@vtxmacro/cli 2026.9.39 → 2026.9.41

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: "2026.9.39",
+  package_version: "2026.9.41",
   codex_package_name: "@openai/codex",
   codex_version: "0.153.3",
   copilot_sdk_package_name: "@github/copilot-sdk",

package/bin/vtx.js

@@ -77,7 +77,7 @@ var init_agent_cli_release = __esm({
   "agent-cli-release.json"() {
     agent_cli_release_default = {
       package_name: "@vtxmacro/cli",
-      package_version: "2026.9.39",
+      package_version: "2026.9.41",
       codex_package_name: "@openai/codex",
       codex_version: "0.153.3",
       copilot_sdk_package_name: "@github/copilot-sdk",
@@ -14722,7 +14722,7 @@ var init_zod = __esm({
 });
 
 // lib/external-inference-contract.ts
-var EXTERNAL_INFERENCE_CONTRACT_VERSION, EXTERNAL_INFERENCE_CONTROLLER_SCHEMA_VERSION, EXTERNAL_INFERENCE_CONTROLLER_SCHEMA_VERSION_V2, EXTERNAL_INFERENCE_ADVERTISEMENT_SCHEMA_VERSION, EXTERNAL_INFERENCE_ENVELOPE_SCHEMA_VERSION, EXTERNAL_INFERENCE_COMPRESSED_INPUT_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_PROFILE_EXTERNAL_INFERENCE_FALLBACKS_PER_LANE, 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, controlModeSchema, dispatchOutcomeSchema, responseOutcomeSchema, nativeModelControllerSchema, legacyExternalAgentModelControllerSchema, externalAgentModelControllerSchema, boundExternalAgentModelControllerSchema, externalAgentUnboundControllerSchema, externalAgentUnboundControllerV2Schema, 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, grokToolValidationDiagnosticSchema, grokResponseIdentityDiagnosticSchema, agentTurnReceiptSchema, agentConnectRequestSchema, agentNextRequestSchema, agentHeartbeatRequestSchema, emptyAgentNextResultSchema, claimedAgentNextResultSchema, agentNextResultSchema, agentCompleteRequestSchema, agentFailRequestSchema, requestedModelSelectionSchema, completedModelProvenanceSchema, attemptModelProvenanceSchema, providerWeeklyQuotaReadSchema, hostHeartbeatRequestSchema, jobClaimRequestBaseShape, jobClaimRequestSchema, emptyJobClaimSchema, claimedJobSchema, claimedDispatchIntentSchema, batchedJobClaimSchema, timedBatchedJobClaimSchema, 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 EXTERNAL_INFERENCE_CONTRACT_VERSION, EXTERNAL_INFERENCE_CONTROLLER_SCHEMA_VERSION, EXTERNAL_INFERENCE_CONTROLLER_SCHEMA_VERSION_V2, EXTERNAL_INFERENCE_ADVERTISEMENT_SCHEMA_VERSION, EXTERNAL_INFERENCE_ENVELOPE_SCHEMA_VERSION, EXTERNAL_INFERENCE_COMPRESSED_INPUT_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_PROFILE_EXTERNAL_INFERENCE_FALLBACKS_PER_LANE, 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, controlModeSchema, dispatchOutcomeSchema, responseOutcomeSchema, nativeModelControllerSchema, legacyExternalAgentModelControllerSchema, externalAgentModelControllerSchema, boundExternalAgentModelControllerSchema, externalAgentUnboundControllerSchema, externalAgentUnboundControllerV2Schema, 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, grokToolValidationDiagnosticSchema, grokResponseIdentityDiagnosticSchema, agentTurnReceiptSchema, agentConnectRequestSchema, agentNextRequestSchema, agentHeartbeatRequestSchema, emptyAgentNextResultSchema, claimedAgentNextResultSchema, agentNextResultSchema, agentCompleteRequestSchema, agentFailRequestSchema, requestedModelSelectionSchema, completedModelProvenanceSchema, attemptModelProvenanceSchema, providerWeeklyQuotaReadSchema, hostRuntimeRecoverySchema, hostHeartbeatRequestSchema, jobClaimRequestBaseShape, jobClaimRequestSchema, emptyJobClaimSchema, claimedJobSchema, claimedDispatchIntentSchema, batchedJobClaimSchema, timedBatchedJobClaimSchema, 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";
@@ -15738,6 +15738,30 @@ var init_external_inference_contract = __esm({
         });
       }
     });
+    hostRuntimeRecoverySchema = external_exports.strictObject({
+      phase: external_exports.enum(["running", "recovering", "draining"]),
