@vtxmacro/cli 2026.9.66 → 2026.9.68

package/README.md

@@ -305,6 +305,7 @@ agent chooses its own cadence and may request zero, one, or many data calls:
 
 ```bash
 vtx inference-host agent-assignment-next --wait-seconds 50 --json
+vtx inference-host agent-review-begin --json
 vtx inference-host agent-data-call --json
 vtx inference-host agent-decision-submit --json
 vtx inference-host agent-decision-status --json

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: "2026.9.66",
+  package_version: "2026.9.68",
   codex_package_name: "@openai/codex",
   codex_version: "0.155.0",
   copilot_sdk_package_name: "@github/copilot-sdk",
@@ -133,6 +133,14 @@ var runtime_policy_generated_default = {
       min_delay_milliseconds: 250,
       max_delay_milliseconds: 1e4
     },
+    trade_refresh_policy: {
+      server_config_stream_timeout_ms: 1e4,
+      config_retry_interval_ms: 5e3,
+      copy_trading_status_poll_ms: 5e3,
+      post_execution_refresh_delay_ms: 3e3,
+      account_read_backoff_min_ms: 2e3,
+      account_read_backoff_max_ms: 6e4
+    },
     deployment_reload_observability: {
       bootstrap_max_error_events_per_page: 4,
       bootstrap_max_profiles: 10
@@ -210,10 +218,12 @@ var runtime_policy_generated_default = {
       },
       server_runtime: {
         prompt_rescue: {
+          request_ttl_seconds: 20,
           snapshot_max_age_seconds: 5,
           snapshot_component_max_skew_seconds: 2,
-          client_long_poll_seconds: 20,
-          client_claim_http_timeout_milliseconds: 25e3,
+          client_claim_http_timeout_milliseconds: 5e3,
+          client_claim_retry_delay_milliseconds: 1e3,
+          client_failure_report_timeout_milliseconds: 2e3,
           client_presence_heartbeat_interval_milliseconds: 3e4,
           client_presence_http_timeout_milliseconds: 5e3
         }

package/bin/vtx-service-launcher.js

@@ -16,6 +16,14 @@ var runtime_policy_generated_default = {
       min_delay_milliseconds: 250,
       max_delay_milliseconds: 1e4
     },
+    trade_refresh_policy: {
+      server_config_stream_timeout_ms: 1e4,
+      config_retry_interval_ms: 5e3,
+      copy_trading_status_poll_ms: 5e3,
+      post_execution_refresh_delay_ms: 3e3,
+      account_read_backoff_min_ms: 2e3,
+      account_read_backoff_max_ms: 6e4
+    },
     deployment_reload_observability: {
       bootstrap_max_error_events_per_page: 4,
       bootstrap_max_profiles: 10
@@ -93,10 +101,12 @@ var runtime_policy_generated_default = {
       },
       server_runtime: {
         prompt_rescue: {
+          request_ttl_seconds: 20,
           snapshot_max_age_seconds: 5,
           snapshot_component_max_skew_seconds: 2,
-          client_long_poll_seconds: 20,
-          client_claim_http_timeout_milliseconds: 25e3,
+          client_claim_http_timeout_milliseconds: 5e3,
+          client_claim_retry_delay_milliseconds: 1e3,
+          client_failure_report_timeout_milliseconds: 2e3,
           client_presence_heartbeat_interval_milliseconds: 3e4,
           client_presence_http_timeout_milliseconds: 5e3
         }

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: "2026.9.66",
+      package_version: "2026.9.68",
       codex_package_name: "@openai/codex",
       codex_version: "0.155.0",
       copilot_sdk_package_name: "@github/copilot-sdk",
@@ -15727,6 +15727,8 @@ var init_external_inference_contract = __esm({
       }
     });
     providerWeeklyQuotaReadSchema = external_exports.strictObject({
+      credits_available: external_exports.boolean().nullable().optional(),
+      rate_limit_reached: external_exports.boolean().nullable().optional(),
       source: external_exports.literal("codex_app_server"),
       used_percent: external_exports.number().finite().min(0).max(1e6),
       resets_at: timestampSchema,
@@ -20516,6 +20518,14 @@ var init_runtime_policy_generated = __esm({
           min_delay_milliseconds: 250,
           max_delay_milliseconds: 1e4
         },
+        trade_refresh_policy: {
+          server_config_stream_timeout_ms: 1e4,
+          config_retry_interval_ms: 5e3,
+          copy_trading_status_poll_ms: 5e3,
+          post_execution_refresh_delay_ms: 3e3,
+          account_read_backoff_min_ms: 2e3,
+          account_read_backoff_max_ms: 6e4
+        },
         deployment_reload_observability: {
           bootstrap_max_error_events_per_page: 4,
           bootstrap_max_profiles: 10
@@ -20593,10 +20603,12 @@ var init_runtime_policy_generated = __esm({
           },
           server_runtime: {
             prompt_rescue: {
+              request_ttl_seconds: 20,
               snapshot_max_age_seconds: 5,
