@vtxmacro/cli 2026.9.22 → 2026.9.23

package/bin/vtx-service-bootstrap.js

@@ -16,7 +16,7 @@ import { fileURLToPath } from "node:url";
 // agent-cli-release.json
 var agent_cli_release_default = {
   package_name: "@vtxmacro/cli",
-  package_version: "2026.9.22",
+  package_version: "2026.9.23",
   codex_package_name: "@openai/codex",
   codex_version: "0.153.3",
   copilot_sdk_package_name: "@github/copilot-sdk",

package/bin/vtx.js

@@ -69,7 +69,7 @@ var init_agent_cli_release = __esm({
   "agent-cli-release.json"() {
     agent_cli_release_default = {
       package_name: "@vtxmacro/cli",
-      package_version: "2026.9.22",
+      package_version: "2026.9.23",
       codex_package_name: "@openai/codex",
       codex_version: "0.153.3",
       copilot_sdk_package_name: "@github/copilot-sdk",
@@ -36140,6 +36140,7 @@ var init_runner = __esm({
         const agentAbort = new AbortController();
         let agentLoop = Promise.resolve();
         let modelCatalogRefresh = null;
+        let providerRateLimitRefresh = null;
         let wakeDrain;
         const drainSignal = new Promise((resolve10) => {
           wakeDrain = resolve10;
@@ -37272,68 +37273,72 @@ var init_runner = __esm({
               providerCooldownReason = null;
               providerQuotaSnapshot = null;
             }
-            if (this.dependencies.codexAdapter.readAccountRateLimits && currentTime >= nextProviderRateLimitRefreshAt) {
-              try {
-                const refreshVersion = providerCooldownVersion;
-                const refreshStartedAt = now();
-                const refreshedRateLimits = await this.dependencies.codexAdapter.readAccountRateLimits(
-                  Date.now() + Math.min(2e3, settings.hostHeartbeatMs),
-                  this.options.signal
-                );
-                if (refreshVersion !== providerCooldownVersion) continue;
-                providerRateLimits = refreshedRateLimits;
-                providerRateLimitsObservedAtMs = now();
-                const refreshedAt = now();
-                const usableSnapshot = usableProviderQuotaSnapshot(
-                  providerRateLimits,
-                  knownProviderQuotaSnapshot,
-                  refreshedAt
-                );
-                providerQuotaRecoveryObservedAtMs = usableSnapshot ? refreshStartedAt : null;
-                retainKnownQuotaWindows(providerRateLimits.rateLimits);
-                const refreshedRetryAt = codexAccountRateLimitReached(providerRateLimits) ? codexRateLimitRetryAtMs(providerRateLimits) ?? refreshedAt + DEFAULT_CODEX_ACCOUNT_COOLDOWN_MS : null;
-                if (refreshedRetryAt !== null && refreshedRetryAt > refreshedAt) {
-                  providerRetryAtMs = Math.max(providerRetryAtMs ?? 0, refreshedRetryAt);
-                  providerCooldownReason = "provider_limit_snapshot";
-                  retainQuotaSnapshot(providerRateLimits);
-                } else if (providerRetryAtMs !== null && !codexAccountRateLimitReached(providerRateLimits) && providerRateLimits.rateLimits.limitId === providerQuotaSnapshot?.limitId && [providerRateLimits.rateLimits.primary, providerRateLimits.rateLimits.secondary].some((window2) => window2 !== null) && ["primary", "secondary"].every((key) => {
-                  const window2 = refreshedRateLimits.rateLimits[key];
-                  const previous = providerQuotaSnapshot?.[key];
-                  if (previous && (!window2 || window2.windowDurationMins !== previous.windowDurationMins)) {
-                    return false;
+            if (this.dependencies.codexAdapter.readAccountRateLimits && currentTime >= nextProviderRateLimitRefreshAt && !providerRateLimitRefresh) {
+              nextProviderRateLimitRefreshAt = Infinity;
+              providerRateLimitRefresh = Promise.resolve().then(async () => {
+                try {
+                  const refreshVersion = providerCooldownVersion;
+                  const refreshStartedAt = now();
+                  const refreshedRateLimits = await this.dependencies.codexAdapter.readAccountRateLimits(
+                    Date.now() + Math.min(2e3, settings.hostHeartbeatMs),
+                    agentAbort.signal
+                  );
+                  if (agentAbort.signal.aborted || refreshVersion !== providerCooldownVersion) return;
+                  providerRateLimits = refreshedRateLimits;
+                  providerRateLimitsObservedAtMs = now();
+                  const refreshedAt = now();
+                  const usableSnapshot = usableProviderQuotaSnapshot(
+                    providerRateLimits,
+                    knownProviderQuotaSnapshot,
+                    refreshedAt
+                  );
+                  providerQuotaRecoveryObservedAtMs = usableSnapshot ? refreshStartedAt : null;
+                  retainKnownQuotaWindows(providerRateLimits.rateLimits);
+                  const refreshedRetryAt = codexAccountRateLimitReached(providerRateLimits) ? codexRateLimitRetryAtMs(providerRateLimits) ?? refreshedAt + DEFAULT_CODEX_ACCOUNT_COOLDOWN_MS : null;
+                  if (refreshedRetryAt !== null && refreshedRetryAt > refreshedAt) {
+                    providerRetryAtMs = Math.max(providerRetryAtMs ?? 0, refreshedRetryAt);
+                    providerCooldownReason = "provider_limit_snapshot";
+                    retainQuotaSnapshot(providerRateLimits);
+                  } else if (providerRetryAtMs !== null && !codexAccountRateLimitReached(providerRateLimits) && providerRateLimits.rateLimits.limitId === providerQuotaSnapshot?.limitId && [providerRateLimits.rateLimits.primary, providerRateLimits.rateLimits.secondary].some((window2) => window2 !== null) && ["primary", "secondary"].every((key) => {
