@vtxmacro/cli 2026.9.37 → 2026.9.38

package/bin/exo-harness.ts

@@ -36,6 +36,8 @@ type ExoModelModule = {
 };
 
 const record = (value: unknown): Json => value !== null && typeof value === 'object' && !Array.isArray(value) ? value as Json : {};
+const validModel = (value: unknown): value is string => typeof value === 'string' && Boolean(value.trim())
+  && value.length <= 256 && !/[\u0000-\u001f\u007f]/u.test(value);
 const count = (value: unknown): number => {
   if (!Number.isSafeInteger(value) || Number(value) < 0) throw new Error('exo_usage_invalid');
   return Number(value);
@@ -286,7 +288,7 @@ export default {
             const choices = Array.isArray(raw.choices) ? raw.choices : [];
             const choice = record(choices[0]);
             terminalKnown = choices.length === 1 && ['stop', 'tool_calls', 'length', 'content_filter'].includes(String(choice.finish_reason));
-            if (raw.model !== input.providerModel || typeof raw.id !== 'string' || !raw.id) throw new Error('exo_response_identity_mismatch');
+            if (!validModel(raw.model) || typeof raw.id !== 'string' || !raw.id) throw new Error('exo_response_identity_mismatch');
             const controls = record(raw.venice_parameters);
             if (controls.disable_thinking !== true || controls.strip_thinking_response !== false
               || controls.include_venice_system_prompt !== false || controls.enable_web_search !== 'off'
@@ -298,7 +300,7 @@ export default {
             addUsage({ input_tokens: nativeUsage.prompt_tokens, output_tokens: nativeUsage.completion_tokens, total_tokens: nativeUsage.total_tokens,
               input_tokens_details: nativeUsage.prompt_tokens_details, output_tokens_details: nativeUsage.completion_tokens_details });
             lastCompletedResponseId = raw.id;
-            currentReceipt = { responseId: raw.id, providerModel: input.providerModel, reasoningEffort: input.reasoningEffort };
+            currentReceipt = { responseId: raw.id, providerModel: raw.model, reasoningEffort: input.reasoningEffort };
             // Exo synthesizes completed status for every Chat response. Inspect
             // the real terminal reason before that normalization can hide loss.
             if (choices.length !== 1 || !['stop', 'tool_calls'].includes(String(choice.finish_reason))) throw new Error('exo_response_incomplete');
@@ -333,14 +335,15 @@ export default {
         // model tool call to the broker, especially decision submission.
         if (response.status !== 'completed') throw new Error('exo_response_incomplete');
         terminalKnown = true;
-        if (response.model !== input.providerModel || typeof response.id !== 'string' || !response.id) throw new Error('exo_response_identity_mismatch');
+        if (!validModel(response.model) || typeof response.id !== 'string' || !response.id) throw new Error('exo_response_identity_mismatch');
         if (input.provider === 'openai') {
           const effectiveEffort = record(response.reasoning).effort ?? null;
           if (effectiveEffort !== input.nativeReasoningEffort) throw new Error('exo_response_effort_mismatch');
           addUsage(record(response.usage));
           lastCompletedResponseId = response.id;
-          currentReceipt = { responseId: response.id, providerModel: input.providerModel, reasoningEffort: input.reasoningEffort };
+          currentReceipt = { responseId: response.id, providerModel: response.model, reasoningEffort: input.reasoningEffort };
         }
+        if (!currentReceipt || currentReceipt.responseId !== response.id || currentReceipt.providerModel !== response.model) throw new Error('exo_response_identity_mismatch');
         const events = model.responseToLinguaEvents(response);
         await context.exoharness.current.turn.addEvents(input.provider === 'venice' ? chatResponseEvents(events) : events);
         const calls = model.responseToolCalls(response);
@@ -354,7 +357,7 @@ export default {
         if (input.mode !== 'agent' || calls.some((call) => !allowed.has(call.request.functionName))) throw new Error('exo_tool_not_allowed');
         for (const call of calls) {
           const result = await broker('/tool', { callId: call.toolCallId, tool: call.request.functionName, arguments: call.request.arguments,
-            effectiveModel: input.requestedModel, effectiveReasoningEffort: input.reasoningEffort }, input.deadlineAtMs);
+            effectiveModel: `${input.provider}/${currentReceipt!.providerModel}`, effectiveReasoningEffort: input.reasoningEffort }, input.deadlineAtMs);
           // Append the entire result directly. The default registry replaces
           // results over 8k with previews, which loses assignment data.
