@vtxmacro/cli 2026.8.26 → 2026.8.28

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: "2026.8.26",
+      package_version: "2026.8.28",
       codex_package_name: "@openai/codex",
       codex_version: "0.147.0",
       platforms: {
@@ -19101,12 +19101,13 @@ async function logoutCodexSubscription(options) {
     await session.close();
   }
 }
-var CODEX_INFERENCE_PERMISSION_PROFILE, CODEX_ACCOUNT_PLAN_TYPES, 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, classifyCodexTurnFailure, scrubbedCodexEnvironment, killWindowsProcessTree, appServerArgs, GUARDIAN_SCRIPT, WINDOWS_RECEIPT_REPLACE_ERROR_CODES, replaceCodexGuardianReceiptFile, writeCodexGuardianSpawnIntent, parseGuardianReceipt, readCodexGuardianReceipt, waitForCodexGuardianState, defaultSpawn, CodexAppServerSession;
+var CODEX_INFERENCE_PERMISSION_PROFILE, CODEX_REASONING_SUMMARY_MAX_UTF8_BYTES, CODEX_ACCOUNT_PLAN_TYPES, 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, classifyCodexTurnFailure, scrubbedCodexEnvironment, killWindowsProcessTree, appServerArgs, GUARDIAN_SCRIPT, WINDOWS_RECEIPT_REPLACE_ERROR_CODES, replaceCodexGuardianReceiptFile, writeCodexGuardianSpawnIntent, parseGuardianReceipt, readCodexGuardianReceipt, waitForCodexGuardianState, defaultSpawn, CodexAppServerSession;
 var init_codex_app_server = __esm({
   "lib/inference-host/codex-app-server.ts"() {
     "use strict";
     init_config();
     CODEX_INFERENCE_PERMISSION_PROFILE = "vtx_inference_readonly";
+    CODEX_REASONING_SUMMARY_MAX_UTF8_BYTES = 32768;
     CODEX_ACCOUNT_PLAN_TYPES = [
       "free",
       "go",
@@ -20258,6 +20259,9 @@ child.once('close', async () => {
           developerInstructions: options.systemPrompt,
           config: {
             model_reasoning_effort: options.requestedReasoningEffort,
+            model_reasoning_summary: "detailed",
+            show_raw_agent_reasoning: false,
+            hide_agent_reasoning: false,
             web_search: "disabled",
             features: {
               apps: false,
@@ -20393,7 +20397,7 @@ child.once('close', async () => {
           if (method === "thread/settings/updated") {
             const settings = objectOrNull(params.threadSettings);
             const permission = objectOrNull(settings?.activePermissionProfile);
-            if (!settings || settings.model !== request.requestedModel || settings.modelProvider !== "openai" || settings.effort !== request.requestedReasoningEffort || String(settings.cwd || "") !== request.workspacePath || settings.approvalPolicy !== "never" || permission?.id !== CODEX_INFERENCE_PERMISSION_PROFILE || permission.extends !== null) {
+            if (!settings || settings.model !== request.requestedModel || settings.modelProvider !== "openai" || settings.effort !== request.requestedReasoningEffort || settings.summary !== "detailed" || String(settings.cwd || "") !== request.workspacePath || settings.approvalPolicy !== "never" || permission?.id !== CODEX_INFERENCE_PERMISSION_PROFILE || permission.extends !== null) {
               terminalReject(new CodexAppServerError({
                 message: "Codex effective turn settings changed after dispatch.",
                 category: "adapter",
@@ -20484,6 +20488,7 @@ child.once('close', async () => {
               model: request.requestedModel,
               serviceTier: "default",
               effort: request.requestedReasoningEffort,
+              summary: "detailed",
               outputSchema: request.outputSchema
             }, {
               timeoutMs: remaining(),
@@ -20672,6 +20677,12 @@ child.once('close', async () => {
           const fallbackMessages = items.filter((item) => item?.type === "agentMessage" && item.phase == null);
           const answerItem = finalMessages.at(-1) ?? fallbackMessages.at(-1);
           const text = typeof answerItem?.text === "string" ? answerItem.text : "";
+          const reasoningSummaryParts = items.flatMap((item) => {
+            if (item?.type !== "reasoning" || !Array.isArray(item.summary)) return [];