               snapshot_component_max_skew_seconds: 2,
-              client_long_poll_seconds: 20,
-              client_claim_http_timeout_milliseconds: 25e3,
+              client_claim_http_timeout_milliseconds: 5e3,
+              client_claim_retry_delay_milliseconds: 1e3,
+              client_failure_report_timeout_milliseconds: 2e3,
               client_presence_heartbeat_interval_milliseconds: 3e4,
               client_presence_http_timeout_milliseconds: 5e3
             }
@@ -20773,7 +20785,7 @@ var init_mcp_transport = __esm({
 });
 
 // lib/inference-host/mcp-client.ts
-var EXTERNAL_INFERENCE_MCP_PROTOCOL_VERSION, EXTERNAL_INFERENCE_AUTOMATED_HOST_TOOLS, EXTERNAL_INFERENCE_OPERATIONAL_TOOLS, ClaimTiming, ExternalInferenceMcpError, DEFAULT_REQUEST_TIMEOUT_MS, MAX_RETRY_AFTER_MS, PUBLIC_TOOL_ERROR_PREFIX, parseRetryAfterMs, identifierSchema2, safeCodeSchema2, publicFailureCodeSchema, publicToolErrorSchema, structuredPublicToolErrorSchema, parsePublicToolError, hostMutationResultSchema, attemptStartResultSchema, agentConnectResultSchema, agentHeartbeatResultSchema, timestampSchema2, positiveGenerationSchema, jsonObjectSchema, agentAssignmentNextArgumentsSchema, agentDataCapabilityDescriptorSchema, agentAssignmentNextResultSchema, agentAssignmentHeartbeatArgumentsSchema, agentAssignmentHeartbeatResultSchema, agentDataCallArgumentsSchema, agentDataCallResultSchema, agentDecisionSubmitArgumentsSchema, agentDecisionSubmitResultSchema, agentDecisionStatusArgumentsSchema, agentDecisionStatusResultSchema, agentAssignmentReleaseArgumentsSchema, agentAssignmentReleaseResultSchema, jobCompletionResultSchema, jobFailureResultSchema, toolContracts, discoveryResultSchema, toolListResultSchema, inlineStructuredContentSchema, protocolMeta, parseJsonDocument, parseSseDocuments, parseResponseDocuments, exactOperationalInventory, exactJson, invalidBoundResult, verifyToolResult, ExternalInferenceMcpClient;
+var EXTERNAL_INFERENCE_MCP_PROTOCOL_VERSION, EXTERNAL_INFERENCE_AUTOMATED_HOST_TOOLS, EXTERNAL_INFERENCE_OPERATIONAL_TOOLS, ClaimTiming, ExternalInferenceMcpError, DEFAULT_REQUEST_TIMEOUT_MS, MAX_RETRY_AFTER_MS, PUBLIC_TOOL_ERROR_PREFIX, parseRetryAfterMs, identifierSchema2, safeCodeSchema2, publicFailureCodeSchema, publicToolErrorSchema, structuredPublicToolErrorSchema, parsePublicToolError, hostMutationResultSchema, attemptStartResultSchema, agentConnectResultSchema, agentHeartbeatResultSchema, timestampSchema2, positiveGenerationSchema, jsonObjectSchema, agentAssignmentNextArgumentsSchema, agentDataCapabilityDescriptorSchema, agentAssignmentNextResultSchema, agentAssignmentHeartbeatArgumentsSchema, agentAssignmentHeartbeatResultSchema, agentDataCallArgumentsSchema, agentDataCallResultSchema, agentDecisionTimingSchema, agentDecisionSubmitArgumentsSchema, agentDecisionSubmitResultSchema, agentDecisionStatusArgumentsSchema, agentDecisionStatusResultSchema, agentAssignmentReleaseArgumentsSchema, agentAssignmentReleaseResultSchema, jobCompletionResultSchema, jobFailureResultSchema, toolContracts, discoveryResultSchema, toolListResultSchema, inlineStructuredContentSchema, protocolMeta, parseJsonDocument, parseSseDocuments, parseResponseDocuments, exactOperationalInventory, exactJson, invalidBoundResult, verifyToolResult, ExternalInferenceMcpClient;
 var init_mcp_client = __esm({
   "lib/inference-host/mcp-client.ts"() {
     "use strict";
@@ -21096,13 +21108,23 @@ var init_mcp_client = __esm({
       warnings: external_exports.array(external_exports.unknown()),
       sources: external_exports.array(external_exports.unknown())
     });
+    agentDecisionTimingSchema = external_exports.strictObject({
+      analysis_started_at: timestampSchema2,
+      model_latency_ms: external_exports.number().int().nonnegative().nullable(),
+      model_latency_source: external_exports.literal("harness_invocation").nullable()
+    }).refine((timing) => timing.model_latency_ms === null === (timing.model_latency_source === null), {
+      message: "Model latency and its measurement source must be supplied together."
+    });
     agentDecisionSubmitArgumentsSchema = external_exports.strictObject({
       assignment_id: identifierSchema2,
       assignment_generation: positiveGenerationSchema,
       operation_id: identifierSchema2,
       candidate: jsonObjectSchema,
       observed_at: timestampSchema2,
+      timing: agentDecisionTimingSchema.optional(),
       provenance: jsonObjectSchema.nullable().optional()
+    }).refine((request) => !request.timing || Date.parse(request.timing.analysis_started_at) <= Date.parse(request.observed_at), {
+      message: "Analysis start must precede decision observation."
