@vtxmacro/cli 2026.8.43 → 2026.8.45

package/bin/vtx.js

@@ -38,7 +38,7 @@ var init_agent_cli_release = __esm({
   "agent-cli-release.json"() {
     agent_cli_release_default = {
       package_name: "@vtxmacro/cli",
-      package_version: "2026.8.43",
+      package_version: "2026.8.45",
       codex_package_name: "@openai/codex",
       codex_version: "0.147.0",
       copilot_sdk_package_name: "@github/copilot-sdk",
@@ -14379,7 +14379,7 @@ var init_zod = __esm({
 });
 
 // lib/external-inference-contract.ts
-var EXTERNAL_INFERENCE_CONTRACT_VERSION, EXTERNAL_INFERENCE_CONTROLLER_SCHEMA_VERSION, EXTERNAL_INFERENCE_ADVERTISEMENT_SCHEMA_VERSION, EXTERNAL_INFERENCE_ENVELOPE_SCHEMA_VERSION, EXTERNAL_INFERENCE_JOB_INPUT_SCHEMA_VERSION, LEGACY_EXTERNAL_INFERENCE_JOB_INPUT_SCHEMA_VERSION, EXTERNAL_INFERENCE_USAGE_SCHEMA_VERSION, EXTERNAL_INFERENCE_INPUT_HASH_DOMAIN, MAX_EXTERNAL_INFERENCE_STATUS_HOSTS, MAX_GENERATION, MAX_SAFE_INTEGER, MAX_PROMPT_BYTES, MAX_IMMUTABLE_JOB_INPUT_BYTES, MAX_CANDIDATE_PLAINTEXT_BYTES, MAX_CIPHERTEXT_CHARS, SHA256_INITIAL_STATE, SHA256_ROUND_CONSTANTS, rotateRight, sha256BytesHex, sha256Utf8Hex, computeImmutableJobInputSha256, identifierSchema, operationIdentifierSchema, sha256HexSchema, generationSchema, zeroBasedGenerationSchema, nonNegativeSafeIntegerSchema, positiveSafeIntegerSchema, sqlIntegerTokenCountSchema, timestampSchema, safeCodeSchema, externalInferenceAdapterIdSchema, reasoningEffortSchema, modelNameSchema, displayNameSchema, authenticatedAccountEmailSchema, protocolVersionSchema, base64UrlAlphabet, isCanonicalBase64Url, base64Url32BytesSchema, base64UrlNonceSchema, base64UrlCiphertextSchema, claimHandleSchema, boundedJsonTextSchema, boundedPromptTextSchema, laneSchema, botModeSchema, executionModeSchema, responseModeSchema, dispatchOutcomeSchema, responseOutcomeSchema, nativeModelControllerSchema, externalAgentModelControllerSchema, externalAgentUnboundControllerSchema, orderedFallbackHostIdsSchema, externalInferenceFallbackHostIdsSchema, liveModelControllerSchema, snapshotModelControllerSchema, modelIdentityMappingSchema, advertisedModelSchema, hostAdvertisementSchema, envelopeAadSchema, validateEnvelopeAad, commonEnvelopeShape, sealedKeyWrapEnvelopeSchema, sealedInputEnvelopeSchema, sealedCandidateEnvelopeSchema, sealedEnvelopeSchema, screenerCandidateRuntimeBindingSchema, screenerClientLeaseBindingSchema, screenerRuntimeBindingSchema, serverRuntimeBindingSchema, clientRuntimeBindingSchema, runtimeBindingSchema, immutableJobInputSchema, canonicalizeJsonValue, serializeImmutableJobInput, reportedTokenUsageSchema, unavailableTokenUsageSchema, tokenUsageSchema, agentConnectRequestSchema, agentNextRequestSchema, agentHeartbeatRequestSchema, emptyAgentNextResultSchema, claimedAgentNextResultSchema, agentNextResultSchema, agentCompleteRequestSchema, agentFailRequestSchema, requestedModelSelectionSchema, completedModelProvenanceSchema, attemptModelProvenanceSchema, providerWeeklyQuotaReadSchema, hostHeartbeatRequestSchema, jobClaimRequestSchema, emptyJobClaimSchema, claimedJobSchema, jobClaimResultSchema, attemptStartRequestSchema, jobHeartbeatRequestSchema, continueHeartbeatDirectiveSchema, cancelHeartbeatDirectiveSchema, jobHeartbeatResultSchema, adapterOutcomeEvidenceSchema, completionBaseShape, validateCompletion, providerResponseCompletionSchema, decisionCandidateCompletionSchema, jobCompleteRequestSchema, jobFailRequestSchema, jobCancelRequestSchema, jobCancellationReceiptSchema, jobDeliveryRequestSchema, deliveryBaseShape, validateDeliveryCandidate, providerResponseDeliverySchema, decisionCandidateDeliverySchema, jobDeliveryResultSchema, jobConsumptionRequestSchema, jobConsumptionReceiptSchema, jobStateSchema, jobStatusReadSchema, hostStatusReadSchema, controllerBindingStatusReadSchema, queueStatusReadSchema, outcomeStatusReadSchema, externalInferenceVolumeWindowReadSchema, externalInferenceFailureCountReadSchema, externalInferenceHostDiagnosticsReadSchema, externalInferenceProfileActivityReadSchema, externalInferenceStatusReadSchema, nativeResearchEvidenceSourceSchema, externalAgentResearchEvidenceSourceSchema, researchEvidenceSourceSchema, reportedResearchTokenUsageSchema, unavailableResearchTokenUsageSchema, researchTokenUsageSchema, externalInferenceProductLineageBaseShape, traderExternalInferenceProductLineageSchema, decisionChatExternalInferenceProductLineageSchema, screenerExternalInferenceProductLineageSchema, externalInferenceProductLineageSchema, normalProductLinkageSchema, evidenceMaterializationMarkerSchema, deliveredExternalAgentSourceIdentitySchema, deliveredRequestEvidenceSchema, researchEvidenceBaseShape, researchEvidenceBaseSchema, validateExactUtf8, validateResearchEvidence, providerResponseResearchEvidenceSchema, decisionCandidateResearchEvidenceSchema, researchEvidenceSchema;
+var EXTERNAL_INFERENCE_CONTRACT_VERSION, EXTERNAL_INFERENCE_CONTROLLER_SCHEMA_VERSION, EXTERNAL_INFERENCE_ADVERTISEMENT_SCHEMA_VERSION, EXTERNAL_INFERENCE_ENVELOPE_SCHEMA_VERSION, EXTERNAL_INFERENCE_JOB_INPUT_SCHEMA_VERSION, LEGACY_EXTERNAL_INFERENCE_JOB_INPUT_SCHEMA_VERSION, EXTERNAL_INFERENCE_USAGE_SCHEMA_VERSION, EXTERNAL_INFERENCE_INPUT_HASH_DOMAIN, MAX_EXTERNAL_INFERENCE_STATUS_HOSTS, MAX_GENERATION, MAX_SAFE_INTEGER, MAX_PROMPT_BYTES, MAX_IMMUTABLE_JOB_INPUT_BYTES, MAX_CANDIDATE_PLAINTEXT_BYTES, MAX_CIPHERTEXT_CHARS, SHA256_INITIAL_STATE, SHA256_ROUND_CONSTANTS, rotateRight, sha256BytesHex, sha256Utf8Hex, computeImmutableJobInputSha256, identifierSchema, operationIdentifierSchema, sha256HexSchema, generationSchema, zeroBasedGenerationSchema, nonNegativeSafeIntegerSchema, positiveSafeIntegerSchema, sqlIntegerTokenCountSchema, timestampSchema, safeCodeSchema, processExitCodeSchema, externalInferenceAdapterIdSchema, reasoningEffortSchema, modelNameSchema, displayNameSchema, authenticatedAccountEmailSchema, protocolVersionSchema, base64UrlAlphabet, isCanonicalBase64Url, base64Url32BytesSchema, base64UrlNonceSchema, base64UrlCiphertextSchema, claimHandleSchema, boundedJsonTextSchema, boundedPromptTextSchema, laneSchema, botModeSchema, executionModeSchema, responseModeSchema, dispatchOutcomeSchema, responseOutcomeSchema, nativeModelControllerSchema, externalAgentModelControllerSchema, externalAgentUnboundControllerSchema, orderedFallbackHostIdsSchema, externalInferenceFallbackHostIdsSchema, liveModelControllerSchema, snapshotModelControllerSchema, modelIdentityMappingSchema, advertisedModelSchema, hostAdvertisementSchema, envelopeAadSchema, validateEnvelopeAad, commonEnvelopeShape, sealedKeyWrapEnvelopeSchema, sealedInputEnvelopeSchema, sealedCandidateEnvelopeSchema, sealedEnvelopeSchema, screenerCandidateRuntimeBindingSchema, screenerClientLeaseBindingSchema, screenerRuntimeBindingSchema, serverRuntimeBindingSchema, clientRuntimeBindingSchema, runtimeBindingSchema, immutableJobInputSchema, canonicalizeJsonValue, serializeImmutableJobInput, reportedTokenUsageSchema, unavailableTokenUsageSchema, tokenUsageSchema, agentConnectRequestSchema, agentNextRequestSchema, agentHeartbeatRequestSchema, emptyAgentNextResultSchema, claimedAgentNextResultSchema, agentNextResultSchema, agentCompleteRequestSchema, agentFailRequestSchema, requestedModelSelectionSchema, completedModelProvenanceSchema, attemptModelProvenanceSchema, providerWeeklyQuotaReadSchema, hostHeartbeatRequestSchema, jobClaimRequestSchema, emptyJobClaimSchema, claimedJobSchema, jobClaimResultSchema, attemptStartRequestSchema, jobHeartbeatRequestSchema, continueHeartbeatDirectiveSchema, cancelHeartbeatDirectiveSchema, jobHeartbeatResultSchema, adapterOutcomeEvidenceSchema, completionBaseShape, validateCompletion, providerResponseCompletionSchema, decisionCandidateCompletionSchema, jobCompleteRequestSchema, codexProcessExitDiagnosticsSchema, jobFailRequestSchema, jobCancelRequestSchema, jobCancellationReceiptSchema, jobDeliveryRequestSchema, deliveryBaseShape, validateDeliveryCandidate, providerResponseDeliverySchema, decisionCandidateDeliverySchema, jobDeliveryResultSchema, jobConsumptionRequestSchema, jobConsumptionReceiptSchema, jobStateSchema, jobStatusReadSchema, hostStatusReadSchema, controllerBindingStatusReadSchema, queueStatusReadSchema, outcomeStatusReadSchema, externalInferenceVolumeWindowReadSchema, externalInferenceFailureCountReadSchema, externalInferenceHostDiagnosticsReadSchema, externalInferenceProfileActivityReadSchema, externalInferenceStatusReadSchema, nativeResearchEvidenceSourceSchema, externalAgentResearchEvidenceSourceSchema, researchEvidenceSourceSchema, reportedResearchTokenUsageSchema, unavailableResearchTokenUsageSchema, researchTokenUsageSchema, externalInferenceProductLineageBaseShape, traderExternalInferenceProductLineageSchema, decisionChatExternalInferenceProductLineageSchema, screenerExternalInferenceProductLineageSchema, externalInferenceProductLineageSchema, normalProductLinkageSchema, evidenceMaterializationMarkerSchema, deliveredExternalAgentSourceIdentitySchema, deliveredRequestEvidenceSchema, researchEvidenceBaseShape, researchEvidenceBaseSchema, validateExactUtf8, validateResearchEvidence, providerResponseResearchEvidenceSchema, decisionCandidateResearchEvidenceSchema, researchEvidenceSchema;