+            return item.summary.filter((part) => typeof part === "string" && part.trim().length > 0);
+          });
+          const reasoningSummaryCandidate = reasoningSummaryParts.length > 0 ? reasoningSummaryParts.join("\n\n") : null;
+          const reasoningSummary = reasoningSummaryCandidate !== null && Buffer.byteLength(reasoningSummaryCandidate, "utf8") <= CODEX_REASONING_SUMMARY_MAX_UTF8_BYTES ? reasoningSummaryCandidate : null;
           const usage = threadUsage ?? rawResponseUsage;
           const responseIdentity = adapterResponseId ?? turnId;
           if (!text || !usage || !responseIdentity) {
@@ -20685,6 +20696,7 @@ child.once('close', async () => {
           }
           return {
             text,
+            reasoningSummary,
             requestedModel: request.requestedModel,
             effectiveModel: request.thread.effectiveModel,
             requestedReasoningEffort: request.requestedReasoningEffort,
@@ -20991,6 +21003,7 @@ var init_codex_adapter = __esm({
     });
     turnResultReceiptSchema = external_exports.strictObject({
       text: external_exports.string().min(1).max(MAX_PROMPT_BYTES2),
+      reasoningSummary: external_exports.string().min(1).max(CODEX_REASONING_SUMMARY_MAX_UTF8_BYTES).nullable().optional().transform((value) => value ?? null),
       requestedModel: external_exports.string().regex(CODEX_MODEL_NAME_PATTERN),
       effectiveModel: external_exports.string().regex(CODEX_MODEL_NAME_PATTERN),
       requestedReasoningEffort: external_exports.string().regex(CODEX_REASONING_EFFORT_PATTERN),
@@ -21002,7 +21015,7 @@ var init_codex_adapter = __esm({
       timeToFirstTokenMs: external_exports.number().int().nonnegative().max(Number.MAX_SAFE_INTEGER).nullable(),
       terminalStatus: external_exports.literal("completed")
     }).superRefine((value, context) => {
-      if (value.usage.cachedInputTokens > value.usage.inputTokens || value.usage.reasoningOutputTokens > value.usage.outputTokens || value.usage.totalTokens !== value.usage.inputTokens + value.usage.outputTokens || value.effectiveModel !== value.requestedModel || value.effectiveReasoningEffort !== value.requestedReasoningEffort) {
+      if (value.reasoningSummary != null && utf8Bytes(value.reasoningSummary) > CODEX_REASONING_SUMMARY_MAX_UTF8_BYTES || value.usage.cachedInputTokens > value.usage.inputTokens || value.usage.reasoningOutputTokens > value.usage.outputTokens || value.usage.totalTokens !== value.usage.inputTokens + value.usage.outputTokens || value.effectiveModel !== value.requestedModel || value.effectiveReasoningEffort !== value.requestedReasoningEffort) {
         context.addIssue({ code: "custom", message: "Invalid Codex recovery usage or identity." });
       }
     });
@@ -28711,14 +28724,24 @@ var init_runner = __esm({
       return models;
     };
     FileInferenceHostRuntimeReceiptStore = class {
-      constructor(path) {
+      constructor(path, readPrivateFile = readInferencePrivateFile) {
         this.path = path;
+        this.readPrivateFile = readPrivateFile;
       }
       async read() {
-        const raw = await readInferencePrivateFile(
-          this.path,
-          "Inference host runtime receipt file"
-        );
+        let raw = null;
+        for (let attempt = 0; attempt < 3; attempt += 1) {
+          try {
+            raw = await this.readPrivateFile(
+              this.path,
+              "Inference host runtime receipt file"
+            );
+            break;
+          } catch (error48) {
+            const changedDuringRead = error48 instanceof Error && error48.message === "Inference host runtime receipt file changed while it was opened.";
+            if (!changedDuringRead || attempt === 2) throw error48;
+          }
+        }
         if (raw === null) return null;
         try {
           return validateRuntimeReceipt(JSON.parse(raw));
@@ -29073,7 +29096,9 @@ var init_runner = __esm({
           providerResponseSchema: null,
           providerResponseSchemaVersion: null,
           providerResponseMaxUtf8Bytes: null,