           await context.exoharness.current.turn.addEvents([harness.toolResultEvent(call.toolCallId,

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: "2026.9.37",
+  package_version: "2026.9.38",
   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: "2026.9.37",
+      package_version: "2026.9.38",
       codex_package_name: "@openai/codex",
       codex_version: "0.153.3",
       copilot_sdk_package_name: "@github/copilot-sdk",
@@ -24222,7 +24222,7 @@ var init_codex_adapter = __esm({
       reasoningContent: external_exports.string().min(1).max(CODEX_REASONING_CONTENT_MAX_UTF8_BYTES).nullable().optional().transform((value) => value ?? null),
       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),
+      effectiveModel: external_exports.string().min(1).max(256).refine((value) => value.trim().length > 0 && !/[\u0000-\u001f\u007f]/u.test(value)),
       requestedReasoningEffort: external_exports.string().regex(CODEX_REASONING_EFFORT_PATTERN),
       effectiveReasoningEffort: external_exports.string().regex(CODEX_REASONING_EFFORT_PATTERN),
       adapterRequestId: external_exports.string().min(1).max(512),
@@ -24232,7 +24232,7 @@ 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.reasoningContent != null && utf8Bytes(value.reasoningContent) > CODEX_REASONING_CONTENT_MAX_UTF8_BYTES || 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) {
+      if (value.reasoningContent != null && utf8Bytes(value.reasoningContent) > CODEX_REASONING_CONTENT_MAX_UTF8_BYTES || 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.effectiveReasoningEffort !== value.requestedReasoningEffort) {
         context.addIssue({ code: "custom", message: "Invalid Codex recovery usage or identity." });
       }
     });
@@ -25232,7 +25232,7 @@ var init_codex_adapter = __esm({
           }
           if (prior.terminal?.kind === "completed") {
             const recovered = prior.terminal.result;
-            if (recovered.requestedModel !== input.requestedModel || recovered.effectiveModel !== input.requestedModel || recovered.requestedReasoningEffort !== input.requestedReasoningEffort || recovered.effectiveReasoningEffort !== input.requestedReasoningEffort || recovered.terminalStatus !== "completed") {
+            if (recovered.requestedModel !== input.requestedModel || typeof recovered.effectiveModel !== "string" || recovered.effectiveModel.trim().length === 0 || recovered.effectiveModel.length > 256 || /[\u0000-\u001f\u007f]/u.test(recovered.effectiveModel) || recovered.requestedReasoningEffort !== input.requestedReasoningEffort || recovered.effectiveReasoningEffort !== input.requestedReasoningEffort || recovered.terminalStatus !== "completed") {
               throw new CodexAppServerError({
                 message: "Codex completed recovery receipt changed immutable model identity.",
                 category: "adapter",
@@ -25849,7 +25849,7 @@ import {
   RuntimeConnection,
   defineTool
 } from "@github/copilot-sdk";
-var COPILOT_SDK_VERSION, MAX_RESULT_BYTES, resolveRuntimePackage, copilotRuntimePackageCandidates, resolvePinnedCopilotCliPath, createPrivateWorkspace, cleanCopilotEnvironment, defaultClient, modelCapabilities, requiredUsageInteger, optionalUsageInteger, copilotAgentToolDefinitions, CopilotSubscriptionAdapter;
+var COPILOT_SDK_VERSION, MAX_RESULT_BYTES, resolveRuntimePackage, copilotRuntimePackageCandidates, resolvePinnedCopilotCliPath, createPrivateWorkspace, cleanCopilotEnvironment, defaultClient, modelCapabilities, requiredUsageInteger, optionalUsageInteger, responseModelLabel, coherentResponseModel, copilotAgentToolDefinitions, CopilotSubscriptionAdapter;
 var init_copilot_adapter = __esm({
   "lib/inference-host/copilot-adapter.ts"() {
     "use strict";
@@ -25948,9 +25948,20 @@ var init_copilot_adapter = __esm({
       }
       return Number(value);
     };
-    copilotAgentToolDefinitions = (input, evidence) => {
+    responseModelLabel = (value) => {
+      if (typeof value !== "string" || value.trim().length === 0 || value.length > 256 || /[\u0000-\u001f\u007f]/u.test(value)) {
+        throw new Error("copilot_effective_model_invalid");
+      }
+      return value;
+    };
+    coherentResponseModel = (models) => {
+      if (models.size !== 1) throw new Error("copilot_effective_model_mismatch");
+      return [...models][0];
+    };
+    copilotAgentToolDefinitions = (input, evidence, beforeExecute) => {
       const execute = async (tool, argumentsValue) => {
         const argumentsRecord = argumentsValue && typeof argumentsValue === "object" && !Array.isArray(argumentsValue) ? argumentsValue : {};
+        await beforeExecute();
         const callId = randomUUID();
         const result2 = await input.executeTool({
           callId,
@@ -26149,7 +26160,7 @@ ${input.outputSchemaJson}`
           let providerCallId = null;
           for (const event of usageEvents) {
             if (event.type !== "assistant.usage") continue;
-            observedModels.add(event.data.model);
+            observedModels.add(responseModelLabel(event.data.model));
             if (event.data.reasoningEffort) observedEfforts.add(event.data.reasoningEffort);
             inputTokens += requiredUsageInteger(event.data.inputTokens, "input_tokens");
             cachedInputTokens += optionalUsageInteger(event.data.cacheReadTokens, "cache_read_tokens");
@@ -26170,10 +26181,8 @@ ${input.outputSchemaJson}`
           if (usageEvents.length === 0) {
             throw new Error("copilot_usage_receipt_missing");
           }
-          if (response.data.model) observedModels.add(response.data.model);
-          if (observedModels.size !== 1 || !observedModels.has(input.requestedModel)) {
-            throw new Error("copilot_effective_model_mismatch");
-          }
+          if (response.data.model !== void 0) observedModels.add(responseModelLabel(response.data.model));
