+    log(
+      `[self-update] "Update CLI now" requested at ${requestedUpdateAt} \u2014 deferring: a self-update is already in flight. Retrying next poll.`
+    );
+    return false;
+  }
+  log(
+    `[self-update] "Update CLI now" requested at ${requestedUpdateAt} \u2014 running window-bypassing self-update now (host idle${relaxed ? ", relaxed gate" : ""}).`
+  );
+  const prevProcessedAt = state6.lastUpdateRequestProcessedAt ?? null;
+  void checkAndUpdateCli({ force: true }).then((outcome) => {
+    if (!shouldConsumeForcedUpdate(outcome)) {
+      log(
+        `[self-update] "Update CLI now" did not complete (no install ran) \u2014 leaving the request pending; the next idle poll retries.`
+      );
+      return;
+    }
+    log(
+      outcome === "updated" ? `[self-update] "Update CLI now" completed \u2014 upgrade installed; manager restart scheduled.` : `[self-update] "Update CLI now" completed \u2014 host was already on the latest in-channel build; nothing to install.`
+    );
+    state6.lastUpdateRequestProcessedAt = requestedUpdateAt;
+    try {
+      atomicWriteFileSync(getStateFile(), JSON.stringify(state6, null, 2));
+    } catch (err) {
+      state6.lastUpdateRequestProcessedAt = prevProcessedAt;
+      log(
+        `[self-update] failed to persist update-request ack; retrying next poll: ${err.message}`
+      );
+    }
+  }).catch((err) => {
+    log(
+      `[self-update] forced check failed: ${err.message} \u2014 leaving the request pending for retry.`
+    );
+  });
+  return true;
+}
 var runningMcpHashes = /* @__PURE__ */ new Map();
 var anchorSessionLifecycle = new AnchorSessionLifecycle();
 var runningMcpServerKeys = /* @__PURE__ */ new Map();
@@ -9307,7 +9378,7 @@ var agentRestartTimezoneInputs = /* @__PURE__ */ new Map();
 var lastVersionCheckAt = 0;
 var VERSION_CHECK_INTERVAL_MS = 5 * 60 * 1e3;
 var lastResponsivenessProbeAt = 0;
-var agtCliVersion = true ? "0.28.409" : "dev";
+var agtCliVersion = true ? "0.28.411" : "dev";
 function resolveBrewPath(execFileSync2) {
   try {
     const out = execFileSync2("which", ["brew"], { timeout: 5e3 }).toString().trim();
@@ -9776,6 +9847,8 @@ var selfUpdatePinInvalidLogged = false;
 var restartAfterUpgrade = false;
 var pendingUpgradeVersion = null;
 var selfUpdateInFlight = false;
+var cachedUpdateRequestedAt = null;
+var forcedUpdateDeferStreak = 0;
 async function checkAndUpdateCli(opts) {
   const force = opts?.force ?? false;
   if (selfUpdateInFlight) return "failed";
@@ -9954,27 +10027,41 @@ async function checkAndUpdateCliViaNpm(force = false) {
     );
     return await installAgtCliViaNpm(pinDecision.version, `pin=${pinDecision.version}`);
   }
-  let latest;
-  try {
-    const res = await fetch(
-      `https://registry.npmjs.org/@integrity-labs/agt-cli/${encodeURIComponent(channel)}`,
-      {
-        signal: AbortSignal.timeout(1e4),
-        headers: { Accept: "application/json" }
+  const REGISTRY_TIMEOUT_MS = 1e4;
+  const registryAttempts = force ? 2 : 1;
+  let latest = null;
+  let lastFetchErr = "";
+  for (let attempt = 1; attempt <= registryAttempts; attempt++) {
+    try {
+      const res = await fetch(
+        `https://registry.npmjs.org/@integrity-labs/agt-cli/${encodeURIComponent(channel)}`,
+        {
+          signal: AbortSignal.timeout(REGISTRY_TIMEOUT_MS),
+          headers: { Accept: "application/json" }
+        }
+      );
+      if (!res.ok) {
+        log(`[self-update] npm registry returned ${res.status} (channel=${channel})`);
+        return "failed";
+      }
+      const body = await res.json();
+      if (!body.version) {
+        log(`[self-update] npm registry response missing version field`);
+        return "failed";
+      }
+      latest = body.version;
+      break;
+    } catch (err) {
+      lastFetchErr = err.message;
+      if (attempt < registryAttempts) {
+        log(
+          `[self-update] npm registry fetch failed (attempt ${attempt}/${registryAttempts}): ${lastFetchErr} \u2014 retrying.`
+        );
       }
-    );
-    if (!res.ok) {