-          providerResponseTextMaxUtf8Bytes: null
+          providerResponseTextMaxUtf8Bytes: null,
+          providerReasoningSummaryMaxUtf8Bytes: null,
+          providerReasoningSummarySupported: false
         };
       }
       const definitions = objectRecord(outputSchema.$defs);
@@ -29081,21 +29106,27 @@ var init_runner = __esm({
       const properties = objectRecord(outputSchema.properties);
       const schemaVersionProperty = objectRecord(properties?.schema_version);
       const responseTextProperty = objectRecord(properties?.response_text);
+      const reasoningSummaryProperty = objectRecord(properties?.reasoning_summary);
       const schemaVersion = schemaVersionProperty?.const;
       const requiredFields = outputSchema.required;
-      const expectedFields = [
+      const baseExpectedFields = [
         "schema_version",
         "response_text",
         "finish_reason",
         "refusal_status"
       ];
+      const providerReasoningSummarySupported = reasoningSummaryProperty !== null;
+      const expectedFields = providerReasoningSummarySupported ? [...baseExpectedFields, "reasoning_summary"] : baseExpectedFields;
       const responseMaxUtf8Bytes = positiveUtf8ByteLimit(
         outputSchema["x-vtx-max-utf8-bytes"]
       );
       const responseTextMaxUtf8Bytes = positiveUtf8ByteLimit(
         responseTextProperty?.["x-vtx-max-utf8-bytes"]
       );
-      if (outputSchema.type !== "object" || outputSchema.additionalProperties !== false || !Array.isArray(requiredFields) || !requiredFields.every((field) => typeof field === "string") || [...requiredFields].sort().join("\0") !== expectedFields.sort().join("\0") || properties === null || !exactObjectKeys(properties, expectedFields) || adapterSchema === null || typeof schemaVersion !== "string" || schemaVersion !== jobInput.output_schema_version || responseTextMaxUtf8Bytes === null || responseMaxUtf8Bytes === null) {
+      const reasoningSummaryMaxUtf8Bytes = positiveUtf8ByteLimit(
+        reasoningSummaryProperty?.["x-vtx-max-utf8-bytes"]
+      );
+      if (outputSchema.type !== "object" || outputSchema.additionalProperties !== false || !Array.isArray(requiredFields) || !requiredFields.every((field) => typeof field === "string") || [...requiredFields].sort().join("\0") !== expectedFields.sort().join("\0") || properties === null || !exactObjectKeys(properties, expectedFields) || adapterSchema === null || typeof schemaVersion !== "string" || schemaVersion !== jobInput.output_schema_version || responseTextMaxUtf8Bytes === null || providerReasoningSummarySupported && reasoningSummaryMaxUtf8Bytes === null || responseMaxUtf8Bytes === null) {
         throw new InferenceHostRunnerError(
           "invalid_output_schema",
           "The provider-response output schema does not match the canonical envelope structure."
@@ -29109,7 +29140,9 @@ var init_runner = __esm({
         providerResponseSchema: outputSchema,
         providerResponseSchemaVersion: schemaVersion,
         providerResponseMaxUtf8Bytes: responseMaxUtf8Bytes,
-        providerResponseTextMaxUtf8Bytes: responseTextMaxUtf8Bytes
+        providerResponseTextMaxUtf8Bytes: responseTextMaxUtf8Bytes,
+        providerReasoningSummaryMaxUtf8Bytes: reasoningSummaryMaxUtf8Bytes,
+        providerReasoningSummarySupported
       };
     };
     safeFailureCode = (value, fallback) => {
@@ -30042,11 +30075,18 @@ var init_runner = __esm({
                 "Codex response text exceeds the immutable provider-response UTF-8 byte limit."
               );
             }
+            if (outputContract.providerReasoningSummarySupported && outputContract.providerReasoningSummaryMaxUtf8Bytes !== null && adapterResult.reasoningSummary !== null && Buffer.byteLength(adapterResult.reasoningSummary, "utf8") > outputContract.providerReasoningSummaryMaxUtf8Bytes) {
+              throw new InferenceHostRunnerError(
+                "output_schema_reasoning_summary_too_large",
+                "Codex reasoning summary exceeds the immutable provider-response UTF-8 byte limit."