+          const effectiveModel = coherentResponseModel(observedModels);
           const effectiveEffort = input.requestedReasoningEffort === "none" ? observedEfforts.size === 0 ? "none" : observedEfforts.size === 1 ? [...observedEfforts][0] : null : observedEfforts.size === 1 ? [...observedEfforts][0] : null;
           if (effectiveEffort !== input.requestedReasoningEffort) {
             throw new Error("copilot_effective_effort_mismatch");
@@ -26192,7 +26201,7 @@ ${input.outputSchemaJson}`
             reasoningContent: response.data.reasoningText ?? null,
             reasoningSummary: null,
             requestedModel: input.requestedModel,
-            effectiveModel: input.requestedModel,
+            effectiveModel,
             requestedReasoningEffort: input.requestedReasoningEffort,
             effectiveReasoningEffort: effectiveEffort,
             adapterRequestId: providerCallId ?? response.data.requestId ?? response.data.messageId,
@@ -26261,9 +26270,22 @@ ${input.outputSchemaJson}`
         let durableCheckpointed = input.durableThread !== null;
         let dispatchEntered = false;
         const toolEvidence = [];
+        const observedModels = /* @__PURE__ */ new Set();
+        let modelReceiptError = null;
+        let thread = null;
+        const checkpointObservedModel = async () => {
+          if (modelReceiptError) throw modelReceiptError;
+          if (observedModels.size === 0 || !thread) return;
+          const effectiveModel = coherentResponseModel(observedModels);
+          if (thread.effectiveModel !== effectiveModel) {
+            const updatedThread = { ...thread, effectiveModel };
+            await input.onThreadReady?.(updatedThread);
+            thread = updatedThread;
+          }
+        };
         try {
           await client.start();
-          const tools = copilotAgentToolDefinitions(input, toolEvidence);
+          const tools = copilotAgentToolDefinitions(input, toolEvidence, checkpointObservedModel);
           const sessionConfig = {
             clientName: "@vtxmacro/cli durable Copilot Agent host",
             model: input.requestedModel,
@@ -26289,7 +26311,7 @@ ${JSON.stringify(input.outputSchema)}`
             workspacePath,
             cleanup: ownedWorkspace?.cleanup ?? null
           };
-          const thread = {
+          thread = {
             threadId: session.sessionId,
             threadPath: workspacePath,
             effectiveModel: input.requestedModel,
@@ -26301,7 +26323,15 @@ ${JSON.stringify(input.outputSchema)}`
           }
           const usageEvents = [];
           session.on((event) => {
-            if (event.type === "assistant.usage" && !event.agentId) usageEvents.push(event);
+            if (event.type === "assistant.usage" && !event.agentId) {
+              usageEvents.push(event);
+              try {
+                observedModels.add(responseModelLabel(event.data.model));
+                coherentResponseModel(observedModels);
+              } catch (error48) {
+                modelReceiptError = error48;
+              }
+            }
           });
           const onAbort = () => {
             void session?.abort().catch(() => void 0);
@@ -26320,7 +26350,6 @@ ${JSON.stringify(input.outputSchema)}`
           if (!response?.data.content || Buffer.byteLength(response.data.content, "utf8") > MAX_RESULT_BYTES) {
             throw new Error("copilot_agent_invalid_result");
           }
-          const observedModels = /* @__PURE__ */ new Set();
           const observedEfforts = /* @__PURE__ */ new Set();
           let inputTokens = 0;
           let cachedInputTokens = 0;
@@ -26332,7 +26361,7 @@ ${JSON.stringify(input.outputSchema)}`
           let providerCallId = null;
           for (const event of usageEvents) {
             if (event.type !== "assistant.usage") continue;
-            observedModels.add(event.data.model);
+            observedModels.add(responseModelLabel(event.data.model));
             if (event.data.reasoningEffort) observedEfforts.add(event.data.reasoningEffort);
             inputTokens += requiredUsageInteger(event.data.inputTokens, "input_tokens");
             cachedInputTokens += optionalUsageInteger(event.data.cacheReadTokens, "cache_read_tokens");
@@ -26348,10 +26377,9 @@ ${JSON.stringify(input.outputSchema)}`
             providerCallId = event.data.providerCallId ?? event.data.apiCallId ?? providerCallId;
           }
           if (usageEvents.length === 0) throw new Error("copilot_agent_usage_receipt_missing");
-          if (response.data.model) observedModels.add(response.data.model);
-          if (observedModels.size !== 1 || !observedModels.has(input.requestedModel)) {
-            throw new Error("copilot_agent_effective_model_mismatch");
-          }
+          if (response.data.model !== void 0) observedModels.add(responseModelLabel(response.data.model));
+          const effectiveModel = coherentResponseModel(observedModels);
+          await checkpointObservedModel();
           const effectiveEffort = input.requestedReasoningEffort === "none" ? observedEfforts.size === 0 ? "none" : observedEfforts.size === 1 ? [...observedEfforts][0] : null : observedEfforts.size === 1 ? [...observedEfforts][0] : null;
           if (effectiveEffort !== input.requestedReasoningEffort) {
             throw new Error("copilot_agent_effective_effort_mismatch");
@@ -26372,7 +26400,7 @@ ${JSON.stringify(input.outputSchema)}`
               reasoningContent: response.data.reasoningText ?? null,
               reasoningSummary: null,
               requestedModel: input.requestedModel,
-              effectiveModel: input.requestedModel,
+              effectiveModel,
               requestedReasoningEffort: input.requestedReasoningEffort,
               effectiveReasoningEffort: effectiveEffort,