     });
     agentDecisionSubmitResultSchema = external_exports.strictObject({
       assignment_id: identifierSchema2,
@@ -21454,6 +21476,7 @@ var init_mcp_client = __esm({
       #requestId = 0;
       #initialized = false;
       #providerQuotaRecoverySupported = false;
+      #providerCreditsSupported = false;
       #runtimeRecoveryDiagnosticsSupported = false;
       #providerHttpStatusTools = /* @__PURE__ */ new Set();
       #grokToolValidationTools = /* @__PURE__ */ new Set();
@@ -21500,6 +21523,9 @@ var init_mcp_client = __esm({
       supportsRuntimeRecoveryDiagnostics() {
         return this.#runtimeRecoveryDiagnosticsSupported;
       }
+      supportsProviderCredits() {
+        return this.#providerCreditsSupported;
+      }
       supportsProviderQuotaRecovery() {
         return this.#providerQuotaRecoverySupported;
       }
@@ -21562,6 +21588,21 @@ var init_mcp_client = __esm({
           "inference.host.heartbeat",
           ["provider_quota_recovery"]
         ) && advertisesFields("inference.job.fail", ["provider_retry_at", "provider_retry_source"]);
+        const hostHeartbeatSchema = listed.tools.find(
+          (tool) => tool.name === "inference.host.heartbeat"
+        )?.inputSchema;
+        const advertisesCredits = (value, depth = 0) => {
+          if (depth > 4 || !value || typeof value !== "object" || Array.isArray(value)) return false;
+          const node = value;
+          const properties = node.properties;
+          if (properties && typeof properties === "object" && Object.hasOwn(properties, "credits_available") && Object.hasOwn(properties, "rate_limit_reached")) return true;
+          if (typeof node.$ref === "string" && /^#\/\$defs\/[^/]+$/u.test(node.$ref)) {
+            const definition = node.$ref.slice("#/$defs/".length).replace(/~1/gu, "/").replace(/~0/gu, "~");
+            return advertisesCredits(hostHeartbeatSchema?.$defs?.[definition], depth + 1);
+          }
+          return Array.isArray(node.anyOf) && node.anyOf.some((branch) => advertisesCredits(branch, depth + 1));
+        };
+        this.#providerCreditsSupported = advertisesCredits(hostHeartbeatSchema?.properties?.provider_weekly_quota);
         this.#grokToolValidationTools.clear();
         if (advertisesFields("inference.job.fail", ["grok_tool_validation"])) {
           this.#grokToolValidationTools.add("inference.job.fail");
@@ -28345,7 +28386,7 @@ async function logoutCodexSubscription(options) {
     await session.close();
   }
 }
-var CODEX_INFERENCE_PERMISSION_PROFILE, CODEX_AGENT_PERMISSION_PROFILE, CODEX_REASONING_CONTENT_MAX_UTF8_BYTES, CODEX_REASONING_SUMMARY_MAX_UTF8_BYTES, CODEX_ACCOUNT_PLAN_TYPES, CODEX_AGENT_DYNAMIC_TOOLS, CodexAppServerError, objectOrNull, finiteToken, tokenUsageFromBreakdown, usageFromNotification, validPlanType, nonnegativeSafeIntegerOrNull, nonnegativeFiniteNumberOrNull, parseRateLimitWindow, RATE_LIMIT_REACHED_TYPES, DEFAULT_CODEX_QUOTA_COOLDOWN_MS, MAX_CODEX_QUOTA_COOLDOWN_MS, CODEX_TRANSIENT_RATE_LIMIT_COOLDOWN_MS, parseRateLimitSnapshot, codexRateLimitRetryAtMs, codexAccountRateLimitReached, forbiddenMethod, forbiddenTerminalItem, forbiddenAgentMethod, forbiddenAgentTerminalItem, createCodexAgentToolSpecs, classifyCodexTurnFailure, scrubbedCodexEnvironment, killWindowsProcessTree, appServerArgs, GUARDIAN_SCRIPT, WINDOWS_RECEIPT_REPLACE_ERROR_CODES, MAX_RETIRED_RESPONSE_IDS, LATE_RESPONSE_SAFE_METHODS, replaceCodexGuardianReceiptFile, writeCodexGuardianSpawnIntent, parseGuardianReceipt, readCodexGuardianReceipt, waitForCodexGuardianState, defaultSpawn, CodexAppServerSession;
+var CODEX_INFERENCE_PERMISSION_PROFILE, CODEX_AGENT_PERMISSION_PROFILE, CODEX_REASONING_CONTENT_MAX_UTF8_BYTES, CODEX_REASONING_SUMMARY_MAX_UTF8_BYTES, CODEX_ACCOUNT_PLAN_TYPES, CODEX_AGENT_DYNAMIC_TOOLS, CodexAppServerError, objectOrNull, finiteToken, tokenUsageFromBreakdown, usageFromNotification, validPlanType, nonnegativeSafeIntegerOrNull, nonnegativeFiniteNumberOrNull, parseRateLimitWindow, RATE_LIMIT_REACHED_TYPES, DEFAULT_CODEX_QUOTA_COOLDOWN_MS, MAX_CODEX_QUOTA_COOLDOWN_MS, CODEX_TRANSIENT_RATE_LIMIT_COOLDOWN_MS, parseCreditsSnapshot, codexCreditsAvailable, parseRateLimitSnapshot, codexRateLimitRetryAtMs, codexAccountRateLimitReached, forbiddenMethod, forbiddenTerminalItem, forbiddenAgentMethod, forbiddenAgentTerminalItem, createCodexAgentToolSpecs, classifyCodexTurnFailure, scrubbedCodexEnvironment, killWindowsProcessTree, appServerArgs, GUARDIAN_SCRIPT, WINDOWS_RECEIPT_REPLACE_ERROR_CODES, MAX_RETIRED_RESPONSE_IDS, LATE_RESPONSE_SAFE_METHODS, replaceCodexGuardianReceiptFile, writeCodexGuardianSpawnIntent, parseGuardianReceipt, readCodexGuardianReceipt, waitForCodexGuardianState, defaultSpawn, CodexAppServerSession;
 var init_codex_app_server = __esm({
   "lib/inference-host/codex-app-server.ts"() {
     "use strict";
@@ -28460,6 +28501,19 @@ var init_codex_app_server = __esm({
     DEFAULT_CODEX_QUOTA_COOLDOWN_MS = 5 * 60 * 1e3;
     MAX_CODEX_QUOTA_COOLDOWN_MS = 8 * 24 * 60 * 60 * 1e3;
     CODEX_TRANSIENT_RATE_LIMIT_COOLDOWN_MS = 30 * 1e3;
+    parseCreditsSnapshot = (value) => {
+      const candidate = objectOrNull(value);
+      if (!candidate || typeof candidate.hasCredits !== "boolean" || typeof candidate.unlimited !== "boolean" || candidate.balance != null && typeof candidate.balance !== "string") return null;