 var init_external_inference_contract = __esm({
   "lib/external-inference-contract.ts"() {
     "use strict";
@@ -14582,6 +14582,7 @@ var init_external_inference_contract = __esm({
     sqlIntegerTokenCountSchema = external_exports.number().int().min(0).max(2147483647);
     timestampSchema = external_exports.iso.datetime({ offset: true });
     safeCodeSchema = external_exports.string().min(1).max(96).regex(/^[a-z0-9][a-z0-9._-]{0,95}$/);
+    processExitCodeSchema = external_exports.number().int().min(-(2 ** 31)).max(2 ** 31 - 1);
     externalInferenceAdapterIdSchema = external_exports.string().min(1).max(64).regex(/^[a-z0-9][a-z0-9._-]{0,63}$/);
     reasoningEffortSchema = external_exports.string().min(1).max(32).regex(/^[a-z][a-z0-9_-]{0,31}$/);
     modelNameSchema = external_exports.string().min(1).max(256);
@@ -15435,6 +15436,55 @@ var init_external_inference_contract = __esm({
       providerResponseCompletionSchema,
       decisionCandidateCompletionSchema
     ]);
+    codexProcessExitDiagnosticsSchema = external_exports.strictObject({
+      schema_version: external_exports.literal("external_inference_process_exit_v1"),
+      runtime: external_exports.literal("codex_app_server"),
+      source: external_exports.enum(["child_receipt", "direct_process", "guardian_process"]),
+      session_generation: generationSchema,
+      expected_shutdown: external_exports.boolean(),
+      exit_class: external_exports.enum(["zero", "nonzero", "signal", "unknown"]),
+      exit_code: processExitCodeSchema.nullable(),
+      signal: safeCodeSchema.nullable(),
+      uptime_ms: nonNegativeSafeIntegerSchema,
+      pending_request_count: nonNegativeSafeIntegerSchema.nullable(),
+      active_thread_count: nonNegativeSafeIntegerSchema.nullable(),
+      stderr_nonempty: external_exports.boolean().nullable(),
+      stderr_bytes_observed: external_exports.number().int().min(0).max(65536).nullable(),
+      stderr_truncated: external_exports.boolean().nullable(),
+      stderr_class: external_exports.enum([
+        "unknown",
+        "none",
+        "out_of_memory",
+        "runtime_fatal",
+        "memory_fault",
+        "pipe_or_connection",
+        "unclassified"
+      ])
+    }).superRefine((value, context) => {
+      const expectedExitClass = value.signal !== null ? "signal" : value.exit_code === null ? "unknown" : value.exit_code === 0 ? "zero" : "nonzero";
+      if (value.exit_class !== expectedExitClass) {
+        context.addIssue({ code: "custom", message: "process exit class does not match code and signal", path: ["exit_class"] });
+      }
+      if (value.signal !== null && value.exit_code !== null) {
+        context.addIssue({ code: "custom", message: "signal exits cannot also report an exit code", path: ["exit_code"] });
+      }
+      if (value.stderr_class === "unknown") {
+        if (value.stderr_nonempty !== null || value.stderr_bytes_observed !== null || value.stderr_truncated !== null) {
+          context.addIssue({ code: "custom", message: "unknown stderr diagnostics must not infer observations", path: ["stderr_class"] });
+        }
+      } else if (value.stderr_nonempty === null || value.stderr_bytes_observed === null || value.stderr_truncated === null) {
+        context.addIssue({ code: "custom", message: "observed stderr diagnostics must be complete", path: ["stderr_class"] });
+      } else if (!value.stderr_nonempty) {
+        if (value.stderr_bytes_observed !== 0 || value.stderr_truncated || value.stderr_class !== "none") {
+          context.addIssue({ code: "custom", message: "empty stderr diagnostics are inconsistent", path: ["stderr_nonempty"] });
+        }
+      } else if (value.stderr_bytes_observed === 0 || value.stderr_class === "none") {
+        context.addIssue({ code: "custom", message: "non-empty stderr diagnostics are inconsistent", path: ["stderr_nonempty"] });
+      }
+      if (value.stderr_truncated && value.stderr_bytes_observed !== 65536) {
+        context.addIssue({ code: "custom", message: "truncated stderr diagnostics must report the observation cap", path: ["stderr_truncated"] });
+      }
+    });
     jobFailRequestSchema = external_exports.strictObject({
       schema_version: external_exports.literal("external_inference_job_fail_v1"),
       contract_version: external_exports.literal(EXTERNAL_INFERENCE_CONTRACT_VERSION),
@@ -15457,6 +15507,7 @@ var init_external_inference_contract = __esm({
       failure_category: safeCodeSchema,
       failure_code: safeCodeSchema,
       retryable: external_exports.boolean(),
+      process_exit: codexProcessExitDiagnosticsSchema.nullable().optional(),
       membership_disposition: external_exports.enum([
         "retry_same_host",
         "cascade_cooldown",
@@ -15473,6 +15524,13 @@ var init_external_inference_contract = __esm({
       if (value.attempt_status !== expectedStatus) {
         context.addIssue({ code: "custom", message: "attempt_status must match dispatch outcome", path: ["attempt_status"] });
       }
+      if (value.process_exit != null && (value.failure_category !== "transport" || value.failure_code !== "codex_process_closed" || value.process_exit.expected_shutdown)) {
+        context.addIssue({
+          code: "custom",
+          message: "process exit diagnostics require an unexpected Codex process-close failure",
+          path: ["process_exit"]
+        });
+      }
       if (value.outcome.dispatch_outcome === "outcome_unknown" && value.membership_disposition !== void 0 && value.membership_disposition !== "quarantine_ambiguous") {
         context.addIssue({ code: "custom", message: "outcome-unknown failures must be quarantined", path: ["membership_disposition"] });
       }
@@ -16196,7 +16254,7 @@ import { spawn } from "node:child_process";
 import { constants } from "node:fs";
 import { lstat, mkdir, open, readFile, realpath, rename, rm } from "node:fs/promises";
 import { homedir } from "node:os";
-import { dirname, isAbsolute, join, relative, resolve } from "node:path";
+import { dirname, isAbsolute, join, relative, resolve, win32 as windowsPath } from "node:path";
 function inferenceBaseDir(env) {
   const configured = String(env.VTX_INFERENCE_HOST_HOME || "").trim();
   return configured || join(homedir(), ".vtx", "inference-host");
@@ -16737,7 +16795,7 @@ async function acquireInferenceHostProcessLock(path, dependencies = {}) {
     }
   };
 }
-var INFERENCE_CREDENTIAL_NAMESPACE, MAX_INFERENCE_PRIVATE_FILE_BYTES, WINDOWS_PRIVATE_ACL_BROKER_SCRIPT, windowsPrivateAclBrokerInvocation, WindowsPrivateAclBroker, defaultWindowsPrivateAclBroker, runWindowsPrivateAcl, windowsAclCache, windowsAclIdentity, aclCacheKey, secureWindowsPrivatePath, invalidateWindowsAclCache, rebindHardenedWindowsAclAfterRename, rebindHardenedWindowsDirectoryAfterOwnedMutation, isDefaultWindowsPrivatePath, secureExistingWindowsPath, linuxProcessIdentity, windowsProcessIdentity, darwinProcessIdentity, runSmallIdentityCommand, cachedSystemBootIdentity, readInferenceSystemBootIdentity, defaultProcessIdentity, defaultCurrentProcessIdentity, processLockObservationCache, inferenceProcessIdentitiesMatch, DEFAULT_INFERENCE_HOST_INSTANCE, SAFE_INFERENCE_HOST_INSTANCE, requireNamedInstancePath, credentialStoreIdentity, inferenceHostCredentialContextPath, inferenceHostCredentialContextTransitionPath, credentialContextForConfig;
+var INFERENCE_CREDENTIAL_NAMESPACE, MAX_INFERENCE_PRIVATE_FILE_BYTES, WINDOWS_PRIVATE_ACL_BROKER_SCRIPT, WINDOWS_IDENTITY_COMMAND_TIMEOUT_MS, WINDOWS_PRIVATE_ACL_BROKER_STARTUP_TIMEOUT_MS, SMALL_IDENTITY_COMMAND_TIMEOUT_MS, windowsPowerShellEnvironment, windowsPrivateAclBrokerInvocation, WindowsPrivateAclBroker, defaultWindowsPrivateAclBroker, runWindowsPrivateAcl, windowsAclCache, windowsAclIdentity, aclCacheKey, secureWindowsPrivatePath, invalidateWindowsAclCache, rebindHardenedWindowsAclAfterRename, rebindHardenedWindowsDirectoryAfterOwnedMutation, isDefaultWindowsPrivatePath, secureExistingWindowsPath, linuxProcessIdentity, windowsProcessIdentityInvocation, windowsBootIdentityInvocation, windowsProcessIdentity, darwinProcessIdentity, runSmallIdentityCommand, cachedSystemBootIdentity, readInferenceSystemBootIdentity, defaultProcessIdentity, defaultCurrentProcessIdentity, processLockObservationCache, inferenceProcessIdentitiesMatch, DEFAULT_INFERENCE_HOST_INSTANCE, SAFE_INFERENCE_HOST_INSTANCE, requireNamedInstancePath, credentialStoreIdentity, inferenceHostCredentialContextPath, inferenceHostCredentialContextTransitionPath, credentialContextForConfig;