               adapterRequestId: providerCallId ?? response.data.requestId ?? response.data.messageId,
@@ -26482,6 +26510,7 @@ var init_deepseek_transport_proxy = __esm({
         this.upstreamRequestStartedValue = false;
         this.providerResponseObservedValue = false;
         this.unansweredUpstreamRequestsValue = 0;
+        this.latestReceiptValue = null;
       }
       get baseUrl() {
         if (!this.urlValue) throw new Error("deepseek_transport_proxy_not_started");
@@ -26490,6 +26519,9 @@ var init_deepseek_transport_proxy = __esm({
       get receipts() {
         return [...this.receiptsValue];
       }
+      get lastReceipt() {
+        return this.latestReceiptValue;
+      }
       get providerResponseObserved() {
         return this.providerResponseObservedValue;
       }
@@ -26528,6 +26560,7 @@ var init_deepseek_transport_proxy = __esm({
             const controller = new AbortController();
             this.controllers.add(controller);
             try {
+              this.latestReceiptValue = null;
               this.upstreamRequestStartedValue = true;
               this.unansweredUpstreamRequestsValue += 1;
               const upstream = await this.fetchImpl(
@@ -26571,6 +26604,7 @@ var init_deepseek_transport_proxy = __esm({
               const decoder2 = new TextDecoder();
               let pending2 = "";
               let observedId = "";
+              let observedModel = "";
               let terminalObserved = false;
               const forwardLine = (line) => {
                 if (!line.startsWith("data:")) {
@@ -26585,25 +26619,18 @@ var init_deepseek_transport_proxy = __esm({
                   return;
                 }
                 if (value === "[DONE]") {
-                  if (terminalObserved || !observedId) {
+                  if (terminalObserved || !observedId || !observedModel) {
                     throw new Error("deepseek_provider_receipt_missing_or_mismatched");
                   }
                   terminalObserved = true;
-                  this.receiptsValue.push({
-                    id: observedId,
-                    model: this.expectedModel,
-                    requestedModel,
-                    requestedReasoningEffort: requestedEffort
-                  });
-                  response.write(`${line}
-`);
                   return;
                 }
                 const event = JSON.parse(value);
                 const eventId = typeof event.id === "string" ? event.id.trim() : "";
-                const eventModel = typeof event.model === "string" ? event.model.trim() : "";
-                if (!eventId || eventModel !== this.expectedModel || observedId && eventId !== observedId) throw new Error("deepseek_provider_receipt_missing_or_mismatched");
+                const eventModel = typeof event.model === "string" ? event.model : "";
+                if (terminalObserved || !eventId || !eventModel.trim() || eventModel.length > 256 || /[\u0000-\u001f\u007f]/u.test(eventModel) || observedModel && eventModel !== observedModel || observedId && eventId !== observedId) throw new Error("deepseek_provider_receipt_missing_or_mismatched");
                 observedId = eventId;
+                observedModel = eventModel;
                 response.write(`${line}
 `);
               };
@@ -26620,6 +26647,10 @@ var init_deepseek_transport_proxy = __esm({
               if (!terminalObserved) {
                 throw new Error("deepseek_provider_receipt_missing_or_mismatched");
               }
+              const receipt = { id: observedId, model: observedModel, requestedModel, requestedReasoningEffort: requestedEffort };
+              this.receiptsValue.push(receipt);
+              this.latestReceiptValue = receipt;
+              response.write("data: [DONE]\n\n");
               response.end();
             } finally {
               this.controllers.delete(controller);
@@ -27145,6 +27176,12 @@ ${JSON.stringify(input.outputSchema)}`,
                 },
                 execute: async (argumentsValue) => {
                   const argumentsRecord = argumentsValue && typeof argumentsValue === "object" && !Array.isArray(argumentsValue) ? argumentsValue : {};
+                  const receipt = proxy.lastReceipt;
+                  if (!receipt) throw new Error("deepseek_harness_agent_receipt_missing");
+                  if (thread.effectiveModel !== receipt.model) {
+                    thread = { ...thread, effectiveModel: receipt.model };
+                    await input.onThreadReady?.(thread);
+                  }
                   const callId = randomUUID2();
                   const result2 = await input.executeTool({ callId, tool: definition.name, arguments: argumentsRecord });
                   evidence.push({ callId, tool: definition.name, arguments: argumentsRecord, success: result2.success });
@@ -27192,7 +27229,7 @@ ${JSON.stringify(input.outputSchema)}`,
               yield* next();
             })();
           });
-          const thread = {
+          let thread = {
             threadId: String(sessionId),
             threadPath: rootPath,
             effectiveModel: input.requestedModel,
@@ -27244,6 +27281,10 @@ ${JSON.stringify(input.outputSchema)}`,
           }
           const text = validateText(textFromBlocks(last.data.message.content, "text"), "deepseek_harness_agent_result_invalid");
           const receipts = proxy.receipts;
+          if (thread.effectiveModel !== receipts.at(-1).model) {
+            thread = { ...thread, effectiveModel: receipts.at(-1).model };
+            await input.onThreadReady?.(thread);
+          }
           await handle.dispose();
           return {
             thread,
@@ -33852,7 +33893,7 @@ var init_pi_transport = __esm({