+      drain_reason: external_exports.enum([
+        "stopped",
+        "sigint",
+        "sigterm",
+        "aborted",
+        "authority_lost",
+        "attempt_terminal_unconfirmed",
+        "agent_recovery_failed",
+        "host_heartbeat_failed",
+        "advertisement_failed",
+        "claim_failed",
+        "once_empty",
+        "once_complete",
+        "service_update",
+        "drain_timeout",
+        "provider_dispatch_watchdog_expired",
+        "receipt_durability_timeout"
+      ]).nullable(),
+      failure_code: external_exports.string().regex(/^[a-z0-9][a-z0-9._-]{0,95}$/u).nullable(),
+      pending_attempts: nonNegativeSafeIntegerSchema,
+      pending_claim: external_exports.boolean()
+    });
     hostHeartbeatRequestSchema = external_exports.strictObject({
       schema_version: external_exports.literal("external_inference_host_heartbeat_v1"),
       contract_version: external_exports.literal(EXTERNAL_INFERENCE_CONTRACT_VERSION),
@@ -15750,6 +15774,7 @@ var init_external_inference_contract = __esm({
       active_attempts: nonNegativeSafeIntegerSchema,
       available_slots: nonNegativeSafeIntegerSchema,
       provider_weekly_quota: providerWeeklyQuotaReadSchema.nullable().optional(),
+      runtime_recovery: hostRuntimeRecoverySchema.optional(),
       provider_quota_recovery: external_exports.strictObject({
         source: external_exports.literal("codex_app_server"),
         observed_at: timestampSchema
@@ -21011,6 +21036,7 @@ var init_mcp_client = __esm({
       #requestId = 0;
       #initialized = false;
       #providerQuotaRecoverySupported = false;
+      #runtimeRecoveryDiagnosticsSupported = false;
       #grokToolValidationTools = /* @__PURE__ */ new Set();
       #grokResponseIdentityTools = /* @__PURE__ */ new Set();
       supportsGrokToolValidationDiagnostics(tool) {
@@ -21037,6 +21063,9 @@ var init_mcp_client = __esm({
         }
         return this.supportsGrokResponseIdentityDiagnostics(tool);
       }
+      supportsRuntimeRecoveryDiagnostics() {
+        return this.#runtimeRecoveryDiagnosticsSupported;
+      }
       supportsProviderQuotaRecovery() {
         return this.#providerQuotaRecoverySupported;
       }
@@ -21090,6 +21119,10 @@ var init_mcp_client = __esm({
           const properties = schema.properties;
           return properties !== null && typeof properties === "object" && fields.every((field) => Object.hasOwn(properties, field));
         };
+        this.#runtimeRecoveryDiagnosticsSupported = advertisesFields(
+          "inference.host.heartbeat",
+          ["runtime_recovery"]
+        );
         this.#providerQuotaRecoverySupported = advertisesFields(
           "inference.host.heartbeat",
           ["provider_quota_recovery"]
@@ -31430,6 +31463,8 @@ var init_runner = __esm({
         });
         let drainRequested = false;
         let drainReason = "stopped";
+        let terminalRecoveryFailureCode = null;
+        let acknowledgedRecoveryDiagnostic = null;
         let receipt;
         let mcp;
         const active = /* @__PURE__ */ new Map();
@@ -31508,6 +31543,7 @@ var init_runner = __esm({
           const emitTerminalRecoveryFailure = (recovery, error48) => {
             const cause = error48 instanceof InferenceHostRunnerError && error48.code === "terminal_outcome_unconfirmed" && error48.cause ? error48.cause : error48;
             const rawCode = cause instanceof ExternalInferenceMcpError || cause instanceof InferenceHostRunnerError || cause instanceof CodexAppServerError ? cause.code : "terminal_recovery_failed";
+            terminalRecoveryFailureCode = cause instanceof ExternalInferenceMcpError ? publicServerFailureCode(cause.serverFailureCode) ?? safeFailureCode(rawCode, "terminal_recovery_failed") : safeFailureCode(rawCode, "terminal_recovery_failed");
             const terminal = recovery.terminal_request;