 var init_config = __esm({
   "lib/inference-host/config.ts"() {
     "use strict";
@@ -16747,6 +16805,7 @@ var init_config = __esm({
 $ProgressPreference='SilentlyContinue'
 while(($line=[Console]::In.ReadLine()) -ne $null) {
   $requestId=$null
+  $failureStage='request_validation'
   try {
     if($line.Length -lt 2 -or $line.Length -gt 65536) { throw 'invalid request size' }
     $request=$line | ConvertFrom-Json
@@ -16761,18 +16820,25 @@ while(($line=[Console]::In.ReadLine()) -ne $null) {
     if($mode -ne 'harden' -and $mode -ne 'verify' -and $mode -ne 'verify-owner') { throw 'invalid ACL mode' }
     $path=[Text.Encoding]::UTF8.GetString([Convert]::FromBase64String([string]$request.path_b64))
     if([string]::IsNullOrWhiteSpace($path) -or $path.Length -gt 32767 -or $path.Contains([char]0)) { throw 'invalid path' }
+    $failureStage='path_lookup'
     if(-not (Test-Path -LiteralPath $path)) {
       [Console]::Out.WriteLine((@{id=$requestId;status='missing'} | ConvertTo-Json -Compress))
       continue
     }
-    $current=[Security.Principal.WindowsIdentity]::GetCurrent().User
+    $failureStage='principal_resolution'
+    $currentIdentity=[Security.Principal.WindowsIdentity]::GetCurrent()
+    $current=$currentIdentity.User
     $system=New-Object Security.Principal.SecurityIdentifier('S-1-5-18')
     $admins=New-Object Security.Principal.SecurityIdentifier('S-1-5-32-544')
+    $currentIsAdmin=(New-Object Security.Principal.WindowsPrincipal($currentIdentity)).IsInRole([Security.Principal.WindowsBuiltInRole]::Administrator)
     $allowed=@($current.Value,$system.Value,$admins.Value)
     if($mode -eq 'harden') {
+      $failureStage='acl_read'
       $acl=Get-Acl -LiteralPath $path
+      $failureStage='owner_validation'
       $existingOwner=(New-Object Security.Principal.NTAccount($acl.Owner)).Translate([Security.Principal.SecurityIdentifier]).Value
-      if($existingOwner -ne $current.Value) { throw 'private ACL owner is invalid' }
+      if($existingOwner -ne $current.Value -and ($existingOwner -ne $admins.Value -or -not $currentIsAdmin)) { throw 'private ACL owner is invalid' }
+      $failureStage='acl_mutation'
       $acl.SetAccessRuleProtection($true,$false)
       foreach($existingRule in @($acl.Access)) { [void]$acl.RemoveAccessRuleSpecific($existingRule) }
       $inherit=if($kind -eq 'directory') {
@@ -16788,11 +16854,14 @@ while(($line=[Console]::In.ReadLine()) -ne $null) {
         )
         [void]$acl.AddAccessRule($rule)
       }
+      $failureStage='acl_write'
       (Get-Item -LiteralPath $path).SetAccessControl($acl)
     }
+    $failureStage='acl_readback'
     $actual=Get-Acl -LiteralPath $path
+    $failureStage='acl_verification'
     $owner=(New-Object Security.Principal.NTAccount($actual.Owner)).Translate([Security.Principal.SecurityIdentifier]).Value
-    if($owner -ne $current.Value) { throw 'private ACL owner is invalid' }
+    if($owner -ne $current.Value -and ($owner -ne $admins.Value -or -not $currentIsAdmin)) { throw 'private ACL owner is invalid' }
     if($mode -ne 'verify-owner') {
       if(-not $actual.AreAccessRulesProtected) { throw 'private ACL inheritance is invalid' }
       $userFull=$false
@@ -16807,9 +16876,28 @@ while(($line=[Console]::In.ReadLine()) -ne $null) {
     }
     [Console]::Out.WriteLine((@{id=$requestId;status='ok'} | ConvertTo-Json -Compress))
   } catch {
-    [Console]::Out.WriteLine((@{id=$requestId;status='error'} | ConvertTo-Json -Compress))
+    [Console]::Out.WriteLine((@{id=$requestId;status='error';error_code=$failureStage} | ConvertTo-Json -Compress))
   }
 }`;
+    WINDOWS_IDENTITY_COMMAND_TIMEOUT_MS = 15e3;
+    WINDOWS_PRIVATE_ACL_BROKER_STARTUP_TIMEOUT_MS = 45e3;
+    SMALL_IDENTITY_COMMAND_TIMEOUT_MS = 5e3;
+    windowsPowerShellEnvironment = (extra = {}) => {
+      const sanitizedExtra = { ...extra };
+      delete sanitizedExtra.PSModulePath;
+      const windowsRoot = process.env.SystemRoot ?? process.env.WINDIR;
+      return {
+        ...sanitizedExtra,
+        NODE_ENV: process.env.NODE_ENV,
+        SystemRoot: process.env.SystemRoot,
+        WINDIR: process.env.WINDIR,
+        ComSpec: process.env.ComSpec,
+        PATHEXT: process.env.PATHEXT,
+        PSModulePath: windowsRoot ? windowsPath.join(windowsRoot, "System32", "WindowsPowerShell", "v1.0", "Modules") : void 0,
+        TEMP: process.env.TEMP,
+        TMP: process.env.TMP
+      };
+    };
     windowsPrivateAclBrokerInvocation = () => ({
       args: [
         "-NoLogo",
@@ -16818,16 +16906,7 @@ while(($line=[Console]::In.ReadLine()) -ne $null) {
         "-EncodedCommand",
         Buffer.from(WINDOWS_PRIVATE_ACL_BROKER_SCRIPT, "utf16le").toString("base64")
       ],
-      env: {
-        NODE_ENV: process.env.NODE_ENV,
-        SystemRoot: process.env.SystemRoot,
-        WINDIR: process.env.WINDIR,
-        ComSpec: process.env.ComSpec,
-        PATHEXT: process.env.PATHEXT,
-        PSModulePath: process.env.PSModulePath,
-        TEMP: process.env.TEMP,
-        TMP: process.env.TMP
-      }
+      env: windowsPowerShellEnvironment()
     });
     WindowsPrivateAclBroker = class {
       constructor(dependencies = {}) {
@@ -16846,6 +16925,7 @@ while(($line=[Console]::In.ReadLine()) -ne $null) {
           await operation;
         };
         this.requestTimeoutMs = dependencies.requestTimeoutMs ?? 15e3;
+        this.startupRequestTimeoutMs = dependencies.startupRequestTimeoutMs ?? WINDOWS_PRIVATE_ACL_BROKER_STARTUP_TIMEOUT_MS;
         this.idleShutdownMs = dependencies.idleShutdownMs ?? 3e4;
         this.spawnProcess = dependencies.spawnProcess ?? (() => {
           const invocation = windowsPrivateAclBrokerInvocation();
@@ -16896,7 +16976,19 @@ while(($line=[Console]::In.ReadLine()) -ne $null) {
         let response;
         try {
           const parsed = JSON.parse(line);
-          if (Object.keys(parsed).sort().join("\0") !== "id\0status" || parsed.id !== pending.id || !["ok", "missing", "error"].includes(String(parsed.status))) throw new Error("invalid response");
+          const status = String(parsed.status);
+          const expectedKeys = status === "error" ? "error_code\0id\0status" : "id\0status";
+          if (Object.keys(parsed).sort().join("\0") !== expectedKeys || parsed.id !== pending.id || !["ok", "missing", "error"].includes(status) || status === "error" && ![
+            "request_validation",
+            "path_lookup",
+            "principal_resolution",
+            "acl_read",
+            "owner_validation",
+            "acl_mutation",
+            "acl_write",
+            "acl_readback",
+            "acl_verification"
+          ].includes(String(parsed.error_code))) throw new Error("invalid response");
           response = parsed;
         } catch {
           this.failChild(child, new Error("Windows private ACL broker returned malformed output."));
@@ -16911,12 +17003,16 @@ while(($line=[Console]::In.ReadLine()) -ne $null) {
           error48.code = "ENOENT";
           pending.reject(error48);
         } else {
-          pending.reject(new Error(`Windows private ACL ${pending.mode} failed.`));
+          pending.reject(new Error(
+            `Windows private ACL ${pending.mode} failed (${response.error_code}).`
+          ));
         }
         this.scheduleIdleShutdown();
       }
       ensureChild() {
-        if (this.child && this.child.exitCode === null && !this.child.killed) return this.child;
+        if (this.child && this.child.exitCode === null && !this.child.killed) {
+          return { child: this.child, startedNow: false };
+        }
         const child = this.spawnProcess();
         this.child = child;
         this.outputBuffer = "";
@@ -16947,7 +17043,7 @@ while(($line=[Console]::In.ReadLine()) -ne $null) {
         for (const stream of [child.stdin, child.stdout, child.stderr]) {
           stream.unref?.();
         }
-        return child;
+        return { child, startedNow: true };
       }
       async dispatch(path, kind, mode) {
         if (this.idleTimer) {
@@ -16965,11 +17061,12 @@ while(($line=[Console]::In.ReadLine()) -ne $null) {
         if (Buffer.byteLength(request, "utf8") > 65536) {
           throw new Error("Windows private ACL broker request exceeds its input limit.");
         }
-        const child = this.ensureChild();
+        const { child, startedNow } = this.ensureChild();
+        const timeoutMs = startedNow ? this.startupRequestTimeoutMs : this.requestTimeoutMs;
         await new Promise((resolvePromise, reject) => {
           const timer = setTimeout(() => {
             this.failChild(child, new Error("Windows private ACL broker request timed out."));
-          }, this.requestTimeoutMs);
+          }, timeoutMs);
           this.pending = { id: id2, mode, resolve: resolvePromise, reject, timer };