         if (new Set(this.receipts.map((receipt) => receipt.id)).size !== this.receipts.length) throw new Error("pi_response_identity_reused");
         for (const receipt of this.receipts) {
           if (receipt.reasoningEffortChanged || (this.options.model.reasoning ? receipt.reasoningEffort !== this.options.expectedEffort.native : receipt.reasoningEffort !== null && receipt.reasoningEffort !== "none")) throw new Error("pi_native_effort_mismatch");
-          if (!receipt.id || receipt.model !== this.options.model.id || !receipt.terminal || receipt.input === null || receipt.output === null || receipt.cachedInput > receipt.input || receipt.input + receipt.output <= 0) {
+          if (!receipt.id || typeof receipt.model !== "string" || !receipt.model.trim() || receipt.model.length > 256 || /[\u0000-\u001f\u007f]/u.test(receipt.model) || !receipt.terminal || receipt.input === null || receipt.output === null || receipt.cachedInput > receipt.input || receipt.input + receipt.output <= 0) {
             throw new Error("pi_native_receipt_unqualified");
           }
         }
@@ -34080,7 +34121,7 @@ var init_pi_adapter = __esm({
             if (this.activeThreads.has(cwd)) throw fail("pi_thread_busy");
             this.activeThreads.add(cwd);
             ownedPath = cwd;
-            if (agent.durableThread && (agent.durableThread.effectiveModel !== input.requestedModel || agent.durableThread.effectiveReasoningEffort !== input.requestedReasoningEffort)) throw fail("pi_thread_binding_changed", "model");
+            if (agent.durableThread && agent.durableThread.effectiveReasoningEffort !== input.requestedReasoningEffort) throw fail("pi_thread_binding_changed", "model");
           }
           const settings = SettingsManager.inMemory({
             retry: { enabled: false, maxRetries: 0, provider: { maxRetries: 0 } },
@@ -34107,6 +34148,8 @@ var init_pi_adapter = __esm({
             if (!info.isFile() || info.isSymbolicLink()) throw fail("pi_session_not_owned");
             manager = SessionManager.open(sessionFile, cwd, cwd);
             if (manager.getSessionId() !== agent.durableThread.threadId || manager.getHeader()?.cwd !== cwd) throw fail("pi_session_binding_changed");
+            const selection = manager.buildSessionContext().model;
+            if (!selection || selection.provider !== model.provider || selection.modelId !== model.id) throw fail("pi_thread_binding_changed", "model");
           } else if (agent) {
             const initial = SessionManager.inMemory(cwd, { id: basename2(cwd).slice("agent-".length) });
             await writeAtomicInferencePrivateFile(sessionFile, `${JSON.stringify(initial.getHeader())}
@@ -34150,6 +34193,11 @@ var init_pi_adapter = __esm({
                 session?.agent.abort();
                 throw boundaryFailure;
               }
+              const effectiveModel2 = `${model.provider}/${transport.lastReceipt.model}`;
+              if (thread && thread.effectiveModel !== effectiveModel2) {
+                thread = { ...thread, effectiveModel: effectiveModel2 };
+                await agent.onThreadReady?.(thread);
+              }
               const tool = definition.name;
               const result2 = await agent.executeTool({ callId, tool, arguments: args });
               tools.push({ callId, tool, arguments: args, success: result2.success });
@@ -34248,7 +34296,7 @@ var init_pi_adapter = __esm({
           for (let i = 0; i < messages.length; i++) {
             const message = messages[i];
             const native = observer.receipts[i];
-            if (message.responseId !== native.id || message.provider !== model.provider || message.model !== model.id || message.usage.input + message.usage.cacheRead + message.usage.cacheWrite !== native.input || message.usage.output !== native.output || message.usage.cacheRead !== native.cachedInput) throw fail("pi_receipt_mismatch");
+            if (message.responseId !== native.id || message.provider !== model.provider || message.usage.input + message.usage.cacheRead + message.usage.cacheWrite !== native.input || message.usage.output !== native.output || message.usage.cacheRead !== native.cachedInput) throw fail("pi_receipt_mismatch");
           }
           if (!last) throw fail("pi_result_missing");
           if (last.stopReason !== "stop") throw fail("pi_terminal_incomplete");
@@ -34262,13 +34310,18 @@ var init_pi_adapter = __esm({
           }
           if (!validateOutput(output3)) throw fail("pi_output_schema_mismatch", "schema");
           const reasoning = last.content.flatMap((part) => part.type === "thinking" ? [part.thinking] : []).join("");
+          const effectiveModel = `${model.provider}/${observer.receipts.at(-1).model}`;
+          if (agent && thread && thread.effectiveModel !== effectiveModel) {
+            thread = { ...thread, effectiveModel };
+            await agent.onThreadReady?.(thread);
+          }
           if (!agent) input.onTerminalDispatchOutcome?.("confirmed_dispatched");
           return { thread, tools, turn: {
             text,
             reasoningContent: reasoning ? Buffer.from(reasoning).subarray(0, 32768).toString("utf8") : null,
             reasoningSummary: null,
             requestedModel: input.requestedModel,
-            effectiveModel: `${model.provider}/${observer.receipts.at(-1).model}`,
+            effectiveModel,
             requestedReasoningEffort: input.requestedReasoningEffort,
             effectiveReasoningEffort: observer.effectiveReasoningEffort(),
             adapterRequestId: observer.receipts[0].id,
@@ -34292,7 +34345,9 @@ var init_pi_adapter = __esm({
             };
             try {
               const effort2 = transport.effectiveReasoningEffort();
-              terminalEvidence = { ...terminalEvidence, effectiveModel: input.requestedModel, effectiveReasoningEffort: effort2 };