             emitDiagnostic("terminal_recovery_failed", {
               job_id: recovery.job_id,
@@ -32023,7 +32059,17 @@ var init_runner = __esm({
             let request;
             await mutateReceipt((current) => {
               const sequence = current.host_heartbeat_sequence + 1;
+              const pendingAttempts = Object.keys(current.attempts).length;
+              const diagnostic = {
+                phase: status === "draining" ? "draining" : pendingAttempts > active.size || current.pending_claim_request ? "recovering" : "running",
+                drain_reason: status === "draining" ? drainReason : null,
+                failure_code: terminalRecoveryFailureCode,
+                pending_attempts: pendingAttempts,
+                pending_claim: Boolean(current.pending_claim_request)
+              };
+              const parsedDiagnostic = hostRuntimeRecoverySchema.safeParse(diagnostic);
               request = {
+                ...mcp.supportsRuntimeRecoveryDiagnostics?.() === true && parsedDiagnostic.success && JSON.stringify(parsedDiagnostic.data) !== acknowledgedRecoveryDiagnostic ? { runtime_recovery: parsedDiagnostic.data } : {},
                 schema_version: "external_inference_host_heartbeat_v1",
                 contract_version: EXTERNAL_INFERENCE_CONTRACT_VERSION,
                 host_id: localState.host_id,
@@ -32074,6 +32120,9 @@ var init_runner = __esm({
               throw error48;
             }
             assertHostHeartbeatResult(request, result2, settings.adapterId, allowOfflineProjection);
+            if (request.runtime_recovery) {
+              acknowledgedRecoveryDiagnostic = JSON.stringify(request.runtime_recovery);
+            }
             if (result2.host.status === "revoked" || result2.host.status === "offline" && !allowOfflineProjection) {
               const error48 = new InferenceHostRunnerError(
                 result2.host.status === "revoked" ? "host_revoked" : "host_offline",
@@ -34474,9 +34523,18 @@ var init_runner = __esm({
               };
             }
           });
+          const wakeScheduledAtMs = this.now();
+          let nextWakeAtMs = null;
+          let wakeError;
+          try {
+            nextWakeAtMs = wakeScheduledAtMs + parseInferenceAgentWake(result2.turn.text, assignment) * 1e3;
+          } catch (error48) {
+            wakeError = error48;
+          }
+          const scheduledWakeAt = nextWakeAtMs === null ? null : new Date(nextWakeAtMs).toISOString();
           state = { ...state, pending_turn: {
             ...state.pending_turn,
-            next_wake_at: null,
+            next_wake_at: scheduledWakeAt,
             terminal_receipt: {
               terminal_status: "completed",
               failure_code: null,
@@ -34495,12 +34553,7 @@ var init_runner = __esm({
             }
           }, updated_at: nowIso() };
           await writeInferenceAgentRuntimeState(this.options.statePath, state);
-          const wakeSeconds = parseInferenceAgentWake(result2.turn.text, assignment);
-          const wakeScheduledAtMs = this.now();
-          const nextWakeAtMs = wakeScheduledAtMs + wakeSeconds * 1e3;
-          const scheduledWakeAt = new Date(nextWakeAtMs).toISOString();
-          state = { ...state, pending_turn: { ...state.pending_turn, next_wake_at: scheduledWakeAt } };
-          await writeInferenceAgentRuntimeState(this.options.statePath, state);
+          if (nextWakeAtMs === null || scheduledWakeAt === null) throw wakeError;
           const wakeHeartbeat = await this.options.controlClient.heartbeat({
             operation_id: randomUUID(),
             host_id: this.options.hostId,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@vtxmacro/cli",
-  "version": "2026.9.39",
+  "version": "2026.9.41",
   "description": "VTX Macro CLI, MCP server, and durable external inference host.",
   "type": "module",
   "license": "UNLICENSED",