+              );
+            }
             const providerResponse = {
               schema_version: outputContract.providerResponseSchemaVersion,
               response_text: adapterResult.text,
               finish_reason: finishReason,
-              refusal_status: refusalStatus
+              refusal_status: refusalStatus,
+              ...outputContract.providerReasoningSummarySupported ? { reasoning_summary: adapterResult.reasoningSummary } : {}
             };
             validateOutput(outputContract.providerResponseSchema, providerResponse);
             sealedPlaintext = JSON.stringify(providerResponse);
@@ -30949,7 +30989,7 @@ async function runInferenceHostCli(argv2, env = process.env, dependencies = {})
       return { exitCode: 0, stdout: INFERENCE_HOST_HELP, stderr: "" };
     }
     const parsed = parseInferenceHostArgs(argv2, env);
-    const config2 = resolveInferenceHostConfig(env);
+    const config2 = await resolveInferenceHostCommandConfig(parsed, env);
     if (parsed.command === "login") {
       return await login(config2, parsed, env, dependencies, warnings);
     }
@@ -31016,7 +31056,7 @@ async function runInferenceHostCli(argv2, env = process.env, dependencies = {})
     };
   }
 }
-var INFERENCE_HOST_CLI_VERSION, runtimeReceiptPath, codexRecoveryPath, codexGuardianReceiptRoot, revocationCheckpointPath, foregroundHostLockPath, AGENT_HEARTBEAT_INTERVAL_MS, AGENT_HEARTBEAT_MAX_RETRY_DELAY_MS, retryableAgentHeartbeatError, assertRevocationCheckpoint, readRevocationCheckpoint, writeRevocationCheckpoint, clearRevocationCheckpoint, render, INFERENCE_HOST_HELP, parsePositiveInteger, parseInferenceHostArgs, defaultOpenBrowser, defaultRegisterLifecycleSignalHandlers, lifecycleCancellation, configuredCredentialStore, accountKeyForState, assertCredentialMatchesState2, revocationCheckpointForCredential, assertRevocationCheckpointMatchesLocalIdentity, fileExistsPrivately, inspectCodexAuthentication, clearLocalRuntimeArtifacts, assertDurableServiceUninstalled, assertLocalRuntimeArtifactsMayBeDiscarded, defaultRunForeground, login, codexLogin, codexLogout, localStatus, doctor, cleanupLogin, runHost, defaultReadStdin, parseAgentStdin, agentOperationId, agentSession, agentConnect, agentRun, agentNext, agentComplete, agentFail, withAgentCommandLock, serviceCommand;
+var INFERENCE_HOST_CLI_VERSION, runtimeReceiptPath, codexRecoveryPath, codexGuardianReceiptRoot, revocationCheckpointPath, foregroundHostLockPath, AGENT_HEARTBEAT_INTERVAL_MS, AGENT_HEARTBEAT_MAX_RETRY_DELAY_MS, retryableAgentHeartbeatError, assertRevocationCheckpoint, readRevocationCheckpoint, writeRevocationCheckpoint, clearRevocationCheckpoint, render, INFERENCE_HOST_HELP, parsePositiveInteger, parseInferenceHostArgs, defaultOpenBrowser, defaultRegisterLifecycleSignalHandlers, lifecycleCancellation, configuredCredentialStore, accountKeyForState, assertCredentialMatchesState2, revocationCheckpointForCredential, assertRevocationCheckpointMatchesLocalIdentity, fileExistsPrivately, inspectCodexAuthentication, clearLocalRuntimeArtifacts, assertDurableServiceUninstalled, assertLocalRuntimeArtifactsMayBeDiscarded, defaultRunForeground, login, codexLogin, codexLogout, localStatus, doctor, cleanupLogin, runHost, defaultReadStdin, parseAgentStdin, agentOperationId, agentSession, agentConnect, agentRun, agentNext, agentComplete, agentFail, withAgentCommandLock, serviceCommand, hasExplicitCredentialStoreConfiguration, commandUsesInstalledServiceCredentials, resolveInferenceHostCommandConfig;
 var init_cli = __esm({
   "lib/inference-host/cli.ts"() {
     "use strict";
@@ -32484,6 +32524,29 @@ Waiting for approval...
       }
       throw new Error("Usage: vtx inference-host service <install|start|stop|status|logs|uninstall>.");
     };
+    hasExplicitCredentialStoreConfiguration = (env) => Boolean(
+      String(env.VTX_INFERENCE_HOST_CREDENTIAL_STORE || "").trim() || String(env.VTX_INFERENCE_HOST_CREDENTIAL_FILE || "").trim()
+    );
+    commandUsesInstalledServiceCredentials = (parsed) => parsed.command === "status" || parsed.command === "doctor" || parsed.command === "service" && parsed.serviceAction === "install";
+    resolveInferenceHostCommandConfig = async (parsed, env) => {
+      const baseConfig = resolveInferenceHostConfig(env);
+      if (!commandUsesInstalledServiceCredentials(parsed) || hasExplicitCredentialStoreConfiguration(env)) {
+        return baseConfig;
+      }
+      const manifestPath = `${baseConfig.statePath}.service.json`;
+      const manifest = await readInferenceHostServiceManifest(manifestPath);
+      if (!manifest) return baseConfig;
+      const installedConfig = resolveInferenceHostConfig({
+        ...env,
+        ...manifest.runtime_environment
+      });
+      if (resolve5(installedConfig.statePath) !== resolve5(baseConfig.statePath)) {
+        throw new Error(
+          "Installed inference-host service manifest does not match the requested local state path."
+        );
+      }
+      return installedConfig;
+    };
   }
 });
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@vtxmacro/cli",
-  "version": "2026.8.26",
+  "version": "2026.8.28",
   "description": "VTX Macro CLI, MCP server, and durable subscription inference host.",
   "type": "module",
   "license": "UNLICENSED",