+              const provider = input.requestedModel.slice(0, input.requestedModel.indexOf("/"));
+              const actualModel = transport.lastReceipt?.model;
+              if (actualModel) terminalEvidence = { ...terminalEvidence, effectiveModel: `${provider}/${actualModel}`, effectiveReasoningEffort: effort2 };
             } catch {
             }
           }
@@ -34300,7 +34355,7 @@ var init_pi_adapter = __esm({
           try {
             const qualified = messages.filter((message, index) => {
               const receipt = transport?.receipts[index];
-              return receipt && receipt.terminal && receipt.id === message.responseId && `${message.provider}/${receipt.model}` === input.requestedModel && message.usage.input + message.usage.cacheRead + message.usage.cacheWrite === receipt.input && message.usage.output === receipt.output && message.usage.cacheRead === receipt.cachedInput && (receipt.input ?? 0) + (receipt.output ?? 0) > 0;
+              return receipt && receipt.terminal && receipt.id === message.responseId && message.provider === input.requestedModel.slice(0, input.requestedModel.indexOf("/")) && typeof receipt.model === "string" && Boolean(receipt.model.trim()) && receipt.model.length <= 256 && !/[\u0000-\u001f\u007f]/u.test(receipt.model) && message.usage.input + message.usage.cacheRead + message.usage.cacheWrite === receipt.input && message.usage.output === receipt.output && message.usage.cacheRead === receipt.cachedInput && (receipt.input ?? 0) + (receipt.output ?? 0) > 0;
             });
             if (qualified.length) usage = usageSum(qualified);
           } catch {
@@ -34920,17 +34975,17 @@ var init_grok_transport_proxy = __esm({
           const event = record3(JSON.parse(data4));
           const native = route === "/responses" ? record3(event.response) : event;
           if (native.id !== void 0 || native.model !== void 0) {
-            if (typeof native.id !== "string" || !native.id || !this.boundIdentity.allowedEffectiveModels.includes(native.model) || id2 && native.id !== id2) {
+            if (typeof native.id !== "string" || !native.id || typeof native.model !== "string" || !native.model.trim() || native.model.length > 256 || /[\u0000-\u001f\u007f]/u.test(native.model) || id2 && native.id !== id2) {
               const idPresent = typeof native.id === "string" && native.id.length > 0;
               const modelPresent = typeof native.model === "string" && native.model.length > 0;
               const modelMatches = native.model === this.options.requestedModel;
-              if (idPresent && !modelMatches && id2 && native.id !== id2 && this.boundIdentity.allowedEffectiveModels.includes(native.model)) {
+              if (idPresent && !modelMatches && id2 && native.id !== id2) {
                 throw new Error("grok_provider_response_id_changed");
               }
               const eventType = route === "/responses" ? event.type : event.object;
               this.responseIdentityFailure = grokResponseIdentityDiagnosticSchema.parse({
                 schema_version: "external_inference_grok_response_identity_v1",
-                reason: !idPresent ? "invalid_id" : !modelMatches ? "model_mismatch" : "changed_id",
+                reason: !idPresent ? "invalid_id" : id2 && native.id !== id2 ? "changed_id" : "model_mismatch",
                 route,
                 response_event: ["response.created", "response.in_progress", "response.completed", "chat.completion.chunk"].includes(String(eventType)) ? eventType : "other",
                 id_present: idPresent,
@@ -35605,11 +35660,11 @@ var init_grok_adapter = __esm({
         isDefault: state.currentModelId === model.modelId
       }];
     });
-    grokPromptUsage = (value, expectedModel, allowedEffectiveModels = [expectedModel]) => {
+    grokPromptUsage = (value, expectedModel) => {
       const usage = grokObject(value);
       if (usage.usageIsIncomplete === true) throw new Error("grok_usage_incomplete");
       const models = Object.keys(grokObject(usage.modelUsage));
-      if (models.length !== 1 || !allowedEffectiveModels.includes(models[0])) throw new Error("grok_effective_model_mismatch");
+      if (models.length !== 1 || models[0] !== expectedModel) throw new Error("grok_effective_model_mismatch");
       const inputTokens = integer2(usage.inputTokens, "input_tokens");
       const outputTokens = integer2(usage.outputTokens, "output_tokens");
       const cachedInputTokens = integer2(usage.cachedReadTokens, "cached_input_tokens");
@@ -35880,7 +35935,7 @@ var init_grok_adapter = __esm({
           const selectedCapability = grokModelCapabilities(catalog).find((entry) => entry.model === input.requestedModel);
           if (!selectedCapability?.supportedReasoningEfforts.includes(input.requestedReasoningEffort)) throw new Error("grok_model_selection_unavailable");
           if (provider && !provider.allowedEffectiveModels) throw new Error("grok_model_identity_contract_missing");
-          const { allowedEffectiveModels, transportModel } = proxy.bindModelIdentity(input.requestedModel, provider?.allowedEffectiveModels);
+          const { transportModel } = proxy.bindModelIdentity(input.requestedModel, provider?.allowedEffectiveModels);
           const metadata = {
             systemPromptOverride: input.systemPrompt,
             agentProfile: agentProfile(Boolean(agent)),
@@ -35947,8 +36002,9 @@ var init_grok_adapter = __esm({
           if (meta3.structuredOutputError || meta3.structuredOutput === void 0) throw new Error("grok_structured_output_missing");
           const text = JSON.stringify(meta3.structuredOutput);
           if (Buffer.byteLength(text) > 3e5) throw new Error("grok_result_too_large");
-          const usage = grokPromptUsage(meta3.usage, input.requestedModel, allowedEffectiveModels);