+      return {
+        hasCredits: candidate.hasCredits,
+        unlimited: candidate.unlimited,
+        balance: candidate.balance ?? null
+      };
+    };
+    codexCreditsAvailable = (snapshot) => {
+      const credits = parseCreditsSnapshot(snapshot.credits);
+      return credits !== null && (credits.hasCredits || credits.unlimited);
+    };
     parseRateLimitSnapshot = (value) => {
       const candidate = objectOrNull(value);
       if (!candidate) throw new Error("invalid rate-limit snapshot");
@@ -28473,6 +28527,7 @@ var init_codex_app_server = __esm({
         limitId,
         limitName,
         planType,
+        credits: parseCreditsSnapshot(candidate.credits),
         primary: parseRateLimitWindow(candidate.primary ?? null),
         secondary: parseRateLimitWindow(candidate.secondary ?? null),
         rateLimitReachedType: reached,
@@ -28480,6 +28535,7 @@ var init_codex_app_server = __esm({
       };
     };
     codexRateLimitRetryAtMs = (accountRateLimits) => {
+      if (!codexAccountRateLimitReached(accountRateLimits)) return null;
       const snapshots = [accountRateLimits.rateLimits];
       const exhaustedResets = [];
       const reachedResets = [];
@@ -28500,7 +28556,12 @@ var init_codex_app_server = __esm({
       const candidates = exhaustedResets.length > 0 ? exhaustedResets : reachedResets.length > 0 ? reachedResets : [];
       return candidates.length > 0 ? Math.max(...candidates) : null;
     };
-    codexAccountRateLimitReached = (accountRateLimits) => [accountRateLimits.rateLimits].some((snapshot) => snapshot.rateLimitReachedType !== null || snapshot.spendControlReached === true || snapshot.primary?.usedPercent !== void 0 && snapshot.primary.usedPercent >= 100 || snapshot.secondary?.usedPercent !== void 0 && snapshot.secondary.usedPercent >= 100);
+    codexAccountRateLimitReached = (accountRateLimits) => [accountRateLimits.rateLimits].some((snapshot) => (
+      // The account snapshot reports generic included-usage exhaustion even when
+      // purchased credits remain. Workspace restrictions are distinct hard stops;
+      // actual turn failures remain authoritative in classifyCodexTurnFailure.
+      snapshot.rateLimitReachedType !== null && (snapshot.rateLimitReachedType !== "rate_limit_reached" || !codexCreditsAvailable(snapshot)) || snapshot.spendControlReached === true || !codexCreditsAvailable(snapshot) && (snapshot.primary?.usedPercent !== void 0 && snapshot.primary.usedPercent >= 100 || snapshot.secondary?.usedPercent !== void 0 && snapshot.secondary.usedPercent >= 100)
+    ));
     forbiddenMethod = (method) => method.startsWith("mcpServer/") || method.startsWith("serverRequest/") || method.startsWith("item/mcpToolCall") || method.startsWith("item/commandExecution") || method.startsWith("item/fileChange") || method.startsWith("item/dynamicToolCall") || method.startsWith("item/collabAgentToolCall");
     forbiddenTerminalItem = (item) => ![
       "userMessage",
@@ -30840,7 +30901,7 @@ async function clearInferenceAgentNextState(statePath) {
     "Agent-driven inference next recovery state"
   );
 }
-var inferenceAgentAttemptStatePath, inferenceAgentNextStatePath, inferenceAgentRuntimeStatePath, inferenceAgentFailureFencePath, isIsoTimestamp, assertPlainObject, hasExactKeys, assertInferenceAgentFailureFence, assertInferenceAgentRuntimeState, readCodexAgentRuntimeState, clearCodexAgentRuntimeState, foregroundAgentControlStatePath, assertRuntimeAssignment, assertForegroundAgentControlState, assertNextState, assertAttemptState;
+var inferenceAgentAttemptStatePath, inferenceAgentNextStatePath, assertDecisionTiming, inferenceAgentRuntimeStatePath, inferenceAgentFailureFencePath, isIsoTimestamp, assertPlainObject, hasExactKeys, assertInferenceAgentFailureFence, assertInferenceAgentRuntimeState, readCodexAgentRuntimeState, clearCodexAgentRuntimeState, foregroundAgentControlStatePath, assertRuntimeAssignment, assertForegroundAgentControlState, assertNextState, assertAttemptState;
 var init_agent_state = __esm({
   "lib/inference-host/agent-state.ts"() {
     "use strict";
@@ -30851,6 +30912,10 @@ var init_agent_state = __esm({
     init_external_inference_contract();
     inferenceAgentAttemptStatePath = (statePath) => `${statePath}.agent-attempt.json`;
     inferenceAgentNextStatePath = (statePath) => `${statePath}.agent-next.json`;
+    assertDecisionTiming = (value, observedAt, message) => {
+      const timing = assertPlainObject(value, message);
+      if (!hasExactKeys(timing, ["analysis_started_at", "model_latency_ms", "model_latency_source"]) || !isIsoTimestamp(timing.analysis_started_at) || Date.parse(timing.analysis_started_at) > Date.parse(String(observedAt)) || (timing.model_latency_ms === null ? timing.model_latency_source !== null : !Number.isSafeInteger(timing.model_latency_ms) || Number(timing.model_latency_ms) < 0 || timing.model_latency_source !== "harness_invocation")) throw new Error(message);
+    };
     inferenceAgentRuntimeStatePath = (statePath) => `${statePath}.codex-agent-runtime.json`;
     inferenceAgentFailureFencePath = (statePath) => `${statePath}.agent-failure-fence.json`;
     isIsoTimestamp = (value) => typeof value === "string" && Number.isFinite(Date.parse(value));