           child.stdin.write(request, "utf8", (error48) => {
             if (error48) {
@@ -17072,26 +17169,40 @@ while(($line=[Console]::In.ReadLine()) -ne $null) {
         return null;
       }
     };
+    windowsProcessIdentityInvocation = (pid) => {
+      const script = `$ErrorActionPreference='Stop';$p=[Diagnostics.Process]::GetProcessById([int]$env:VTX_PROCESS_PID);[Console]::Out.Write($p.StartTime.ToUniversalTime().ToFileTimeUtc())`;
+      return {
+        args: [
+          "-NoLogo",
+          "-NoProfile",
+          "-NonInteractive",
+          "-EncodedCommand",
+          Buffer.from(script, "utf16le").toString("base64")
+        ],
+        env: windowsPowerShellEnvironment({ VTX_PROCESS_PID: String(pid) }),
+        timeoutMs: WINDOWS_IDENTITY_COMMAND_TIMEOUT_MS
+      };
+    };
+    windowsBootIdentityInvocation = () => {
+      const script = `$ErrorActionPreference='Stop';$s=[System.Management.ManagementObjectSearcher]::new('SELECT LastBootUpTime FROM Win32_OperatingSystem');try{$c=$s.Get();$e=$c.GetEnumerator();if(-not $e.MoveNext()){throw 'boot identity unavailable'};$raw=[string]$e.Current['LastBootUpTime'];$v=[System.Management.ManagementDateTimeConverter]::ToDateTime($raw).ToUniversalTime().ToFileTimeUtc();[Console]::Out.Write($v)}finally{if($c){$c.Dispose()};$s.Dispose()}`;
+      return {
+        args: [
+          "-NoLogo",
+          "-NoProfile",
+          "-NonInteractive",
+          "-EncodedCommand",
+          Buffer.from(script, "utf16le").toString("base64")
+        ],
+        env: windowsPowerShellEnvironment(),
+        timeoutMs: WINDOWS_IDENTITY_COMMAND_TIMEOUT_MS
+      };
+    };
     windowsProcessIdentity = async (pid) => await new Promise((resolvePromise) => {
-      const script = `$ErrorActionPreference='Stop';$p=Get-Process -Id ([int]$env:VTX_PROCESS_PID);[Console]::Out.Write($p.StartTime.ToUniversalTime().ToFileTimeUtc())`;
-      const child = spawn("powershell.exe", [
-        "-NoLogo",
-        "-NoProfile",
-        "-NonInteractive",
-        "-EncodedCommand",
-        Buffer.from(script, "utf16le").toString("base64")
-      ], {
+      const invocation = windowsProcessIdentityInvocation(pid);
+      const child = spawn("powershell.exe", [...invocation.args], {
         windowsHide: true,
         stdio: ["ignore", "pipe", "ignore"],
-        env: {
-          NODE_ENV: process.env.NODE_ENV,
-          SystemRoot: process.env.SystemRoot,
-          WINDIR: process.env.WINDIR,
-          ComSpec: process.env.ComSpec,
-          PATHEXT: process.env.PATHEXT,
-          PSModulePath: process.env.PSModulePath,
-          VTX_PROCESS_PID: String(pid)
-        }
+        env: invocation.env
       });
       const output3 = [];
       let settled = false;
@@ -17104,7 +17215,7 @@ while(($line=[Console]::In.ReadLine()) -ne $null) {
       const timer = setTimeout(() => {
         child.kill("SIGKILL");
         settle(null);
-      }, 5e3);
+      }, invocation.timeoutMs);
       child.stdout.on("data", (chunk) => {
         if (output3.reduce((total, item) => total + item.byteLength, 0) + Buffer.byteLength(chunk) > 1024) {
           child.kill("SIGKILL");
@@ -17142,7 +17253,7 @@ while(($line=[Console]::In.ReadLine()) -ne $null) {
         settle(code === 0 && Number.isFinite(started) ? `darwin:${Math.floor(started / 1e3)}` : null);
       });
     });
-    runSmallIdentityCommand = async (command, args, env) => await new Promise((resolvePromise) => {
+    runSmallIdentityCommand = async (command, args, env, timeoutMs = SMALL_IDENTITY_COMMAND_TIMEOUT_MS) => await new Promise((resolvePromise) => {
       const child = spawn(command, [...args], {
         windowsHide: true,
         stdio: ["ignore", "pipe", "ignore"],
@@ -17160,7 +17271,7 @@ while(($line=[Console]::In.ReadLine()) -ne $null) {
       const timer = setTimeout(() => {
         child.kill("SIGKILL");
         settle(null);
-      }, 5e3);
+      }, timeoutMs);
       child.stdout.on("data", (chunk) => {
         bytes3 += Buffer.byteLength(chunk);
         if (bytes3 > 4096) {
@@ -17189,21 +17300,13 @@ while(($line=[Console]::In.ReadLine()) -ne $null) {
             identity = null;
           }
         } else if (platform === "win32") {
-          const script = `$ErrorActionPreference='Stop';$v=(Get-CimInstance Win32_OperatingSystem).LastBootUpTime.ToUniversalTime().ToFileTimeUtc();[Console]::Out.Write($v)`;
-          identity = await runSmallIdentityCommand("powershell.exe", [
-            "-NoLogo",
-            "-NoProfile",
-            "-NonInteractive",
-            "-EncodedCommand",
-            Buffer.from(script, "utf16le").toString("base64")
-          ], {
-            NODE_ENV: process.env.NODE_ENV,
-            SystemRoot: process.env.SystemRoot,
-            WINDIR: process.env.WINDIR,
-            ComSpec: process.env.ComSpec,
-            PATHEXT: process.env.PATHEXT,
-            PSModulePath: process.env.PSModulePath
-          });
+          const invocation = windowsBootIdentityInvocation();
+          identity = await runSmallIdentityCommand(
+            "powershell.exe",
+            invocation.args,
+            invocation.env,
+            invocation.timeoutMs
+          );
         } else if (platform === "darwin") {
           identity = await runSmallIdentityCommand("sysctl", ["-n", "kern.boottime"], void 0);
         }
@@ -19548,6 +19651,7 @@ var init_codex_app_server = __esm({
         this.retryAtMs = options.retryAtMs ?? null;
         this.usage = options.usage ?? null;
         this.rateLimits = options.rateLimits ?? null;
+        this.processExit = options.processExit ?? null;
       }
     };
     objectOrNull = (value) => value !== null && typeof value === "object" && !Array.isArray(value) ? value : null;
@@ -19852,7 +19956,7 @@ var init_codex_app_server = __esm({
     };
     GUARDIAN_SCRIPT = String.raw`
 import { spawn } from 'node:child_process';
-import { chmod, open, rename, rm, writeFile } from 'node:fs/promises';
+import { chmod, open, readFile, rename, rm, writeFile } from 'node:fs/promises';
 import { dirname } from 'node:path';
 const input = JSON.parse(Buffer.from(process.argv[1], 'base64url').toString('utf8'));
 // The guardian owns the process boundary for every attempt. Apply a private
@@ -19882,7 +19986,7 @@ const replaceReceipt = async (temporary) => {
 };
 let receiptWriteSequence = 0;
 let receiptWriteChain = Promise.resolve();
-const writeReceipt = (state, childPid) => {
+const writeReceipt = (state, childPid, termination = null) => {
   const write = async () => {
     const body = JSON.stringify({
       schema_version: 'vtx_codex_guardian_v1',
@@ -19891,6 +19995,7 @@ const writeReceipt = (state, childPid) => {
       guardian_pid: process.pid,
       child_pid: childPid ?? null,
       boot_identity: input.bootIdentity,
+      termination,
       updated_at: new Date().toISOString(),
     }) + '\n';
     receiptWriteSequence += 1;
@@ -19915,8 +20020,11 @@ const writeReceipt = (state, childPid) => {
 };
 let child = null;
 let shuttingDown = false;
+let shutdownRequested = false;
+let childStartedAt = null;
 let forceTimer = null;
-const shutdown = () => {
+const stopChild = (expected) => {
+  if (expected) shutdownRequested = true;
   if (shuttingDown) return;
   shuttingDown = true;
   if (!child) return;
@@ -19926,6 +20034,19 @@ const shutdown = () => {
   }, input.shutdownGraceMs);
   forceTimer.unref();
 };
+let shutdownStarted = false;
+const shutdown = async () => {
+  if (shutdownStarted) return;
+  shutdownStarted = true;
+  let expected = false;
+  try {
+    expected = (await readFile(input.shutdownIntentPath, 'utf8')) === input.processToken;
+    await rm(input.shutdownIntentPath, { force: true });
+  } catch {
+    expected = false;
+  }
+  stopChild(expected);
+};
 process.stdin.once('end', shutdown);
 process.stdin.once('close', shutdown);
 process.stdout.on('error', () => undefined);
@@ -19941,6 +20062,7 @@ child = spawn(input.binary, input.args, {
   windowsHide: true,
   detached: false,
 });
+childStartedAt = Date.now();
 child.stdin.on('error', () => undefined);
 process.stdin.pipe(child.stdin, { end: false });
 child.stdout.pipe(process.stdout);
@@ -19949,18 +20071,33 @@ child.once('spawn', async () => {
   try {
     await writeReceipt('running', child.pid ?? null);
   } catch {
-    shutdown();
+    stopChild(false);
   }
 });
-child.once('error', shutdown);
-child.once('close', async () => {
+child.once('error', () => stopChild(false));
+child.once('close', async (code, signal) => {
   if (forceTimer) clearTimeout(forceTimer);
   // A pending promise alone does not keep Node's event loop alive after the
   // child and parent stdin have both closed. Hold one local handle until the
   // serialized terminal tombstone is durably replaced.