-          const actualModel = Object.keys(grokObject(grokObject(meta3.usage).modelUsage))[0];
+          const actualModel = proxy.observedModel;
+          if (!actualModel) throw new Error("grok_transport_receipt_missing");
+          const usage = grokPromptUsage(meta3.usage, actualModel);
           if (!proxy.dispatched || !proxy.responded || proxy.terminalFailure) throw new Error(proxy.failureCode ?? "grok_transport_receipt_missing");
           const receipts = proxy.receipts;
           if (!receipts.length || proxy.pendingRequest || receipts.some((receipt) => receipt.model !== actualModel || receipt.requestedReasoningEffort !== input.requestedReasoningEffort || receipt.input === null || receipt.output === null || receipt.cachedInput === null) || receipts.reduce((sum, receipt) => sum + receipt.input, 0) !== usage.inputTokens || receipts.reduce((sum, receipt) => sum + receipt.output, 0) !== usage.outputTokens || receipts.reduce((sum, receipt) => sum + receipt.cachedInput, 0) !== usage.cachedInputTokens || integer2(grokObject(meta3.usage).modelCalls, "model_calls") !== receipts.length) {
@@ -36118,7 +36174,7 @@ import { createServer as createServer5 } from "node:http";
 import { access as access3, lstat as lstat8, readdir as readdir4, realpath as realpath7, rm as rm9 } from "node:fs/promises";
 import { basename as basename4, dirname as dirname9, join as join12, resolve as resolve7 } from "node:path";
 import { fileURLToPath as fileURLToPath2 } from "node:url";
-var import_ajv2, failure, object4, receiptUsage, assertOwnedRuntime, cleanupExoOwnedState, ExoForegroundAdapter;
+var import_ajv2, failure, object4, validModel, receiptUsage, assertOwnedRuntime, cleanupExoOwnedState, ExoForegroundAdapter;
 var init_exo_adapter = __esm({
   "lib/inference-host/exo-adapter.ts"() {
     "use strict";
@@ -36137,6 +36193,7 @@ var init_exo_adapter = __esm({
       dispatchOutcome: dispatched ? "outcome_unknown" : "not_dispatched"
     });
     object4 = (value) => value && typeof value === "object" && !Array.isArray(value) ? value : {};
+    validModel = (value) => typeof value === "string" && Boolean(value.trim()) && value.length <= 256 && !/[\u0000-\u001f\u007f]/u.test(value);
     receiptUsage = (value) => {
       const usage = object4(value);
       for (const key of ["inputTokens", "cachedInputTokens", "outputTokens", "reasoningOutputTokens", "totalTokens"]) {
@@ -36159,6 +36216,7 @@ var init_exo_adapter = __esm({
       const raw = await readInferencePrivateFile(join12(root, "vtx-owner.json"), "Exo state owner");
       const owner2 = raw === null ? {} : object4(JSON.parse(raw));
       if (owner2.adapter !== "exo" || owner2.id !== id2 || owner2.root !== root) throw failure("exo_state_not_owned");
+      return owner2;
     };
     cleanupExoOwnedState = async (stateRoot) => {
       try {
@@ -36297,8 +36355,10 @@ var init_exo_adapter = __esm({
         if (!credential || credential.provider !== provider) throw failure("exo_credential_required", false, "auth");
         const id2 = agent?.durableThread?.threadId ?? randomUUID6();
         const runtimeRoot = agent?.durableThread ? await this.ownedRoot(agent.durableThread) : join12(await realpath7(this.options.stateRoot), `${agent ? "agent" : "attempt"}-${id2}`);
-        const thread = { threadId: id2, threadPath: runtimeRoot, effectiveModel: input.requestedModel, effectiveReasoningEffort: input.requestedReasoningEffort };
-        if (agent?.durableThread && (agent.durableThread.effectiveModel !== thread.effectiveModel || agent.durableThread.effectiveReasoningEffort !== thread.effectiveReasoningEffort)) throw failure("exo_thread_selection_mismatch");
+        let thread = { threadId: id2, threadPath: runtimeRoot, effectiveModel: input.requestedModel, effectiveReasoningEffort: input.requestedReasoningEffort };
+        const owner2 = agent?.durableThread ? await assertOwnedRuntime(runtimeRoot, id2) : null;
+        const boundSelection = owner2?.requestedModel ?? agent?.durableThread?.effectiveModel;
+        if (agent?.durableThread && (boundSelection !== input.requestedModel || agent.durableThread.effectiveReasoningEffort !== thread.effectiveReasoningEffort)) throw failure("exo_thread_selection_mismatch");
         const schema = agent ? agent.outputSchema : JSON.parse(input.outputSchemaJson);
         const validate = new import_ajv2.default({ allErrors: true }).compile(schema);
         const started = Date.now();
@@ -36373,11 +36433,15 @@ var init_exo_adapter = __esm({
               return;
             }
             if (request.url === "/tool") {
-              if (!agent || !dispatched || toolBusy || typeof body.callId !== "string" || toolIds.has(body.callId) || !CODEX_AGENT_DYNAMIC_TOOLS.includes(body.tool) || body.effectiveModel !== input.requestedModel || body.effectiveReasoningEffort !== input.requestedReasoningEffort) throw failure("exo_tool_not_authorized");
+              if (!agent || !dispatched || toolBusy || typeof body.callId !== "string" || toolIds.has(body.callId) || !CODEX_AGENT_DYNAMIC_TOOLS.includes(body.tool) || !validModel(body.effectiveModel) || body.effectiveReasoningEffort !== input.requestedReasoningEffort) throw failure("exo_tool_not_authorized");
               toolBusy = true;
               toolIds.add(body.callId);
               const call = { callId: body.callId, tool: body.tool, arguments: object4(body.arguments) };
               try {
+                if (thread.effectiveModel !== body.effectiveModel) {
+                  thread = { ...thread, effectiveModel: body.effectiveModel };
+                  await agent.onThreadReady?.(thread);
+                }
                 let result2;
                 try {
                   result2 = await agent.executeTool(call);
@@ -36411,7 +36475,10 @@ var init_exo_adapter = __esm({
         try {
           input.signal?.throwIfAborted();