@@ -30962,9 +31027,14 @@ var init_agent_state = __esm({
           "observed_at",
           "provenance",
           "first_transmit_at",
-          "last_status_check_at"
+          "last_status_check_at",
+          ..."timing" in pending3 ? ["timing"] : []
         ]) || typeof pending3.operation_id !== "string" || !pending3.operation_id || !pending3.candidate || typeof pending3.candidate !== "object" || Array.isArray(pending3.candidate) || !isIsoTimestamp(pending3.observed_at) || pending3.provenance !== null && (typeof pending3.provenance !== "object" || Array.isArray(pending3.provenance)) || pending3.first_transmit_at !== null && !isIsoTimestamp(pending3.first_transmit_at) || pending3.last_status_check_at !== null && !isIsoTimestamp(pending3.last_status_check_at)) throw new Error(message);
       }
+      if (state.pending_decision && "timing" in state.pending_decision) {
+        const pending3 = state.pending_decision;
+        assertDecisionTiming(pending3.timing, pending3.observed_at, message);
+      }
       return state;
     };
     readCodexAgentRuntimeState = readInferenceAgentRuntimeState;
@@ -31012,9 +31082,15 @@ var init_agent_state = __esm({
         "pending_next",
         "pending_decision",
         "next_wake_at",
-        "updated_at"
+        "updated_at",
+        ..."review" in state ? ["review"] : []
       ]) || state.schema_version !== "vtx_foreground_agent_control_v1" || typeof state.host_id !== "string" || !state.host_id || state.next_wake_at !== null && !isIsoTimestamp(state.next_wake_at) || !isIsoTimestamp(state.updated_at)) throw new Error(message);
       if (state.assignment !== null) assertRuntimeAssignment(state.assignment, message);
+      if (state.review !== void 0 && state.review !== null) {
+        const review = assertPlainObject(state.review, message);
+        const assignment = state.assignment;
+        if (!hasExactKeys(review, ["assignment_id", "assignment_generation", "analysis_started_at"]) || !assignment || review.assignment_id !== assignment.assignment_id || review.assignment_generation !== assignment.assignment_generation || !isIsoTimestamp(review.analysis_started_at)) throw new Error(message);
+      }
       if (state.pending_next !== null) {
         const pending3 = assertPlainObject(state.pending_next, message);
         if (!hasExactKeys(pending3, ["operation_id", "requested_at"]) || typeof pending3.operation_id !== "string" || !pending3.operation_id || !isIsoTimestamp(pending3.requested_at)) throw new Error(message);
@@ -31028,13 +31104,15 @@ var init_agent_state = __esm({
           "last_status_check_at"
         ])) throw new Error(message);
         const request = assertPlainObject(pending3.request, message);
+        if ("timing" in request) assertDecisionTiming(request.timing, request.observed_at, message);
         if (!hasExactKeys(request, [
           "assignment_id",
           "assignment_generation",
           "operation_id",
           "candidate",
           "observed_at",
-          "provenance"
+          "provenance",
+          ..."timing" in request ? ["timing"] : []
         ]) || request.assignment_id !== state.assignment.assignment_id || request.assignment_generation !== state.assignment.assignment_generation || typeof request.operation_id !== "string" || !request.operation_id || !request.candidate || typeof request.candidate !== "object" || Array.isArray(request.candidate) || !isIsoTimestamp(request.observed_at) || !request.provenance || typeof request.provenance !== "object" || Array.isArray(request.provenance) || pending3.first_transmit_at !== null && !isIsoTimestamp(pending3.first_transmit_at) || pending3.last_status_check_at !== null && !isIsoTimestamp(pending3.last_status_check_at)) throw new Error(message);
       }
       return state;
@@ -31487,12 +31565,16 @@ var init_runner = __esm({
       started_at: startedAt
     });
     isoAt = (milliseconds) => new Date(milliseconds).toISOString();
-    providerWeeklyQuotaFromRateLimits = (rateLimits, observedAtMs, currentTimeMs = observedAtMs) => {
+    providerWeeklyQuotaFromRateLimits = (rateLimits, observedAtMs, currentTimeMs = observedAtMs, includeCredits = true) => {
       if (!rateLimits) return null;
       for (const window2 of [rateLimits.rateLimits.primary, rateLimits.rateLimits.secondary]) {
         if (window2?.windowDurationMins !== 7 * 24 * 60 || window2.resetsAt === null || window2.resetsAt * 1e3 <= currentTimeMs) continue;
         return {
           source: "codex_app_server",
+          ...includeCredits ? {
+            credits_available: codexCreditsAvailable(rateLimits.rateLimits),
+            rate_limit_reached: codexAccountRateLimitReached(rateLimits)
+          } : {},
           used_percent: window2.usedPercent,
           resets_at: isoAt(window2.resetsAt * 1e3),
           observed_at: isoAt(observedAtMs)
@@ -31504,7 +31586,7 @@ var init_runner = __esm({
       const window2 = rateLimits.rateLimits[key];
       const known = previous?.[key];
       if (known && (!window2 || window2.windowDurationMins !== known.windowDurationMins)) return false;
-      return window2 === null || window2.resetsAt !== null && window2.resetsAt * 1e3 > nowMs && window2.usedPercent < 100;