   const receiptKeepAlive = setInterval(() => undefined, 1_000);
   try {
-    await writeReceipt('terminated', child.pid ?? null);
+    const normalizedSignal = typeof signal === 'string' && /^SIG[A-Z0-9]{1,28}$/.test(signal)
+      ? signal.toLowerCase()
+      : null;
+    const normalizedCode = normalizedSignal === null && Number.isInteger(code)
+      && code >= -(2 ** 31) && code <= (2 ** 31) - 1
+      ? code
+      : null;
+    await writeReceipt('terminated', child.pid ?? null, {
+      exit_code: normalizedCode,
+      signal: normalizedSignal,
+      expected: shutdownRequested,
+      uptime_ms: Math.min(
+        Math.max(0, Date.now() - (childStartedAt ?? Date.now())),
+        Number.MAX_SAFE_INTEGER,
+      ),
+    });
   } finally {
     clearInterval(receiptKeepAlive);
     process.exit(0);
@@ -20012,6 +20149,7 @@ child.once('close', async () => {
         guardian_pid: null,
         child_pid: null,
         boot_identity: guardian.bootIdentity ?? null,
+        termination: null,
         updated_at: (/* @__PURE__ */ new Date()).toISOString()
       })}
 `;
@@ -20036,12 +20174,15 @@ child.once('close', async () => {
     };
     parseGuardianReceipt = (raw) => {
       const value = JSON.parse(raw);
-      if (value.schema_version !== "vtx_codex_guardian_v1" || typeof value.process_token !== "string" || !/^[0-9a-f]{64}$/u.test(value.process_token) || !["spawn_intent", "starting", "running", "terminated"].includes(String(value.state)) || value.guardian_pid !== null && !Number.isSafeInteger(value.guardian_pid) || value.state === "spawn_intent" && value.guardian_pid !== null || value.state !== "spawn_intent" && !Number.isSafeInteger(value.guardian_pid) || value.child_pid !== null && !Number.isSafeInteger(value.child_pid) || value.boot_identity !== null && value.boot_identity !== void 0 && typeof value.boot_identity !== "string" || typeof value.updated_at !== "string" || !Number.isFinite(Date.parse(value.updated_at))) {
+      const rawTermination = value.termination;
+      const termination = rawTermination === null || rawTermination === void 0 ? null : objectOrNull(rawTermination);
+      if (value.schema_version !== "vtx_codex_guardian_v1" || typeof value.process_token !== "string" || !/^[0-9a-f]{64}$/u.test(value.process_token) || !["spawn_intent", "starting", "running", "terminated"].includes(String(value.state)) || value.guardian_pid !== null && !Number.isSafeInteger(value.guardian_pid) || value.state === "spawn_intent" && value.guardian_pid !== null || value.state !== "spawn_intent" && !Number.isSafeInteger(value.guardian_pid) || value.child_pid !== null && !Number.isSafeInteger(value.child_pid) || value.boot_identity !== null && value.boot_identity !== void 0 && typeof value.boot_identity !== "string" || typeof value.updated_at !== "string" || !Number.isFinite(Date.parse(value.updated_at)) || rawTermination !== null && rawTermination !== void 0 && termination === null || termination !== null && (value.state !== "terminated" || termination.exit_code !== null && (!Number.isInteger(termination.exit_code) || Number(termination.exit_code) < -(2 ** 31) || Number(termination.exit_code) > 2 ** 31 - 1) || termination.signal !== null && (typeof termination.signal !== "string" || !/^sig[a-z0-9]{1,28}$/u.test(termination.signal)) || termination.signal !== null && termination.exit_code !== null || typeof termination.expected !== "boolean" || !Number.isSafeInteger(termination.uptime_ms) || Number(termination.uptime_ms) < 0)) {
         throw new Error("Codex guardian receipt is invalid.");
       }
       return {
         ...value,
-        boot_identity: typeof value.boot_identity === "string" ? value.boot_identity : null
+        boot_identity: typeof value.boot_identity === "string" ? value.boot_identity : null,
+        termination
       };
     };
     readCodexGuardianReceipt = async (path) => {
@@ -20085,6 +20226,7 @@ child.once('close', async () => {
         args,
         processToken: options.guardian.processToken,
         receiptPath: options.guardian.receiptPath,
+        shutdownIntentPath: `${options.guardian.receiptPath}.shutdown-intent`,
         shutdownGraceMs: Math.max(25, options.guardian.shutdownGraceMs ?? 2e3),
         bootIdentity: options.guardian.bootIdentity
       }), "utf8").toString("base64url");
@@ -20106,13 +20248,33 @@ child.once('close', async () => {
         this.nextRequestId = 1;
         this.fatalError = null;
         this.closed = false;
+        this.processCloseObserved = false;
+        this.shutdownRequested = false;
+        this.stderrBytesObserved = 0;
+        this.stderrNonempty = false;
+        this.stderrTruncated = false;
+        this.stderrTail = "";
+        this.stderrClass = "none";
         this.process = options.process;
         this.codexHome = options.codexHome;
-        this.process.stderr.resume();
+        this.sessionGeneration = options.sessionGeneration;
+        this.startedAtMs = Date.now();
+        this.guardian = options.guardian;
+        this.onProcessExit = options.onProcessExit;
+        this.process.stderr.on("data", (chunk) => this.observeStderr(chunk));
         const lines = createInterface({ input: this.process.stdout, crlfDelay: Infinity });
         lines.on("line", (line) => this.handleLine(line));
-        this.process.once("error", () => this.failTransport("codex_process_error"));
-        this.process.once("close", () => this.failTransport("codex_process_closed"));
+        this.process.once("error", () => {
+        });
+        this.process.once("close", (code, signal) => {
+          void this.handleProcessClose(code, signal);
+        });
+      }
+      static {
+        this.STDERR_OBSERVATION_CAP_BYTES = 65536;
+      }
+      static {
+        this.STDERR_CLASSIFICATION_TAIL_CHARS = 1024;
       }
       static async start(options) {
         const spawnProcess = options.spawnProcess ?? defaultSpawn;
@@ -20173,7 +20335,13 @@ child.once('close', async () => {
             });
           }
         }
-        const session = new _CodexAppServerSession({ process: child, codexHome: options.codexHome });
+        const session = new _CodexAppServerSession({
+          process: child,
+          codexHome: options.codexHome,
+          sessionGeneration: options.sessionGeneration ?? 1,
+          guardian: guardian ?? null,
+          onProcessExit: options.onProcessExit ?? null
+        });
         try {
           const initialized = await session.request("initialize", {
             clientInfo: {
@@ -20288,13 +20456,14 @@ child.once('close', async () => {
         }
         for (const listener of this.notificationListeners) listener({ method, params });
       }
-      failTransport(code) {
+      failTransport(code, processExit = null) {
         if (this.fatalError) return;
         this.fatalError = new CodexAppServerError({
           message: "Codex app-server transport became unavailable.",
           category: "transport",
           code,
-          retryable: true
+          retryable: true,
+          processExit
         });
         for (const pending of this.pending.values()) {
           if (pending.timer) clearTimeout(pending.timer);
@@ -20395,6 +20564,77 @@ child.once('close', async () => {
         this.fatalListeners.add(listener);
         return () => this.fatalListeners.delete(listener);
       }
+      observeStderr(chunk) {
+        const bytes3 = Buffer.isBuffer(chunk) ? chunk : Buffer.from(chunk, "utf8");
+        if (bytes3.length === 0) return;
+        this.stderrNonempty = true;
+        const remaining = Math.max(
+          0,
+          _CodexAppServerSession.STDERR_OBSERVATION_CAP_BYTES - this.stderrBytesObserved
+        );
+        if (remaining === 0) {
+          this.stderrTruncated = true;
+          return;
+        }
+        const observed = bytes3.subarray(0, remaining);
+        this.stderrBytesObserved += observed.length;
+        if (bytes3.length > remaining) this.stderrTruncated = true;
+        this.stderrTail = `${this.stderrTail}${observed.toString("utf8")}`.slice(-_CodexAppServerSession.STDERR_CLASSIFICATION_TAIL_CHARS).toLowerCase();
+        const classified = /heap out of memory|allocation failed|out of memory/u.test(this.stderrTail) ? "out_of_memory" : /fatal runtime error|fatal error|\bpanic(?:ked)?\b/u.test(this.stderrTail) ? "runtime_fatal" : /access violation|segmentation fault|\bsegfault\b/u.test(this.stderrTail) ? "memory_fault" : /broken pipe|econnreset|connection reset/u.test(this.stderrTail) ? "pipe_or_connection" : null;
+        if (classified) this.stderrClass = classified;
+        else if (this.stderrClass === "none") this.stderrClass = "unclassified";
+      }
+      async handleProcessClose(rawCode, rawSignal) {
+        this.processCloseObserved = true;
+        let termination = null;
+        if (this.guardian) {
+          try {
+            const receipt = await readCodexGuardianReceipt(this.guardian.receiptPath);
+            if (receipt?.process_token === this.guardian.processToken && receipt.state === "terminated") {
+              termination = receipt.termination;
+            }
+          } catch {
+          }
+        }
+        const diagnostics = this.processExitDiagnostics(rawCode, rawSignal, termination);
+        if (this.guardian) {
+          await rm2(`${this.guardian.receiptPath}.shutdown-intent`, { force: true }).catch(() => void 0);
+        }
+        try {
+          this.onProcessExit?.(diagnostics);
+        } catch {
+        }
+        this.failTransport("codex_process_closed", diagnostics);
+      }
+      processExitDiagnostics(rawCode, rawSignal, termination) {
+        const source = termination ? "child_receipt" : this.guardian ? "guardian_process" : "direct_process";
+        const reportedCode = termination?.exit_code ?? rawCode;
+        const reportedSignal = termination?.signal ?? rawSignal;
+        const signal = typeof reportedSignal === "string" && /^(?:SIG[A-Z0-9]{1,28}|sig[a-z0-9]{1,28})$/u.test(reportedSignal) ? reportedSignal.toLowerCase() : null;
+        const exitCode = signal === null && typeof reportedCode === "number" && Number.isInteger(reportedCode) && reportedCode >= -(2 ** 31) && reportedCode <= 2 ** 31 - 1 ? reportedCode : null;
+        const diagnostics = {
+          schema_version: "external_inference_process_exit_v1",
+          runtime: "codex_app_server",
+          source,