-      log(`[self-update] npm registry returned ${res.status} (channel=${channel})`);
-      return "failed";
-    }
-    const body = await res.json();
-    if (!body.version) {
-      log(`[self-update] npm registry response missing version field`);
-      return "failed";
     }
-    latest = body.version;
-  } catch (err) {
-    log(`[self-update] npm registry fetch failed: ${err.message}`);
+  }
+  if (latest === null) {
+    log(`[self-update] npm registry fetch failed: ${lastFetchErr}`);
     return "failed";
   }
   const urgentTarget = await fetchUrgentDistTagVersion();
@@ -10439,7 +10526,7 @@ function flushRestartedAgentDiagnostics(hostId, codeNames) {
   if (codeNames.length === 0) return;
   void (async () => {
     try {
-      const { collectDiagnostics } = await import("../persistent-session-W5QSIM7F.js");
+      const { collectDiagnostics } = await import("../persistent-session-YE2Q5LSK.js");
       await api.post("/host/heartbeat", {
         host_id: hostId,
         agent_diagnostics: collectDiagnostics(codeNames, quarantineEntriesFor)
@@ -10515,7 +10602,15 @@ async function pollCycle() {
   } catch (err) {
     log(`[restart-handler] processRestartFlags threw: ${err.message}`);
   }
-  checkAndUpdateCli().catch((err) => log(`[self-update] Check failed: ${err.message}`));
+  let forcedUpdateKicked = false;
+  try {
+    forcedUpdateKicked = runPendingForcedUpdate();
+  } catch (err) {
+    log(`[self-update] runPendingForcedUpdate threw: ${err.message}`);
+  }
+  if (!forcedUpdateKicked) {
+    checkAndUpdateCli().catch((err) => log(`[self-update] Check failed: ${err.message}`));
+  }
   maybeUpgradeClaudeCode().catch((err) => log(`[claude-code-upgrade] Check failed: ${err.message}`));
   try {
     registeredAgentsCache.clear();
@@ -10538,7 +10633,7 @@ async function pollCycle() {
     }
     try {
       const { detectHostSecurity } = await import("../host-security-6PDFG7F5.js");
-      const { collectDiagnostics } = await import("../persistent-session-W5QSIM7F.js");
+      const { collectDiagnostics } = await import("../persistent-session-YE2Q5LSK.js");
       const diagCodeNames = [...agentState.persistentSessionAgents];
       const agentDiagnostics = diagCodeNames.length > 0 ? collectDiagnostics(diagCodeNames, quarantineEntriesFor) : void 0;
       let tailscaleHostname;
@@ -10625,49 +10720,7 @@ async function pollCycle() {
           cachedDesiredAgtCliVersion = hbResp.desired_agt_cli_version ?? null;
         }
       }
-      const requestedUpdateAt = hbResp?.update_requested_at ?? null;
-      const forcedUpdate = decideForcedUpdate({
-        requestedAt: requestedUpdateAt,
-        lastProcessedAt: state6.lastUpdateRequestProcessedAt ?? null,
-        hostBusy: isHostBusyForForcedUpdate()
-      });
-      if (forcedUpdate === "defer-idle") {
-        log(
-          `[self-update] "Update CLI now" requested at ${requestedUpdateAt} \u2014 deferring: an agent is mid-turn / recently active. Retrying next poll.`
-        );
-      } else if (forcedUpdate === "consume") {
-        if (selfUpdateInFlight) {
-          log(
-            `[self-update] "Update CLI now" requested at ${requestedUpdateAt} \u2014 deferring: a self-update is already in flight. Retrying next poll.`
-          );
-        } else {
-          log(
-            `[self-update] "Update CLI now" requested at ${requestedUpdateAt} \u2014 running window-bypassing self-update now (host idle).`
-          );
-          const prevProcessedAt = state6.lastUpdateRequestProcessedAt ?? null;
-          void checkAndUpdateCli({ force: true }).then((outcome) => {
-            if (!shouldConsumeForcedUpdate(outcome)) {
-              log(
-                `[self-update] "Update CLI now" did not complete (no install ran) \u2014 leaving the request pending; the next idle heartbeat retries.`
-              );
-              return;
-            }
-            state6.lastUpdateRequestProcessedAt = requestedUpdateAt;
-            try {
-              atomicWriteFileSync(getStateFile(), JSON.stringify(state6, null, 2));
-            } catch (err) {