           await ensureInferencePrivateDirectory(runtimeRoot);
-          if (!agent?.durableThread) await writeAtomicInferencePrivateFile(join12(runtimeRoot, "vtx-owner.json"), JSON.stringify({ adapter: "exo", id: id2, root: runtimeRoot }));
+          if (!owner2 || owner2.requestedModel === void 0) await writeAtomicInferencePrivateFile(
+            join12(runtimeRoot, "vtx-owner.json"),
+            JSON.stringify({ adapter: "exo", id: id2, root: runtimeRoot, requestedModel: input.requestedModel })
+          );
           await new Promise((done, reject) => {
             server.once("error", reject);
             server.listen(0, "127.0.0.1", done);
@@ -36432,7 +36499,7 @@ var init_exo_adapter = __esm({
           }
           const result2 = terminal;
           if (!result2 || workerFailure) throw failure("exo_terminal_receipt_missing", dispatched);
-          if (result2.providerModel !== providerModel || result2.reasoningEffort !== input.requestedReasoningEffort) throw failure("exo_effective_identity_mismatch", dispatched, "model");
+          if (!validModel(result2.providerModel) || result2.reasoningEffort !== input.requestedReasoningEffort) throw failure("exo_effective_identity_mismatch", dispatched, "model");
           const usage = receiptUsage(result2.usage);
           if (!usage) throw failure("exo_usage_invalid", dispatched);
           let validOutput = false;
@@ -36444,13 +36511,18 @@ var init_exo_adapter = __esm({
           }
           if (!validOutput || typeof result2.text !== "string") throw failure("exo_output_schema_invalid", dispatched, "schema");
           if (typeof result2.responseId !== "string" || !result2.responseId) throw failure("exo_response_id_missing", dispatched);
+          const effectiveModel = `${provider}/${result2.providerModel}`;
+          if (thread.effectiveModel !== effectiveModel) {
+            thread = { ...thread, effectiveModel };
+            await agent?.onThreadReady?.(thread);
+          }
           if (!agent) input.onTerminalDispatchOutcome?.("confirmed_dispatched");
           return { thread, tools, turn: {
             text: result2.text,
             reasoningContent: null,
             reasoningSummary: null,
             requestedModel: input.requestedModel,
-            effectiveModel: input.requestedModel,
+            effectiveModel,
             requestedReasoningEffort: input.requestedReasoningEffort,
             effectiveReasoningEffort: input.requestedReasoningEffort,
             adapterRequestId: id2,
@@ -36466,7 +36538,7 @@ var init_exo_adapter = __esm({
           const receipt = terminal;
           const completed = receipt !== null || knownFailure?.terminalKnown === true;
           const evidence = receipt ?? knownFailure;
-          const identityQualified = evidence?.providerModel === providerModel && evidence?.reasoningEffort === input.requestedReasoningEffort;
+          const identityQualified = validModel(evidence?.providerModel) && evidence?.reasoningEffort === input.requestedReasoningEffort;
           const outcome = !dispatched ? "not_dispatched" : completed ? "confirmed_dispatched" : "outcome_unknown";
           const knownCode = typeof knownFailure?.code === "string" && /^exo_[a-z_]+$/u.test(knownFailure.code) ? knownFailure.code : null;
           const observedCode = completed ? knownCode ?? (error48 instanceof CodexAppServerError ? error48.code : "exo_turn_failed") : abort.signal.aborted ? "exo_interrupted" : "exo_unconfirmed_result";
@@ -36482,7 +36554,7 @@ var init_exo_adapter = __esm({
             httpStatusCode: ["quota_exceeded", "provider_rate_limited"].includes(code) ? 429 : null,
             retryAtMs: completed && Number.isSafeInteger(evidence?.retryAtMs) && Number(evidence?.retryAtMs) > Date.now() ? Number(evidence?.retryAtMs) : null,
             terminalEvidence: completed ? {
-              effectiveModel: identityQualified ? input.requestedModel : null,
+              effectiveModel: identityQualified ? `${provider}/${evidence.providerModel}` : null,
               effectiveReasoningEffort: identityQualified ? input.requestedReasoningEffort : null,
               adapterRequestId: id2,
               adapterResponseId: typeof evidence?.responseId === "string" && evidence.responseId ? evidence.responseId : null
@@ -40784,7 +40856,8 @@ var init_runner = __esm({
         const mapping2 = jobInput.allowed_model_identities.find(
           (candidate) => candidate.requested_model === jobInput.requested_model
         );
-        if (result2.requestedModel !== jobInput.requested_model || result2.requestedReasoningEffort !== jobInput.requested_reasoning_effort || !mapping2?.effective_models.includes(result2.effectiveModel) || !jobInput.allowed_effective_reasoning_efforts.includes(result2.effectiveReasoningEffort)) {
+        const effectiveModelMatches = mapping2 !== void 0 && jobInput.controller.adapter === (this.options.adapterId ?? "codex") && typeof result2.effectiveModel === "string" && result2.effectiveModel.trim().length > 0 && result2.effectiveModel.length <= 256 && !/[\u0000-\u001f\u007f]/u.test(result2.effectiveModel);
+        if (result2.requestedModel !== jobInput.requested_model || result2.requestedReasoningEffort !== jobInput.requested_reasoning_effort || !effectiveModelMatches || !jobInput.allowed_effective_reasoning_efforts.includes(result2.effectiveReasoningEffort)) {
           throw new InferenceHostRunnerError(
             "effective_model_mismatch",
             "The effective adapter model provenance did not match the immutable allowance."

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@vtxmacro/cli",
-  "version": "2026.9.37",
+  "version": "2026.9.38",
   "description": "VTX Macro CLI, MCP server, and durable external inference host.",
   "type": "module",
   "license": "UNLICENSED",