+      return window2 === null || window2.resetsAt !== null && window2.resetsAt * 1e3 > nowMs && (window2.usedPercent < 100 || codexCreditsAvailable(rateLimits.rateLimits));
     });
     finitePositiveOption = (value, fallback, label) => {
       const result2 = value ?? fallback;
@@ -32536,7 +32618,8 @@ var init_runner = __esm({
                 provider_weekly_quota: providerWeeklyQuotaFromRateLimits(
                   providerRateLimits,
                   providerRateLimitsObservedAtMs,
-                  now()
+                  now(),
+                  mcp.supportsProviderCredits?.() === true
                 ),
                 ...mcp.supportsProviderQuotaRecovery?.() === true && providerQuotaRecoveryObservedAtMs !== null && status === "healthy" && active.size === 0 && (providerRetryAtMs ?? 0) <= now() && providerFailureRetryAtMs <= now() && now() - providerQuotaRecoveryObservedAtMs <= 2 * settings.providerRateLimitRefreshMs && providerRateLimits !== null && usableProviderQuotaSnapshot(providerRateLimits, knownProviderQuotaSnapshot, now()) ? { provider_quota_recovery: {
                   source: "codex_app_server",
@@ -33154,14 +33237,7 @@ var init_runner = __esm({
                     providerRetryAtMs = Math.max(providerRetryAtMs ?? 0, refreshedRetryAt);
                     providerCooldownReason = "provider_limit_snapshot";
                     retainQuotaSnapshot(providerRateLimits);
-                  } else if (providerRetryAtMs !== null && !codexAccountRateLimitReached(providerRateLimits) && providerRateLimits.rateLimits.limitId === providerQuotaSnapshot?.limitId && [providerRateLimits.rateLimits.primary, providerRateLimits.rateLimits.secondary].some((window2) => window2 !== null) && ["primary", "secondary"].every((key) => {
-                    const window2 = refreshedRateLimits.rateLimits[key];
-                    const previous = providerQuotaSnapshot?.[key];
-                    if (previous && (!window2 || window2.windowDurationMins !== previous.windowDurationMins)) {
-                      return false;
-                    }
-                    return window2 === null || window2.resetsAt !== null && window2.resetsAt * 1e3 > refreshedAt && window2.usedPercent < 100;
-                  })) {
+                  } else if (providerRetryAtMs !== null && usableProviderQuotaSnapshot(providerRateLimits, providerQuotaSnapshot, refreshedAt)) {
                     emitDiagnostic("provider_cooldown_cleared", {
                       reason: providerCooldownReason,
                       recovery: "provider_quota_restored",
@@ -34956,6 +35032,11 @@ var init_runner = __esm({
                   operation_id: randomUUID(),
                   candidate,
                   observed_at: nowIso(),
+                  timing: {
+                    analysis_started_at: state.pending_turn.started_at,
+                    model_latency_ms: null,
+                    model_latency_source: null
+                  },
                   provenance: {
                     source: "external_agent",
                     agent_run_id: liveThread.threadId,
@@ -34993,6 +35074,7 @@ var init_runner = __esm({
                     operation_id: pending3.operation_id,
                     candidate: pending3.candidate,
                     observed_at: pending3.observed_at,
+                    ...pending3.timing ? { timing: pending3.timing } : {},
                     provenance: pending3.provenance
                   }, { signal: turnAbort.signal });
                 } catch {
@@ -42609,6 +42691,14 @@ var init_runtime_policy_generated2 = __esm({
           min_delay_milliseconds: 250,
           max_delay_milliseconds: 1e4
         },
+        trade_refresh_policy: {
+          server_config_stream_timeout_ms: 1e4,
+          config_retry_interval_ms: 5e3,
+          copy_trading_status_poll_ms: 5e3,
+          post_execution_refresh_delay_ms: 3e3,
+          account_read_backoff_min_ms: 2e3,
+          account_read_backoff_max_ms: 6e4
+        },
         deployment_reload_observability: {
           bootstrap_max_error_events_per_page: 4,
           bootstrap_max_profiles: 10
@@ -42686,10 +42776,12 @@ var init_runtime_policy_generated2 = __esm({
           },
           server_runtime: {
             prompt_rescue: {
+              request_ttl_seconds: 20,
               snapshot_max_age_seconds: 5,
               snapshot_component_max_skew_seconds: 2,
-              client_long_poll_seconds: 20,
-              client_claim_http_timeout_milliseconds: 25e3,
+              client_claim_http_timeout_milliseconds: 5e3,
+              client_claim_retry_delay_milliseconds: 1e3,
+              client_failure_report_timeout_milliseconds: 2e3,
               client_presence_heartbeat_interval_milliseconds: 3e4,
               client_presence_http_timeout_milliseconds: 5e3
             }
@@ -44537,6 +44629,13 @@ async function runInferenceHostCli(argv2, env = process.env, dependencies = {})
         dependencies
       );
     }
+    if (parsed.command === "agent-review-begin") {
+      return await withAgentCommandLock(
+        config2,
+        async () => await foregroundReviewBegin(config2, parsed, dependencies, warnings),
+        dependencies
+      );
+    }
     if (parsed.command === "agent-data-call") {