+          session_generation: this.sessionGeneration,
+          expected_shutdown: termination?.expected ?? this.shutdownRequested,
+          exit_class: signal !== null ? "signal" : exitCode === null ? "unknown" : exitCode === 0 ? "zero" : "nonzero",
+          exit_code: exitCode,
+          signal,
+          uptime_ms: termination?.uptime_ms ?? Math.min(
+            Math.max(0, Date.now() - this.startedAtMs),
+            Number.MAX_SAFE_INTEGER
+          ),
+          pending_request_count: this.pending.size,
+          active_thread_count: this.threads.size,
+          stderr_nonempty: this.stderrNonempty,
+          stderr_bytes_observed: this.stderrBytesObserved,
+          stderr_truncated: this.stderrTruncated,
+          stderr_class: this.stderrNonempty ? this.stderrClass : "none"
+        };
+        this.stderrTail = "";
+        return diagnostics;
+      }
       isUsable() {
         return !this.closed && this.fatalError === null && this.process.exitCode === null;
       }
@@ -20819,6 +21059,7 @@ child.once('close', async () => {
             code: error48.code,
             retryable: error48.retryable,
             dispatchOutcome: requestWritten ? "outcome_unknown" : "not_dispatched",
+            processExit: error48.processExit,
             cause: error48
           }));
         });
@@ -21022,6 +21263,7 @@ child.once('close', async () => {
                 code: error48.code,
                 retryable: error48.retryable,
                 dispatchOutcome: "outcome_unknown",
+                processExit: error48.processExit,
                 cause: error48
               });
             }
@@ -21274,6 +21516,20 @@ child.once('close', async () => {
       async close() {
         if (this.closed) return;
         this.closed = true;
+        if (this.processCloseObserved || this.process.exitCode !== null) {
+          if (this.guardian) {
+            await rm2(`${this.guardian.receiptPath}.shutdown-intent`, { force: true }).catch(() => void 0);
+          }
+          return;
+        }
+        this.shutdownRequested = true;
+        if (this.guardian) {
+          await writeFile(
+            `${this.guardian.receiptPath}.shutdown-intent`,
+            this.guardian.processToken,
+            { mode: 384 }
+          ).catch(() => void 0);
+        }
         this.process.stdin.end();
         const exited = await new Promise((resolve6) => {
           if (this.process.exitCode !== null) {
@@ -21482,11 +21738,12 @@ import {
 import { createHash as createHash4, randomBytes as randomBytes3 } from "node:crypto";
 import { tmpdir as tmpdir2 } from "node:os";
 import { isAbsolute as isAbsolute3, join as join4, relative as relative2, resolve as resolve3, sep as sep2 } from "node:path";
-var MAX_PROMPT_BYTES2, CODEX_MODEL_NAME_PATTERN, CODEX_REASONING_EFFORT_PATTERN, codexRecoveryCheckpointProcessFingerprint, proveCodexSameBootQuiescence, tokenUsageReceiptSchema, turnResultReceiptSchema, terminalReceiptSchema, recoveryCheckpointSchema, recoveryFileSchema, FileCodexAttemptRecoveryStore, utf8Bytes, assertAttemptActive, tomlString, permissionConfig, ensureDedicatedCodexHome, createIsolatedCodexAttemptResources, createIsolatedCodexHostResources, validateAttemptInput, isStrictDescendant, assertRecoveryResourceScope, removeRecoveredThread, confirmGuardianTerminatedForRecovery, reconcileCodexAttemptRecovery, CodexSubscriptionAdapter;
+var MAX_PROMPT_BYTES2, CODEX_MODEL_NAME_PATTERN, CODEX_REASONING_EFFORT_PATTERN, codexRecoveryCheckpointProcessFingerprint, proveCodexSameBootQuiescence, tokenUsageReceiptSchema, turnResultReceiptSchema, terminalReceiptSchema, recoveryCheckpointSchema, recoveryFileSchema, FileCodexAttemptRecoveryStore, utf8Bytes, assertAttemptActive, tomlString, permissionConfig, ensureDedicatedCodexHome, createIsolatedCodexAttemptResources, createIsolatedCodexHostResources, validateAttemptInput, isStrictDescendant, assertRecoveryResourceScope, removeRecoveredThread, confirmGuardianTerminatedForRecovery, recoveredCodexProcessExit, reconcileCodexAttemptRecovery, CodexSubscriptionAdapter;
 var init_codex_adapter = __esm({
   "lib/inference-host/codex-adapter.ts"() {
     "use strict";
     init_zod();
+    init_external_inference_contract();
     init_codex_app_server();
     init_codex_binary();
     init_config();
@@ -21561,7 +21818,8 @@ var init_codex_adapter = __esm({
           dispatchOutcome: external_exports.enum(["not_dispatched", "confirmed_dispatched", "outcome_unknown"]),
           httpStatusCode: external_exports.number().int().min(100).max(599).nullable().optional(),
           retryAtMs: external_exports.number().int().nonnegative().max(Number.MAX_SAFE_INTEGER).nullable().optional(),
-          usage: tokenUsageReceiptSchema.nullable().optional()
+          usage: tokenUsageReceiptSchema.nullable().optional(),
+          processExit: codexProcessExitDiagnosticsSchema.nullable().optional()
         })
       })
     ]);
@@ -21577,6 +21835,7 @@ var init_codex_adapter = __esm({
       processToken: external_exports.string().regex(/^[0-9a-f]{64}$/u).nullable(),
       processReceiptPath: external_exports.string().min(1).max(4096).nullable(),
       processState: external_exports.enum(["unmanaged", "spawn_intent", "running", "terminated"]),
+      sessionGeneration: external_exports.number().int().positive().max(2147483647).nullable().optional(),
       bootIdentity: external_exports.string().min(1).max(4096).nullable().optional(),
       cleanupConfirmed: external_exports.boolean(),
       adapterRequestId: external_exports.string().min(1).max(512).nullable(),
@@ -21971,6 +22230,30 @@ var init_codex_adapter = __esm({
         throw error48;
       }
     };
+    recoveredCodexProcessExit = (checkpoint, receipt) => {
+      const termination = receipt?.termination;
+      const sessionGeneration = checkpoint.sessionGeneration ?? null;
+      if (!termination || termination.expected || sessionGeneration === null) return null;
+      const signal = termination.signal;
+      const exitCode = signal === null ? termination.exit_code : null;
+      return {
+        schema_version: "external_inference_process_exit_v1",
+        runtime: "codex_app_server",
+        source: "child_receipt",
+        session_generation: sessionGeneration,
+        expected_shutdown: false,
+        exit_class: signal !== null ? "signal" : exitCode === null ? "unknown" : exitCode === 0 ? "zero" : "nonzero",
+        exit_code: exitCode,
+        signal,
+        uptime_ms: termination.uptime_ms,
+        pending_request_count: null,
+        active_thread_count: null,
+        stderr_nonempty: null,
+        stderr_bytes_observed: null,
+        stderr_truncated: null,
+        stderr_class: "unknown"
+      };
+    };
     reconcileCodexAttemptRecovery = async (options) => {
       const attempts = await options.recoveryHooks.loadAll?.() ?? {};
       const currentBootIdentity = options.bootIdentity ?? await readInferenceSystemBootIdentity();
@@ -22011,11 +22294,27 @@ var init_codex_adapter = __esm({
         assertRecoveryResourceScope(checkpoint);
         await removeRecoveredThread(checkpoint, options.codexHome);
         await rm3(checkpoint.resourceRoot, { recursive: true, force: true });
+        const recoveredProcessExit = recoveredCodexProcessExit(
+          checkpoint,
+          await readCodexGuardianReceipt(guardian.receiptPath)
+        );
+        const dispatchOutcome = crossedBoot ? "outcome_unknown" : checkpoint.dispatchOutcome;
         await options.recoveryHooks.save({
           ...checkpoint,
           processState: "terminated",
           cleanupConfirmed: true,
-          dispatchOutcome: crossedBoot ? "outcome_unknown" : checkpoint.dispatchOutcome
+          dispatchOutcome,
+          terminal: checkpoint.terminal ?? (recoveredProcessExit ? {
+            kind: "failed",
+            error: {
+              message: "Codex app-server transport became unavailable.",
+              category: "transport",
+              code: "codex_process_closed",
+              retryable: true,
+              dispatchOutcome,
+              processExit: recoveredProcessExit
+            }
+          } : null)
         });
         reconciled += 1;
       }
@@ -22029,8 +22328,15 @@ var init_codex_adapter = __esm({
         this.durableSessionStart = null;
         this.closing = false;
         this.adapterClosePromise = null;
+        this.sessionGeneration = 0;
         this.dependencies = dependencies;
       }
+      emitDiagnostic(event) {
+        try {
+          this.dependencies.emitDiagnosticEvent?.(event);
+        } catch {
+        }
+      }
       async readAccountRateLimits(deadlineAtMs, signal) {
         const codexHome = this.dependencies.codexHome;
         if (!codexHome) {
@@ -22049,6 +22355,9 @@ var init_codex_adapter = __esm({
         return await durableSession.session.readAccountRateLimits(deadlineAtMs, signal);
       }
       async startDurableSession(codexHome, deadlineAtMs) {
+        this.sessionGeneration = Math.min(this.sessionGeneration + 1, 2147483647);
+        const sessionGeneration = this.sessionGeneration;
+        const activeAttemptIds = /* @__PURE__ */ new Set();
         const binary = await (this.dependencies.resolveBinary ?? resolvePinnedCodexBinary)();
         const guardianManaged = Boolean(
           this.dependencies.recoveryHooks && !this.dependencies.spawnProcess
@@ -22063,10 +22372,26 @@ var init_codex_adapter = __esm({
           codexHome,
           deadlineAtMs,
           spawnProcess: this.dependencies.spawnProcess,
-          guardian: processToken && processReceiptPath ? { processToken, receiptPath: processReceiptPath, bootIdentity: bootIdentity ?? void 0 } : void 0
+          guardian: processToken && processReceiptPath ? { processToken, receiptPath: processReceiptPath, bootIdentity: bootIdentity ?? void 0 } : void 0,
+          sessionGeneration,
+          onProcessExit: (diagnostics) => {
+            this.emitDiagnostic({
+              event: "codex_app_server_closed",
+              process_exit: diagnostics,
+              active_attempt_ids: [...activeAttemptIds].sort().slice(0, 64),
+              active_attempt_ids_truncated: activeAttemptIds.size > 64
+            });
+          }
+        });
+        this.emitDiagnostic({
+          event: "codex_app_server_session_started",
+          session_generation: sessionGeneration,