+                    const window2 = refreshedRateLimits.rateLimits[key];
+                    const previous = providerQuotaSnapshot?.[key];
+                    if (previous && (!window2 || window2.windowDurationMins !== previous.windowDurationMins)) {
+                      return false;
+                    }
+                    return window2 === null || window2.resetsAt !== null && window2.resetsAt * 1e3 > refreshedAt && window2.usedPercent < 100;
+                  })) {
+                    emitDiagnostic("provider_cooldown_cleared", {
+                      reason: providerCooldownReason,
+                      recovery: "provider_quota_restored",
+                      active_attempts: active.size,
+                      provider_weekly_quota: providerWeeklyQuotaFromRateLimits(
+                        providerRateLimits,
+                        providerRateLimitsObservedAtMs,
+                        refreshedAt
+                      )
+                    });
+                    providerRetryAtMs = null;
+                    providerCooldownReason = null;
+                    providerQuotaSnapshot = null;
                   }
-                  return window2 === null || window2.resetsAt !== null && window2.resetsAt * 1e3 > refreshedAt && window2.usedPercent < 100;
-                })) {
-                  emitDiagnostic("provider_cooldown_cleared", {
-                    reason: providerCooldownReason,
-                    recovery: "provider_quota_restored",
-                    active_attempts: active.size,
+                  emitDiagnostic("provider_rate_limits_refreshed", {
                     provider_weekly_quota: providerWeeklyQuotaFromRateLimits(
                       providerRateLimits,
                       providerRateLimitsObservedAtMs,
-                      refreshedAt
-                    )
+                      now()
+                    ),
+                    active_attempts: active.size
                   });
-                  providerRetryAtMs = null;
-                  providerCooldownReason = null;
-                  providerQuotaSnapshot = null;
+                } catch (error48) {
+                  emitDiagnostic("provider_rate_limits_refresh_failed", {
+                    failure_category: error48 instanceof CodexAppServerError ? error48.category : "adapter",
+                    failure_code: error48 instanceof CodexAppServerError ? error48.code : "provider_rate_limits_refresh_failed"
+                  });
+                } finally {
+                  nextProviderRateLimitRefreshAt = now() + settings.providerRateLimitRefreshMs;
+                  providerRateLimitRefresh = null;
                 }
-                emitDiagnostic("provider_rate_limits_refreshed", {
-                  provider_weekly_quota: providerWeeklyQuotaFromRateLimits(
-                    providerRateLimits,
-                    providerRateLimitsObservedAtMs,
-                    now()
-                  ),
-                  active_attempts: active.size
-                });
-              } catch (error48) {
-                emitDiagnostic("provider_rate_limits_refresh_failed", {
-                  failure_category: error48 instanceof CodexAppServerError ? error48.category : "adapter",
-                  failure_code: error48 instanceof CodexAppServerError ? error48.code : "provider_rate_limits_refresh_failed"
-                });
-              } finally {
-                nextProviderRateLimitRefreshAt = now() + settings.providerRateLimitRefreshMs;
-              }
+              });
             }
             const expiresAt = receipt.advertisement_expires_at ? Date.parse(receipt.advertisement_expires_at) : 0;
             const renewalDue = currentTime >= nextAdvertisementAttemptAt && expiresAt - currentTime <= settings.advertisementRefreshLeadMs;
@@ -37497,6 +37502,14 @@ var init_runner = __esm({
                   0,
                   claimCallRoundTripMs - (claim.schema_version === "external_inference_job_claim_result_v3" ? claim.server_processing_ms : 0)
                 );
+                emitDiagnostic("claim_batch_received", {
+                  claim_count: claim.claims.length,
+                  round_trip_ms: claimCallRoundTripMs,
+                  server_processing_ms: claim.schema_version === "external_inference_job_claim_result_v3" ? claim.server_processing_ms : null,
+                  transport_residual_ms: claimTransportResidualMs,
+                  minimum_freshness_remaining_ms: claim.claims.length ? Math.min(...claim.claims.map((item) => item.provider_dispatch_freshness_remaining_ms)) : null,
+                  active_attempts: active.size
+                });
                 if (claim.claims.length === 0) {
                   await mutateReceipt((current) => ({
                     ...current,
@@ -37674,6 +37687,7 @@ var init_runner = __esm({
           const agentCleanup = Promise.allSettled([
             agentLoop,
             modelCatalogRefresh,
+            providerRateLimitRefresh,
             Promise.resolve().then(async () => {
               await settings.agentRuntime?.adapter.close();
             })

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@vtxmacro/cli",
-  "version": "2026.9.22",
+  "version": "2026.9.23",
   "description": "VTX Macro CLI, MCP server, and durable external inference host.",
   "type": "module",
   "license": "UNLICENSED",