       return await foregroundDataCall(config2, parsed, dependencies, warnings);
     }
@@ -44597,7 +44696,7 @@ async function runInferenceHostCli(argv2, env = process.env, dependencies = {})
       return await cleanupLogin(config2, parsed, dependencies, warnings, true);
     }
     throw new Error(
-      "Usage: vtx inference-host <login|codex-login|grok-login|deepseek-login|pi-login|exo-login|run|agent-connect|agent-run|agent-next|agent-complete|agent-fail|agent-assignment-next|agent-assignment-heartbeat|agent-data-call|agent-decision-submit|agent-decision-status|agent-assignment-release|service|status|doctor|logout|codex-logout|grok-logout|deepseek-logout|pi-logout|exo-logout|revoke> [--json]"
+      "Usage: vtx inference-host <login|codex-login|grok-login|deepseek-login|pi-login|exo-login|run|agent-connect|agent-run|agent-next|agent-complete|agent-fail|agent-assignment-next|agent-assignment-heartbeat|agent-review-begin|agent-data-call|agent-decision-submit|agent-decision-status|agent-assignment-release|service|status|doctor|logout|codex-logout|grok-logout|deepseek-logout|pi-logout|exo-logout|revoke> [--json]"
     );
   } catch (error48) {
     return {
@@ -44608,7 +44707,7 @@ async function runInferenceHostCli(argv2, env = process.env, dependencies = {})
     };
   }
 }
-var INFERENCE_HOST_CLI_VERSION, runtimeReceiptPath, codexRecoveryPath, codexGuardianReceiptRoot, revocationCheckpointPath, foregroundHostLockPath, credentialLifecycleLockPath, serviceRecoveryCommandLockPath, AGENT_HEARTBEAT_INTERVAL_MS, AGENT_HEARTBEAT_MAX_RETRY_DELAY_MS, retryableAgentHeartbeatError, assertRevocationCheckpoint, readRevocationCheckpoint, writeRevocationCheckpoint, clearRevocationCheckpoint, render, INFERENCE_HOST_HELP, parseHostConcurrency, parseInferenceHostArgs, resolveDurableServiceDisplayName, defaultOpenBrowser, registerInferenceHostServiceControlInput, defaultRegisterLifecycleSignalHandlers, lifecycleCancellation, configuredCredentialStore, restoreInferenceHostCredentialContext, recoverInferenceHostCredentialContextTransition, accountKeyForState, assertCredentialMatchesState2, revocationCheckpointForCredential, assertRevocationCheckpointMatchesLocalIdentity, fileExistsPrivately, inspectCodexAuthentication, clearLocalRuntimeArtifacts, assertDurableServiceUninstalled, assertLocalRuntimeArtifactsMayBeDiscarded, defaultRunForeground, preparePortableDurableAdapter, prepareExoAdapter, defaultRunPortableDurableAdapter, login, codexLogin, codexLogout, grokAuthentication, deepSeekCredentialStore, deepSeekLogin, deepSeekLogout, piLogin, piLogout, exoLogin, exoLogout, localStatus, doctor, cleanupLogin, runHost, defaultReadStdin, parseAgentStdin, parseOptionalAgentStdin, agentOperationId, agentSession, agentConnect, foregroundAgentControlState, requireForegroundAgentState, assignmentFromClaim, foregroundAssignmentNext, foregroundAssignmentHeartbeat, foregroundDataCall, foregroundDecisionStatusRequest, foregroundAssignmentLeaseEnded, handleForegroundCancelledAssignment, foregroundDecisionFollowup, foregroundDecisionStatus, foregroundDecisionSubmit, foregroundAssignmentRelease, agentRun, agentNext, agentComplete, agentFail, withAgentCommandLock, recoveryBackupPath, serviceRecoveryTransactionPath, readServiceRecoveryTransaction, assertResumableCodexRecoveryEvidence, recoverInstalledCodexService, serviceCommand, hasExplicitCredentialStoreConfiguration, credentialEnvironmentForIdentity, commandRequiresRetainedCredentialStore, commandUsesInstalledServiceCredentials, resolveInferenceHostCommandConfig;
+var INFERENCE_HOST_CLI_VERSION, runtimeReceiptPath, codexRecoveryPath, codexGuardianReceiptRoot, revocationCheckpointPath, foregroundHostLockPath, credentialLifecycleLockPath, serviceRecoveryCommandLockPath, AGENT_HEARTBEAT_INTERVAL_MS, AGENT_HEARTBEAT_MAX_RETRY_DELAY_MS, retryableAgentHeartbeatError, assertRevocationCheckpoint, readRevocationCheckpoint, writeRevocationCheckpoint, clearRevocationCheckpoint, render, INFERENCE_HOST_HELP, parseHostConcurrency, parseInferenceHostArgs, resolveDurableServiceDisplayName, defaultOpenBrowser, registerInferenceHostServiceControlInput, defaultRegisterLifecycleSignalHandlers, lifecycleCancellation, configuredCredentialStore, restoreInferenceHostCredentialContext, recoverInferenceHostCredentialContextTransition, accountKeyForState, assertCredentialMatchesState2, revocationCheckpointForCredential, assertRevocationCheckpointMatchesLocalIdentity, fileExistsPrivately, inspectCodexAuthentication, clearLocalRuntimeArtifacts, assertDurableServiceUninstalled, assertLocalRuntimeArtifactsMayBeDiscarded, defaultRunForeground, preparePortableDurableAdapter, prepareExoAdapter, defaultRunPortableDurableAdapter, login, codexLogin, codexLogout, grokAuthentication, deepSeekCredentialStore, deepSeekLogin, deepSeekLogout, piLogin, piLogout, exoLogin, exoLogout, localStatus, doctor, cleanupLogin, runHost, defaultReadStdin, parseAgentStdin, parseOptionalAgentStdin, agentOperationId, agentSession, agentConnect, foregroundAgentControlState, requireForegroundAgentState, assignmentFromClaim, foregroundAssignmentNext, foregroundAssignmentHeartbeat, foregroundReviewBegin, foregroundDataCall, foregroundDecisionStatusRequest, foregroundAssignmentLeaseEnded, handleForegroundCancelledAssignment, foregroundDecisionFollowup, foregroundDecisionStatus, foregroundDecisionSubmit, foregroundAssignmentRelease, agentRun, agentNext, agentComplete, agentFail, withAgentCommandLock, recoveryBackupPath, serviceRecoveryTransactionPath, readServiceRecoveryTransaction, assertResumableCodexRecoveryEvidence, recoverInstalledCodexService, serviceCommand, hasExplicitCredentialStoreConfiguration, credentialEnvironmentForIdentity, commandRequiresRetainedCredentialStore, commandUsesInstalledServiceCredentials, resolveInferenceHostCommandConfig;