+          replacement: sessionGeneration > 1
         });
         return {
           session,
+          sessionGeneration,
+          activeAttemptIds,
           codexHome,
           capabilityPreflight: null,
           processToken,
@@ -22306,6 +22631,7 @@ var init_codex_adapter = __esm({
             const receipt = await readCodexGuardianReceipt(checkpoint.processReceiptPath);
             if (checkpoint.processState === "terminated" || receipt === null || receipt.state === "terminated") {
               await rm3(checkpoint.processReceiptPath, { force: true });
+              await rm3(`${checkpoint.processReceiptPath}.shutdown-intent`, { force: true });
             }
           }
         }
@@ -22353,6 +22679,7 @@ var init_codex_adapter = __esm({
             for (const session of sessions) {
               if (session.processReceiptPath && unconfirmedReceiptPaths !== null && !unconfirmedReceiptPaths.has(resolve3(session.processReceiptPath))) {
                 await rm3(session.processReceiptPath, { force: true });
+                await rm3(`${session.processReceiptPath}.shutdown-intent`, { force: true });
               }
             }
           })();
@@ -22397,7 +22724,8 @@ var init_codex_adapter = __esm({
               dispatchOutcome: recovered.dispatchOutcome,
               httpStatusCode: recovered.httpStatusCode,
               retryAtMs: recovered.retryAtMs,
-              usage: recovered.usage
+              usage: recovered.usage,
+              processExit: recovered.processExit
             });
           }
           throw new CodexAppServerError({
@@ -22430,6 +22758,7 @@ var init_codex_adapter = __esm({
             input.deadlineAtMs,
             input.signal
           );
+          sessionLease.activeAttemptIds.add(input.attemptId);
           session = sessionLease.session;
           baseCheckpoint = {
             schemaVersion: "vtx_codex_attempt_recovery_v1",
@@ -22443,6 +22772,7 @@ var init_codex_adapter = __esm({
             processToken: sessionLease.processToken,
             processReceiptPath: sessionLease.processReceiptPath,
             processState: sessionLease.processToken ? "running" : "unmanaged",
+            sessionGeneration: sessionLease.sessionGeneration,
             bootIdentity: sessionLease.bootIdentity,
             cleanupConfirmed: false,
             adapterRequestId: null,
@@ -22541,7 +22871,8 @@ var init_codex_adapter = __esm({
                     dispatchOutcome,
                     httpStatusCode: error48.httpStatusCode,
                     retryAtMs: error48.retryAtMs,
-                    usage: error48.usage
+                    usage: error48.usage,
+                    processExit: error48.processExit
                   }
                 }
               };
@@ -22557,6 +22888,7 @@ var init_codex_adapter = __esm({
               httpStatusCode: error48.httpStatusCode,
               retryAtMs: error48.retryAtMs,
               usage: error48.usage,
+              processExit: error48.processExit,
               cause: error48
             });
           }
@@ -22587,6 +22919,7 @@ var init_codex_adapter = __esm({
           }
           throw classified;
         } finally {
+          sessionLease?.activeAttemptIds.delete(input.attemptId);
           const cleanup = async () => {
             let cleanupError = sessionInvalidationError;
             if (session && threadId) {
@@ -29717,7 +30050,7 @@ function createDefaultInferenceHostRunnerDependencies(options) {
     envelopePublicKey: options.envelopePublicKey
   };
 }
-var import_ajv, RUNTIME_RECEIPT_SCHEMA_VERSION, ATTEMPT_RECEIPT_SCHEMA_VERSION, DEFAULT_ADVERTISEMENT_TTL_MS, DEFAULT_ADVERTISEMENT_REFRESH_LEAD_MS, DEFAULT_HOST_HEARTBEAT_MS, DEFAULT_PROVIDER_RATE_LIMIT_REFRESH_MS, DEFAULT_ATTEMPT_HEARTBEAT_MS, DEFAULT_DRAIN_TIMEOUT_MS, DEFAULT_REMOTE_RETRY_LIMIT, MIN_SLEEP_MS, DEFAULT_CODEX_ACCOUNT_COOLDOWN_MS, MIN_CLAIM_START_WINDOW_MS, UNBOUNDED_AVAILABLE_SLOTS, buildInferenceAdvertisedModels, summarizeInferenceHostRuntimeRecovery, FileInferenceHostRuntimeReceiptStore, InferenceHostRecoveryRequiredError, InferenceHostRunnerError, defaultSleep, abortError, defaultRegisterSignalHandlers, safeInteger, exactObjectKeys, validIso, validateAttemptReceipt, validateRuntimeReceipt, sha256, stableOperationId, buildAttemptStartRequest, isoAt, providerWeeklyQuotaFromRateLimits, finitePositiveOption, validateRunnerOptions, assertLocalCredentialIdentity, retryableRemoteError, remoteRetryAfterMs, controlPlaneFatal, defaultValidateOutput, objectRecord, positiveUtf8ByteLimit, assertCompilableOutputSchema, parseOutputContract, membershipFailureDisposition, safeFailureCode, exactJson2, runnerIdentityMismatch, assertAdvertisementResult, assertHostHeartbeatResult, assertClaimResult, assertStartResult, assertJobHeartbeatResult, assertTerminalResult, classifyFailure, ambiguousHeartbeatTransport, reportedTokenUsage, reportedUsage, InferenceHostRunner;
+var import_ajv, RUNTIME_RECEIPT_SCHEMA_VERSION, ATTEMPT_RECEIPT_SCHEMA_VERSION, DEFAULT_ADVERTISEMENT_TTL_MS, DEFAULT_ADVERTISEMENT_REFRESH_LEAD_MS, DEFAULT_HOST_HEARTBEAT_MS, DEFAULT_PROVIDER_RATE_LIMIT_REFRESH_MS, DEFAULT_ATTEMPT_HEARTBEAT_MS, DEFAULT_DRAIN_TIMEOUT_MS, DEFAULT_REMOTE_RETRY_LIMIT, MIN_SLEEP_MS, MIN_HOST_HEARTBEAT_GAP_DIAGNOSTIC_MS, HOST_HEARTBEAT_GAP_DIAGNOSTIC_FACTOR, DEFAULT_CODEX_ACCOUNT_COOLDOWN_MS, MIN_CLAIM_START_WINDOW_MS, MAX_ATTEMPT_START_RETRY_DELAY_MS, UNBOUNDED_AVAILABLE_SLOTS, buildInferenceAdvertisedModels, summarizeInferenceHostRuntimeRecovery, FileInferenceHostRuntimeReceiptStore, InferenceHostRecoveryRequiredError, InferenceHostRunnerError, defaultSleep, abortError, defaultRegisterSignalHandlers, safeInteger, exactObjectKeys, validIso, validateAttemptReceipt, validateRuntimeReceipt, sha256, stableOperationId, buildAttemptStartRequest, isoAt, providerWeeklyQuotaFromRateLimits, finitePositiveOption, validateRunnerOptions, assertLocalCredentialIdentity, retryableRemoteError, remoteRetryAfterMs, controlPlaneFatal, defaultValidateOutput, objectRecord, positiveUtf8ByteLimit, assertCompilableOutputSchema, parseOutputContract, membershipFailureDisposition, safeFailureCode, attemptStartRetryFallbackMs, exactJson2, runnerIdentityMismatch, assertAdvertisementResult, assertHostHeartbeatResult, assertClaimResult, assertStartResult, assertJobHeartbeatResult, assertTerminalResult, classifyFailure, ambiguousHeartbeatTransport, reportedTokenUsage, reportedUsage, InferenceHostRunner;
 var init_runner = __esm({
   "lib/inference-host/runner.ts"() {
     "use strict";
@@ -29739,8 +30072,11 @@ var init_runner = __esm({
     DEFAULT_DRAIN_TIMEOUT_MS = 3e4;
     DEFAULT_REMOTE_RETRY_LIMIT = 3;
     MIN_SLEEP_MS = 10;
+    MIN_HOST_HEARTBEAT_GAP_DIAGNOSTIC_MS = 1e4;
+    HOST_HEARTBEAT_GAP_DIAGNOSTIC_FACTOR = 3;
     DEFAULT_CODEX_ACCOUNT_COOLDOWN_MS = 5 * 60 * 1e3;
     MIN_CLAIM_START_WINDOW_MS = 5e3;
+    MAX_ATTEMPT_START_RETRY_DELAY_MS = 15e3;
     UNBOUNDED_AVAILABLE_SLOTS = Number.MAX_SAFE_INTEGER;
     buildInferenceAdvertisedModels = (capabilities, adapterRuntimeVersion, adapter = "codex", structuredOutput = true, sameAttemptRecovery = adapter === "codex") => {
       const visible = capabilities.filter((capability) => !capability.hidden);
@@ -30276,6 +30612,10 @@ var init_runner = __esm({
       const normalized = value.trim().toLowerCase().replace(/[^a-z0-9._-]+/gu, "_").slice(0, 96);
       return /^[a-z0-9]/u.test(normalized) ? normalized : fallback;
     };
+    attemptStartRetryFallbackMs = (retryCount) => Math.min(
+      1e3 * 2 ** Math.min(Math.max(0, retryCount - 1), 4),
+      MAX_ATTEMPT_START_RETRY_DELAY_MS
+    );
     exactJson2 = (left, right) => JSON.stringify(left) === JSON.stringify(right);
     runnerIdentityMismatch = (operation) => {
       throw new InferenceHostRunnerError(
@@ -30729,6 +31069,26 @@ var init_runner = __esm({
             }
           };
           await hostHeartbeat("healthy");
+          let lastHostHeartbeatSucceededAt = now();
+          let hostHeartbeatGapOpen = false;
+          const hostHeartbeatGapThresholdMs = Math.max(
+            MIN_HOST_HEARTBEAT_GAP_DIAGNOSTIC_MS,
+            settings.hostHeartbeatMs * HOST_HEARTBEAT_GAP_DIAGNOSTIC_FACTOR
+          );
+          const observeHostHeartbeatGap = (observedAt) => {
+            const gapMs = Math.max(0, observedAt - lastHostHeartbeatSucceededAt);
+            if (gapMs >= hostHeartbeatGapThresholdMs && !hostHeartbeatGapOpen) {
+              hostHeartbeatGapOpen = true;
+              emitDiagnostic("host_heartbeat_gap_detected", {
+                gap_ms: gapMs,
+                configured_interval_ms: settings.hostHeartbeatMs,
+                diagnostic_threshold_ms: hostHeartbeatGapThresholdMs,
+                active_attempts: active.size,
+                host_heartbeat_sequence: receipt.host_heartbeat_sequence
+              });
+            }
+            return gapMs;
+          };
           let nextHostHeartbeatAt = now() + settings.hostHeartbeatMs;
           let nextAdvertisementAttemptAt = 0;
           let nextProviderRateLimitRefreshAt = now() + settings.providerRateLimitRefreshMs;
@@ -30736,6 +31096,7 @@ var init_runner = __esm({
           let pendingClaimPromotions = 0;
           let onceClaimed = false;
           const recoveryQueue = [];
+          const attemptStartRetryCounts = /* @__PURE__ */ new Map();
           const launchClaim = (claim, recovery, claimRequest) => {
             if (!recovery) {
               claimed += 1;
@@ -30748,6 +31109,10 @@ var init_runner = __esm({
               claim.attempt_index,
               claim.input_sha256
             ]);
+            const nextStartRetryCount = Math.min(
+              (attemptStartRetryCounts.get(attemptId) ?? 0) + 1,
+              Number.MAX_SAFE_INTEGER
+            );
             const controller = new AbortController();
             attemptControllers.set(attemptId, controller);
             let claimPromotionSettled = claimRequest === void 0;
@@ -30812,13 +31177,27 @@ var init_runner = __esm({
                   active_attempts: active.size
                 });
               },
+              startRetryCount: nextStartRetryCount,
+              onAttemptStartRetry: (observation) => {
+                attemptStartRetryCounts.set(attemptId, observation.retry_count);
+                emitDiagnostic("attempt_start_retry_scheduled", {
+                  ...observation,
