From f3361dc92870fa951d1e4f205877e57af12f27be Mon Sep 17 00:00:00 2001
From: Peter Steinberger <steipete@gmail.com>
Date: Thu, 14 May 2026 01:36:33 +0100
Subject: [PATCH] test(agents): surface live OpenAI replay auth failures

---
 src/agents/models.profiles.live.test.ts    | 35 ++++++++++++++++++++++
 src/agents/openai-transport-stream.test.ts | 31 +++++++++++++++++++
 src/agents/openai-transport-stream.ts      |  5 ++++
 3 files changed, 71 insertions(+)

diff --git a/src/agents/models.profiles.live.test.ts b/src/agents/models.profiles.live.test.ts
index 12768549902..3fc6554ffa5 100644
--- a/src/agents/models.profiles.live.test.ts
+++ b/src/agents/models.profiles.live.test.ts
@@ -545,6 +545,34 @@ async function completeSimpleWithTimeout<TApi extends Api>(
   }
 }
 
+function requireToolChoicePayload(payload: unknown): unknown | undefined {
+  if (!payload || typeof payload !== "object" || Array.isArray(payload)) {
+    return undefined;
+  }
+  const candidate = payload as { tools?: unknown; tool_choice?: unknown };
+  if (!Array.isArray(candidate.tools) || candidate.tools.length === 0) {
+    return undefined;
+  }
+  return {
+    ...candidate,
+    tool_choice: { type: "function", name: "noop" },
+  };
+}
+
+describe("requireToolChoicePayload", () => {
+  it("requires tool use when a Responses payload has tools", () => {
+    expect(requireToolChoicePayload({ model: "gpt", tools: [{ name: "noop" }] })).toEqual({
+      model: "gpt",
+      tools: [{ name: "noop" }],
+      tool_choice: { type: "function", name: "noop" },
+    });
+  });
+
+  it("leaves payloads without tools unchanged", () => {
+    expect(requireToolChoicePayload({ model: "gpt", tools: [] })).toBeUndefined();
+  });
+});
+
 async function completeOkWithRetry(params: {
   model: Model<Api>;
   apiKey: string;
@@ -982,6 +1010,7 @@ describeLive("live models (profile keys)", () => {
                   apiKey,
                   reasoning: resolveTestReasoning(model),
                   maxTokens: 128,
+                  onPayload: requireToolChoicePayload,
                 },
                 perModelTimeoutMs,
                 `${progressLabel}: tool-only regression first call`,
@@ -1012,6 +1041,7 @@ describeLive("live models (profile keys)", () => {
                     apiKey,
                     reasoning: resolveTestReasoning(model),
                     maxTokens: 128,
+                    onPayload: requireToolChoicePayload,
                   },
                   perModelTimeoutMs,
                   `${progressLabel}: tool-only regression retry ${i + 1}`,
@@ -1025,6 +1055,11 @@ describeLive("live models (profile keys)", () => {
                   .trim();
               }
 
+              if (first.stopReason === "error") {
+                throw new Error(
+                  first.errorMessage || "tool-only regression returned error with no message",
+                );
+              }
               expect(firstText.length).toBe(0);
               if (!toolCall || toolCall.type !== "toolCall") {
                 throw new Error("expected tool call");
diff --git a/src/agents/openai-transport-stream.test.ts b/src/agents/openai-transport-stream.test.ts
index b0d5485bd4d..676f6ceac67 100644
--- a/src/agents/openai-transport-stream.test.ts
+++ b/src/agents/openai-transport-stream.test.ts
@@ -2314,6 +2314,37 @@ describe("openai transport stream", () => {
     });
   });
 
+  it("passes explicit Responses tool_choice when tools are present", () => {
+    const params = buildOpenAIResponsesParams(
+      {
+        id: "gpt-5.4",
+        name: "GPT-5.4",
+        api: "openai-responses",
+        provider: "openai",
+        baseUrl: "https://api.openai.com/v1",
+        reasoning: true,
+        input: ["text"],
+        cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+        contextWindow: 200000,
+        maxTokens: 8192,
+      } satisfies Model<"openai-responses">,
+      {
+        systemPrompt: "system",
+        messages: [],
+        tools: [
+          {
+            name: "lookup_weather",
+            description: "Get forecast",
+            parameters: { type: "object", properties: {}, additionalProperties: false },
+          },
+        ],
+      } as never,
+      { toolChoice: "required" } as never,
+    ) as { tool_choice?: string };
+
+    expect(params.tool_choice).toBe("required");
+  });
+
   it("falls back to strict:false when a native OpenAI tool schema is not strict-compatible", () => {
     const params = buildOpenAIResponsesParams(
       {
diff --git a/src/agents/openai-transport-stream.ts b/src/agents/openai-transport-stream.ts
index d07bd457e58..e2f168f67ee 100644
--- a/src/agents/openai-transport-stream.ts
+++ b/src/agents/openai-transport-stream.ts
@@ -94,6 +94,7 @@ type OpenAIResponsesOptions = BaseStreamOptions & {
   reasoningEffort?: OpenAIReasoningEffort;
   reasoningSummary?: "auto" | "detailed" | "concise" | null;
   serviceTier?: ResponseCreateParamsStreaming["service_tier"];
+  toolChoice?: ResponseCreateParamsStreaming["tool_choice"];
 };
 
 type OpenAICompletionsOptions = BaseStreamOptions & {
@@ -1364,6 +1365,9 @@ export function buildOpenAIResponsesParams(
         transport: "stream",
       }),
     });
+    if (options?.toolChoice) {
+      params.tool_choice = options.toolChoice;
+    }
   }
   if (model.reasoning) {
     if (options?.reasoningEffort || options?.reasoning || options?.reasoningSummary) {
@@ -2188,6 +2192,7 @@ type OpenAIResponsesRequestParams = {
   top_p?: number;
   service_tier?: ResponseCreateParamsStreaming["service_tier"];
   tools?: FunctionTool[];
+  tool_choice?: ResponseCreateParamsStreaming["tool_choice"];
   reasoning?:
     | { effort: OpenAIApiReasoningEffort }
     | {