 var init_cli = __esm({
   "lib/inference-host/cli.ts"() {
     "use strict";
@@ -44729,6 +44828,7 @@ Commands:
   agent-fail      Submit one truthful agent failure from stdin
   agent-assignment-next       Claim the assigned Main bot for Agent control
   agent-assignment-heartbeat  Renew the current Agent assignment
+  agent-review-begin          Record review start before reading decision context
   agent-data-call             Request allowed assignment-scoped VTX data
   agent-decision-submit       Submit the normal structured trading decision
   agent-decision-status       Read a submitted decision outcome or recover uncertain submission
@@ -46446,6 +46546,7 @@ Waiting for approval...
             state = {
               ...state,
               assignment,
+              review: state.assignment?.assignment_id === assignment.assignment_id && state.assignment?.assignment_generation === assignment.assignment_generation ? state.review ?? null : null,
               pending_next: null,
               next_wake_at: null,
               updated_at: now().toISOString()
@@ -46462,7 +46563,8 @@ Waiting for approval...
             exitCode: 0,
             stdout: render({
               ...result2,
-              instructions: "Use your harness scheduler for the requested cadence and keep agent-run open for liveness. Use agent-data-call for canonical VTX data and your authorized native Insights/external tools. Trading JSON is optional; submit a decision only when you choose, using the exact output_schema. The keeper does not schedule or restart your harness.",
+              instructions: "Use your harness scheduler for the requested cadence and keep agent-run open for liveness. Call agent-review-begin before each review to record its actual analysis start. Use agent-data-call for canonical VTX data and your authorized native Insights/external tools. Trading JSON is optional; submit a decision only when you choose, using the exact output_schema. The keeper does not schedule or restart your harness.",
+              review_begin_command: "vtx inference-host agent-review-begin --json",
               data_command: "vtx inference-host agent-data-call --json",
               decision_command: "vtx inference-host agent-decision-submit --json",
               decision_input: {
@@ -46532,6 +46634,31 @@ Waiting for approval...
 ` : ""
       };
     };
+    foregroundReviewBegin = async (config2, parsed, dependencies, warnings) => {
+      const session = await agentSession(config2, dependencies, warnings);
+      const state = await requireForegroundAgentState(config2, session.localState.host_id);
+      if (state.pending_decision || state.pending_next) {
+        throw new Error("Resolve the pending Agent operation before beginning a review.");
+      }
+      await parseOptionalAgentStdin(dependencies, []);
+      const startedAt = (dependencies.now ?? (() => /* @__PURE__ */ new Date()))().toISOString();
+      const review = {
+        assignment_id: state.assignment.assignment_id,
+        assignment_generation: state.assignment.assignment_generation,
+        analysis_started_at: startedAt
+      };
+      await writeInferenceForegroundAgentControlState(config2.statePath, {
+        ...state,
+        review,
+        updated_at: startedAt
+      });
+      return { exitCode: 0, stdout: render({
+        ...review,
+        model_latency_ms: null,
+        model_latency_source: null
+      }, parsed.json), stderr: warnings.length ? `${warnings.join("\n")}
+` : "" };
+    };
     foregroundDataCall = async (config2, parsed, dependencies, warnings) => {
       const session = await agentSession(config2, dependencies, warnings);
       const state = await withAgentCommandLock(
@@ -46671,6 +46798,7 @@ Waiting for approval...
         const observedAt = now().toISOString();
         state = {
           ...state,
+          review: null,
           pending_decision: {
             request: {
               assignment_id: state.assignment.assignment_id,
@@ -46678,6 +46806,11 @@ Waiting for approval...
               operation_id: agentOperationId("agent-decision-submit"),
               candidate: input.candidate,
               observed_at: observedAt,
+              ...state.review ? { timing: {
+                analysis_started_at: state.review.analysis_started_at,
+                model_latency_ms: null,
+                model_latency_source: null
+              } } : {},
               provenance: input.provenance
             },
             first_transmit_at: null,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@vtxmacro/cli",
-  "version": "2026.9.66",
+  "version": "2026.9.68",
   "description": "VTX Macro CLI, MCP server, and durable external inference host.",
   "type": "module",
   "license": "UNLICENSED",