+                  job_id: claim.job_id,
+                  attempt_id: attemptId,
+                  active_attempts: active.size
+                });
+              },
               resumeReceipt: recovery,
               claimRequest,
               onClaimPersisted: settleClaimPromotion
             }).then((outcome) => {
-              if (outcome === "completed") completed += 1;
-              else if (outcome === "failed") failed += 1;
-              else {
+              if (outcome === "completed") {
+                attemptStartRetryCounts.delete(attemptId);
+                completed += 1;
+              } else if (outcome === "failed") {
+                attemptStartRetryCounts.delete(attemptId);
+                failed += 1;
+              } else {
                 const recovery2 = receipt.attempts[attemptId];
                 if (recovery2?.phase !== "claimed") {
                   throw new InferenceHostRecoveryRequiredError(
@@ -30828,6 +31207,7 @@ var init_runner = __esm({
                 retryRecovery = recovery2;
               }
             }).catch((error48) => {
+              attemptStartRetryCounts.delete(attemptId);
               failed += 1;
               const recovery2 = receipt.attempts[attemptId];
               if (recovery2?.phase === "terminal_pending") {
@@ -30840,22 +31220,19 @@ var init_runner = __esm({
               attemptControllers.delete(attemptId);
               if (retryRecovery) {
                 recoveryQueue.push(retryRecovery);
-                emitDiagnostic("attempt_start_retry_scheduled", {
-                  job_id: claim.job_id,
-                  attempt_id: attemptId,
-                  active_attempts: active.size
-                });
               }
             });
             active.set(attemptId, promise2);
-            emitDiagnostic("attempt_started", {
-              job_id: claim.job_id,
-              attempt_id: attemptId,
-              attempt_index: claim.attempt_index,
-              requested_model: claim.requested_model,
-              requested_reasoning_effort: claim.requested_reasoning_effort,
-              active_attempts: active.size
-            });
+            if (!recovery) {
+              emitDiagnostic("attempt_started", {
+                job_id: claim.job_id,
+                attempt_id: attemptId,
+                attempt_index: claim.attempt_index,
+                requested_model: claim.requested_model,
+                requested_reasoning_effort: claim.requested_reasoning_effort,
+                active_attempts: active.size
+              });
+            }
             return attemptId;
           };
           const removeRecoveredAttempt = async (attemptId) => {
@@ -30978,6 +31355,7 @@ var init_runner = __esm({
               }
             }
             if (currentTime >= nextHostHeartbeatAt) {
+              observeHostHeartbeatGap(currentTime);
               try {
                 const refreshPendingAcrossExpiry = receipt.pending_advertisement !== null && expiresAt <= currentTime && nextAdvertisementAttemptAt > currentTime;
                 await hostHeartbeat(
@@ -30986,6 +31364,19 @@ var init_runner = __esm({
                   false,
                   refreshPendingAcrossExpiry
                 );
+                const heartbeatSucceededAt = now();
+                const gapMs = observeHostHeartbeatGap(heartbeatSucceededAt);
+                if (hostHeartbeatGapOpen) {
+                  emitDiagnostic("host_heartbeat_gap_recovered", {
+                    gap_ms: gapMs,
+                    configured_interval_ms: settings.hostHeartbeatMs,
+                    diagnostic_threshold_ms: hostHeartbeatGapThresholdMs,
+                    active_attempts: active.size,
+                    host_heartbeat_sequence: receipt.host_heartbeat_sequence
+                  });
+                  hostHeartbeatGapOpen = false;
+                }
+                lastHostHeartbeatSucceededAt = heartbeatSucceededAt;
               } catch (error48) {
                 if (controlPlaneFatal(error48)) {
                   requestDrain("authority_lost");
@@ -31262,10 +31653,28 @@ var init_runner = __esm({
               MIN_SLEEP_MS,
               attemptDeadlineAtMs - now() - MIN_CLAIM_START_WINDOW_MS
             );
-            await options.sleep(
-              Math.min(remoteRetryAfterMs(error48, 1e3), remainingMs),
-              signal
+            const retryCount = Math.min(
+              Math.max(1, options.startRetryCount),
+              Number.MAX_SAFE_INTEGER
             );
+            const retryDelayMs = Math.min(
+              remoteRetryAfterMs(error48, attemptStartRetryFallbackMs(retryCount)),
+              remainingMs
+            );
+            const rawErrorCode = error48 && typeof error48 === "object" && "code" in error48 ? String(error48.code) : "start_acknowledgement_failed";
+            options.onAttemptStartRetry?.({
+              failure_stage: "job_start_acknowledgement",
+              error_code: safeFailureCode(rawErrorCode, "start_acknowledgement_failed"),
+              http_status_code: error48 instanceof ExternalInferenceMcpError ? error48.httpStatusCode : null,
+              retryable: retryableRemoteError(error48),
+              definitively_not_applied: Boolean(
+                error48 && typeof error48 === "object" && "definitivelyNotApplied" in error48 && error48.definitivelyNotApplied === true
+              ),
+              retry_after_ms: error48 instanceof ExternalInferenceMcpError ? error48.retryAfterMs : null,
+              retry_count: retryCount,
+              next_delay_ms: retryDelayMs
+            });
+            await options.sleep(retryDelayMs, signal);
             return "retry_claimed";
           }
           assertStartResult(startRequest, startResult);
@@ -31527,6 +31936,7 @@ var init_runner = __esm({
             });
           }
           const classified = classifyFailure(heartbeatFailure ?? error48, adapterResult);
+          const processExit = heartbeatFailure === void 0 && error48 instanceof CodexAppServerError && error48.code === "codex_process_closed" && error48.processExit?.expected_shutdown === false ? error48.processExit : null;
           const heartbeatAmbiguous = ambiguousHeartbeatTransport(heartbeatFailure);
           const failure = heartbeatAmbiguous ? {
             ...classified,
@@ -31580,6 +31990,7 @@ var init_runner = __esm({
             failure_category: failure.category,
             failure_code: failure.code,
             retryable: failure.retryable,
+            ...processExit ? { process_exit: processExit } : {},
             membership_disposition: membershipFailureDisposition({
               ...failure,
               dispatchOutcome
@@ -31614,6 +32025,7 @@ var init_runner = __esm({
             failure_category: failureRequest.failure_category,
             failure_code: failureRequest.failure_code,
             retryable: failureRequest.retryable,
+            process_exit: failureRequest.process_exit ?? null,
             dispatch_outcome: failureRequest.outcome.dispatch_outcome,
             response_outcome: failureRequest.outcome.response_outcome,
             latency_ms: adapterResult?.latencyMs ?? null,
@@ -32983,8 +33395,8 @@ ${result2.stderr}`)) {
           return !nextWorker || !sameWorker(record2.worker, nextWorker);
         });
         for (const record2 of retiring) {
-          await updateRecord(record2, "draining");
           record2.controller.abort();
+          await updateRecord(record2, "draining");
         }
         const settled = await Promise.allSettled(retiring.map((record2) => record2.promise));
         const failedDrain = settled.find(
@@ -33018,8 +33430,8 @@ ${result2.stderr}`)) {
       const drainWorkers = async () => {
         const draining = [...workers.values()];
         for (const record2 of draining) {
-          await updateRecord(record2, "draining");
           record2.controller.abort();
+          await updateRecord(record2, "draining");
         }
         const settled = await Promise.allSettled(draining.map((record2) => record2.promise));
         const failedDrain = settled.find(
@@ -33607,6 +34019,17 @@ Durable service:
       }
     };
     defaultRunForeground = async (options) => {
+      const localState = await readInferenceHostLocalState(options.config.statePath);
+      const emitCodexDiagnostic = (event) => {
+        options.emitDiagnosticEvent?.({
+          ...event,
+          schema_version: "vtx_inference_host_diagnostic_v1",
+          at: (/* @__PURE__ */ new Date()).toISOString(),
+          instance_name: options.config.instanceName,
+          host_id: localState?.host_id ?? null,
+          display_name: options.displayName
+        });
+      };
       const adapter = new CodexSubscriptionAdapter({
         resolveBinary: async () => options.verifiedBinary,
         createAttemptResources: options.hostResources.createAttemptResources,
@@ -33614,7 +34037,8 @@ Durable service:
         guardianReceiptRoot: codexGuardianReceiptRoot(options.config),
         recoveryHooks: new FileCodexAttemptRecoveryStore(
           codexRecoveryPath(options.config)
-        )
+        ),
+        emitDiagnosticEvent: emitCodexDiagnostic
       });
       const dependencies = createDefaultInferenceHostRunnerDependencies({
         apiUrl: options.config.apiUrl,
@@ -34291,7 +34715,15 @@ Waiting for approval...
         const emitDiagnosticEvent = parsed.json ? (event) => (dependencies.emitStdout ?? ((text) => {
           process.stdout.write(text);
         }))(`${JSON.stringify(event)}
-`) : void 0;
+`) : adapterId === "codex" ? (event) => {
+          if (!["codex_app_server_closed", "codex_app_server_session_started"].includes(
+            String(event.event ?? "")
+          )) return;
+          (dependencies.emitStderr ?? ((text) => {
+            process.stderr.write(text);
+          }))(`[VTX inference-host] ${JSON.stringify(event)}
+`);
+        } : void 0;
         if (adapterId !== "codex") {
           const prepared = await (dependencies.prepareDurableAdapter ?? preparePortableDurableAdapter)(
             adapterId,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@vtxmacro/cli",
-  "version": "2026.8.43",
+  "version": "2026.8.45",
   "description": "VTX Macro CLI, MCP server, and durable subscription inference host.",
   "type": "module",
   "license": "UNLICENSED",