fix: preserve forced streaming for json clients

Keep provider-required streaming when client prefers JSON. The
Accept: application/json branch no longer flips stream back to false
for forceStream providers, fixing 400 errors on stream-only providers
(e.g. Command Code) for Hermes / Claude Code / other JSON clients.

Fixes #2031

Co-authored-by: Cursor <cursoragent@cursor.com>

open-sse/handlers/chatCore.js

@@ -89,7 +89,7 @@ export async function handleChatCore({ body, modelInfo, credentials, log, onCred
   const acceptHeader = clientRawRequest?.headers?.accept || "";
   const clientPrefersJson = acceptHeader.includes("application/json");
   const clientPrefersSSE = acceptHeader.includes("text/event-stream");
-  if (clientPrefersJson && !clientPrefersSSE && body.stream !== true) {
+  if (clientPrefersJson && !clientPrefersSSE && body.stream !== true && !providerRequiresStreaming) {
     stream = false;
   }
 

tests/unit/force-stream-config.test.js

@@ -1,9 +1,136 @@
 // Guards forceStream moved from chatCore hardcode → PROVIDERS schema (#5).
-import { describe, it, expect } from "vitest";
+import { beforeEach, describe, expect, it, vi } from "vitest";
+
+const { executeMock } = vi.hoisted(() => ({
+  executeMock: vi.fn(),
+}));
+
+vi.mock("../../open-sse/executors/index.js", () => ({
+  getExecutor: vi.fn(() => ({
+    execute: executeMock,
+    refreshCredentials: vi.fn().mockResolvedValue(null),
+  })),
+}));
+
+vi.mock("../../open-sse/utils/requestLogger.js", () => ({
+  createRequestLogger: vi.fn(async () => ({
+    logClientRawRequest: vi.fn(),
+    logRawRequest: vi.fn(),
+    logTargetRequest: vi.fn(),
+    logError: vi.fn(),
+  })),
+}));
+
+vi.mock("../../open-sse/utils/clientDetector.js", () => ({
+  detectClientTool: vi.fn(() => null),
+  isNativePassthrough: vi.fn(() => false),
+}));
+
+vi.mock("../../open-sse/utils/bypassHandler.js", () => ({
+  handleBypassRequest: vi.fn(() => null),
+}));
+
+vi.mock("../../open-sse/utils/streamHandler.js", () => ({
+  createStreamController: vi.fn(() => ({
+    signal: undefined,
+    handleComplete: vi.fn(),
+    handleError: vi.fn(),
+  })),
+}));
+
+vi.mock("../../open-sse/services/tokenRefresh.js", () => ({
+  refreshWithRetry: vi.fn(),
+}));
+
+vi.mock("../../open-sse/utils/proxyFetch.js", () => ({
+  default: vi.fn(),
+  proxyAwareFetch: vi.fn(),
+}));
+
+vi.mock("../../open-sse/translator/formats/claude.js", () => ({
+  normalizeClaudePassthrough: vi.fn(),
+}));
+
+vi.mock("../../open-sse/utils/toolDeduper.js", () => ({
+  dedupeTools: vi.fn((tools) => ({ tools, stripped: [] })),
+}));
+
+vi.mock("../../open-sse/rtk/caveman.js", () => ({
+  injectCaveman: vi.fn(),
+}));
+
+vi.mock("../../open-sse/rtk/ponytail.js", () => ({
+  injectPonytail: vi.fn(),
+}));
+
+vi.mock("../../open-sse/rtk/index.js", () => ({
+  compressMessages: vi.fn(() => null),
+  formatRtkLog: vi.fn(() => ""),
+}));
+
+vi.mock("../../open-sse/rtk/headroom.js", () => ({
+  compressWithHeadroom: vi.fn(async () => null),
+  formatHeadroomLog: vi.fn(() => ""),
+}));
+
+vi.mock("../../open-sse/providers/capabilities.js", () => ({
+  getCapabilitiesForModel: vi.fn(() => ({})),
+}));
+
+vi.mock("../../open-sse/translator/concerns/modality.js", () => ({
+  stripUnsupportedModalities: vi.fn(() => false),
+}));
+
+vi.mock("../../open-sse/translator/concerns/prefetch.js", () => ({
+  prefetchRemoteImages: vi.fn(async () => 0),
+}));
+
+vi.mock("../../open-sse/handlers/chatCore/requestDetail.js", () => ({
+  buildRequestDetail: vi.fn((detail) => detail),
+  extractRequestConfig: vi.fn((body, stream) => ({ body, stream })),
+}));
+
+vi.mock("../../open-sse/utils/error.js", () => ({
+  createErrorResult: vi.fn((status, message) => ({ success: false, status, error: message })),
+  formatProviderError: vi.fn((error) => error.message),
+  parseUpstreamError: vi.fn(),
+}));
+
+vi.mock("@/lib/usageDb.js", () => ({
+  trackPendingRequest: vi.fn(),
+  appendRequestLog: vi.fn(() => Promise.resolve()),
+  saveRequestDetail: vi.fn(() => Promise.resolve()),
+}));
 
 const FORCED = ["openai", "codex", "commandcode"];
 
+function makeOptions(bodyStream) {
+  const body = {
+    model: "gpt-4.1",
+    messages: [{ role: "user", content: "hello" }],
+  };
+  if (bodyStream !== undefined) body.stream = bodyStream;
+
+  return {
+    body,
+    modelInfo: { provider: "openai", model: "gpt-4.1" },
+    credentials: { apiKey: "sk-test" },
+    clientRawRequest: {
+      endpoint: "/v1/chat/completions",
+      body,
+      headers: { accept: "application/json" },
+    },
+    connectionId: "test-connection",
+    log: { debug: vi.fn(), info: vi.fn(), warn: vi.fn(), error: vi.fn() },
+  };
+}
+
 describe("forceStream provider config", () => {
+  beforeEach(() => {
+    executeMock.mockReset();
+    executeMock.mockRejectedValue(new Error("boom"));
+  });
+
   it("only openai/codex/commandcode force streaming", async () => {
     const { PROVIDERS } = await import("../../open-sse/config/providers.js");
     for (const id of FORCED) {
@@ -14,4 +141,13 @@ describe("forceStream provider config", () => {
       expect(PROVIDERS[id]?.forceStream, `${id} not forced`).not.toBe(true);
     }
   });
+
+  it.each([undefined, false])( "keeps forced-stream providers streaming for JSON clients when body.stream is %s", async (bodyStream) => {
+    const { handleChatCore } = await import("../../open-sse/handlers/chatCore.js");
+
+    await handleChatCore(makeOptions(bodyStream));
+
+    expect(executeMock).toHaveBeenCalledTimes(1);
+    expect(executeMock.mock.calls[0][0].stream).toBe(true);
+  });
 });