@qwen-code/qwen-code 0.20.0-preview.0 → 0.20.0

package/chunks/{chunk-JJEFX6U5.js → chunk-AQIEINAY.js}

@@ -1,23 +1,53 @@
 // Force strict mode and setup for ESM
 "use strict";
+import {
+  InvalidStreamError,
+  OpenAIContentConverter,
+  ProtocolTagSanitizedEvent,
+  TaggedThinkingParser,
+  getRateLimitErrorDetails,
+  logProtocolTagSanitized,
+  openaiRequestCaptureContext
+} from "./chunk-ZFKNWQGQ.js";
+import {
+  RequestTokenizer
+} from "./chunk-FLSVD6BS.js";
 import {
   DASHSCOPE_PROXY_BASE_URL,
   DEFAULT_DASHSCOPE_BASE_URL,
   DEFAULT_MAX_RETRIES,
+  DEFAULT_STREAM_IDLE_TIMEOUT_MS,
+  MAX_STREAM_IDLE_TIMEOUT_MS,
+  QWEN_STREAM_IDLE_TIMEOUT_MS_ENV,
   resolveRequestTimeout
-} from "./chunk-GLNBDYUX.js";
+} from "./chunk-FEOZBUMT.js";
 import {
-  buildRuntimeFetchOptions
-} from "./chunk-5I5BM454.js";
+  createChildAbortController
+} from "./chunk-DJ2GSRLV.js";
+import {
+  buildRuntimeFetchOptions,
+  redactProxyError
+} from "./chunk-6EFTQYW7.js";
 import {
   defaultOutputCeiling,
+  getToolCallPreparations,
   hasExplicitOutputLimit,
   parsePositiveIntegerEnvValue,
+  reconcileMaxTokens,
+  runtimeDiagnostics,
+  safeJsonParse,
   tokenLimit
-} from "./chunk-JYS55H6Z.js";
+} from "./chunk-MAQQGNQ5.js";
+import {
+  GenerateContentResponse
+} from "./chunk-TYQMGACW.js";
 import {
-  createDebugLogger
-} from "./chunk-OQB4EXO4.js";
+  createDebugLogger,
+  getErrorMessage,
+  getErrorStatus,
+  getErrorType,
+  isAbortError
+} from "./chunk-C27BYUX6.js";
 import {
   init_esbuild_shims
 } from "./chunk-5O2XNYP6.js";
@@ -25,6 +55,24 @@ import {
   __name
 } from "./chunk-J2S4EL5Y.js";
 
+// packages/core/src/core/openaiContentGenerator/index.ts
+init_esbuild_shims();
+
+// packages/core/src/core/openaiContentGenerator/openaiContentGenerator.ts
+init_esbuild_shims();
+
+// packages/core/src/core/openaiContentGenerator/provider/index.ts
+init_esbuild_shims();
+
+// packages/core/src/core/openaiContentGenerator/provider/modelscope.ts
+init_esbuild_shims();
+
+// packages/core/src/core/openaiContentGenerator/provider/default.ts
+init_esbuild_shims();
+
+// node_modules/openai/index.mjs
+init_esbuild_shims();
+
 // node_modules/openai/client.mjs
 init_esbuild_shims();
 
@@ -70,12 +118,12 @@ init_esbuild_shims();
 
 // node_modules/openai/internal/errors.mjs
 init_esbuild_shims();
-function isAbortError(err) {
+function isAbortError2(err) {
   return typeof err === "object" && err !== null && // Spec-compliant fetch implementations
   ("name" in err && err.name === "AbortError" || // Expo fetch
   "message" in err && String(err.message).includes("FetchRequestCanceledException"));
 }
-__name(isAbortError, "isAbortError");
+__name(isAbortError2, "isAbortError");
 var castToError = /* @__PURE__ */ __name((err) => {
   if (err instanceof Error)
     return err;
@@ -1180,7 +1228,7 @@ var Stream = class _Stream {
         }
         done = true;
       } catch (e) {
-        if (isAbortError(e))
+        if (isAbortError2(e))
           return;
         throw e;
       } finally {
@@ -1225,7 +1273,7 @@ var Stream = class _Stream {
         }
         done = true;
       } catch (e) {
-        if (isAbortError(e))
+        if (isAbortError2(e))
           return;
         throw e;
       } finally {
@@ -6619,7 +6667,7 @@ var OpenAI = class {
       if (options.signal?.aborted) {
         throw new APIUserAbortError();
       }
-      const isTimeout = isAbortError(response) || /timed? ?out/i.test(String(response) + ("cause" in response ? String(response.cause) : ""));
+      const isTimeout = isAbortError2(response) || /timed? ?out/i.test(String(response) + ("cause" in response ? String(response.cause) : ""));
       if (retriesRemaining) {
         loggerFor(this).info(`[${requestLogID}] connection ${isTimeout ? "timed out" : "failed"} - ${retryMessage}`);
         loggerFor(this).debug(`[${requestLogID}] connection ${isTimeout ? "timed out" : "failed"} (${retryMessage})`, formatRequestDetails({
@@ -6869,14 +6917,10 @@ OpenAI.Responses = Responses;
 OpenAI.Evals = Evals;
 OpenAI.Containers = Containers;
 