-              state6.lastUpdateRequestProcessedAt = prevProcessedAt;
-              log(
-                `[self-update] failed to persist update-request ack; retrying next heartbeat: ${err.message}`
-              );
-            }
-          }).catch((err) => {
-            log(
-              `[self-update] forced check failed: ${err.message} \u2014 leaving the request pending for retry.`
-            );
-          });
-        }
-      }
+      cachedUpdateRequestedAt = hbResp?.update_requested_at ?? null;
       if (hbResp?.feature_flags) {
         hostFlagStore().applyHeartbeat(hbResp.feature_flags, hbResp.flags_schema_version);
       }
@@ -10693,7 +10746,7 @@ async function pollCycle() {
       const {
         collectResponsivenessProbes,
         getResponsivenessIntervalMs
-      } = await import("../responsiveness-probe-B7XCQQN4.js");
+      } = await import("../responsiveness-probe-QO4EHFCH.js");
       const probeIntervalMs = getResponsivenessIntervalMs();
       if (now - lastResponsivenessProbeAt > probeIntervalMs) {
         const probeCodeNames = [...agentState.persistentSessionAgents];
@@ -10725,7 +10778,7 @@ async function pollCycle() {
           collectResponsivenessProbes,
           livePendingInboundOldestAgeSeconds,
           parkPendingInbound
-        } = await import("../responsiveness-probe-B7XCQQN4.js");
+        } = await import("../responsiveness-probe-QO4EHFCH.js");
         const { getProjectDir: wedgeProjectDir } = await import("../scheduler-engine-HYOOQV4A.js");
         const wedgeNow = /* @__PURE__ */ new Date();
         const liveAgents = agentState.persistentSessionAgents;
@@ -13445,6 +13498,7 @@ async function ensureOpencodeRuntime(agent, refreshData, agentTimezone) {
   };
   if (openRouterRaw?.auth_token) serveEnv["OPENROUTER_API_KEY"] = openRouterRaw.auth_token;
   if (anthropicRaw?.api_key) serveEnv["ANTHROPIC_API_KEY"] = anthropicRaw.api_key;
+  const startedAtBefore = getOpencodeSessionState(codeName)?.startedAt ?? null;
   const session = await startOpencodeSession({
     codeName,
     agentId: agent.agent_id,
@@ -13458,18 +13512,19 @@ async function ensureOpencodeRuntime(agent, refreshData, agentTimezone) {
     log
   });
   agentState.persistentSessionAgents.add(codeName);
+  const spawnAttempted = session.startedAt !== startedAtBefore;
   if (session.status === "crashed") {
     const tail = session.lastFailureTail ?? readOpencodePaneLogTail(codeName);
     return {
       decision: "spawn",
-      spawnAttempted: true,
+      spawnAttempted,
       sessionHealthyAfter: isOpencodeSessionHealthy(codeName),
       detail: tail ? `opencode spawn failed: ${tail.slice(0, 200)}` : "opencode spawn failed"
     };
   }
   return {
     decision: "spawn",
-    spawnAttempted: true,
+    spawnAttempted,
     sessionHealthyAfter: isOpencodeSessionHealthy(codeName),
     detail: `opencode serve on 127.0.0.1:${session.port ?? "?"}`
   };
@@ -14027,7 +14082,7 @@ 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-W5QSIM7F.js");
+        const { collectDiagnostics } = await import("../persistent-session-YE2Q5LSK.js");
         await api.post("/host/heartbeat", {
           host_id: hostId,
           agent_diagnostics: collectDiagnostics([codeName], quarantineEntriesFor)
@@ -14077,7 +14132,7 @@ async function respawnAgentAfterMcpStop(codeName, reason) {
     }
     try {
       const hostId = await getHostId();
-      const { collectDiagnostics } = await import("../persistent-session-W5QSIM7F.js");
+      const { collectDiagnostics } = await import("../persistent-session-YE2Q5LSK.js");
       await api.post("/host/heartbeat", {
         host_id: hostId,
         agent_diagnostics: collectDiagnostics([codeName], quarantineEntriesFor)
@@ -14530,7 +14585,7 @@ async function processClaudePairSessions(agents) {
     killPairSession,
     pairTmuxSession,
     finalizeClaudePairOnboarding
-  } = await import("../claude-pair-runtime-XGPENQB7.js");
+  } = await import("../claude-pair-runtime-253N4REG.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));