-// node_modules/openai/index.mjs
-init_esbuild_shims();
-
 // node_modules/openai/azure.mjs
 init_esbuild_shims();
 
 // packages/core/src/core/openaiContentGenerator/provider/default.ts
-init_esbuild_shims();
 function shouldMirrorReasoningContentForQwen3(model) {
   return model.toLowerCase().includes("qwen3");
 }
@@ -7012,14 +7056,40 @@ var DefaultOpenAICompatibleProvider = class {
   }
 };
 
+// packages/core/src/core/openaiContentGenerator/provider/modelscope.ts
+var ModelScopeOpenAICompatibleProvider = class extends DefaultOpenAICompatibleProvider {
+  static {
+    __name(this, "ModelScopeOpenAICompatibleProvider");
+  }
+  /**
+   * Checks if the configuration is for ModelScope.
+   */
+  static isModelScopeProvider(config) {
+    const baseUrl = config.baseUrl ?? "";
+    if (!baseUrl) return false;
+    try {
+      const hostname = new URL(baseUrl).hostname.toLowerCase();
+      return hostname === "modelscope.cn" || hostname.endsWith(".modelscope.cn");
+    } catch {
+      return false;
+    }
+  }
+  /**
+   * ModelScope does not support `stream_options` when `stream` is false.
+   * This method removes `stream_options` if `stream` is not true.
+   */
+  buildRequest(request, userPromptId) {
+    const newRequest = super.buildRequest(request, userPromptId);
+    if (!newRequest.stream) {
+      delete newRequest.stream_options;
+    }
+    return newRequest;
+  }
+};
+
 // packages/core/src/core/openaiContentGenerator/provider/dashscope.ts
 init_esbuild_shims();
 var debugLogger = createDebugLogger("DashScopeOpenAICompatibleProvider");
-var DASHSCOPE_REGIONAL_HOSTS = [
-  "dashscope.aliyuncs.com",
-  "dashscope-intl.aliyuncs.com",
-  "dashscope-us.aliyuncs.com"
-];
 var DashScopeOpenAICompatibleProvider = class _DashScopeOpenAICompatibleProvider extends DefaultOpenAICompatibleProvider {
   static {
     __name(this, "DashScopeOpenAICompatibleProvider");
@@ -7032,7 +7102,7 @@ var DashScopeOpenAICompatibleProvider = class _DashScopeOpenAICompatibleProvider
   }
   /**
    * Determines whether to use the DashScope-compatible provider.
-   * Covers the official regional hosts (DASHSCOPE_REGIONAL_HOSTS),
+   * Covers dashscope.aliyuncs.com, dashscope-intl.aliyuncs.com,
    * Token Plan endpoints under token-plan.<region>.maas.aliyuncs.com,
    * internal Alibaba domains (*.alibaba-inc.com, *.aliyun-inc.com),
    * and proxy matches.
@@ -7052,9 +7122,7 @@ var DashScopeOpenAICompatibleProvider = class _DashScopeOpenAICompatibleProvider
     } catch {
       hostname = null;
     }
-    const isDashscopeOrigin = hostname !== null && DASHSCOPE_REGIONAL_HOSTS.some(
-      (host) => hostname === host || hostname.endsWith("." + host)
-    );
+    const isDashscopeOrigin = hostname !== null && (hostname === "dashscope.aliyuncs.com" || hostname === "dashscope-intl.aliyuncs.com" || hostname.endsWith(".dashscope.aliyuncs.com") || hostname.endsWith(".dashscope-intl.aliyuncs.com"));
     const isTokenPlanOrigin = hostname !== null && hostname.startsWith("token-plan.") && hostname.endsWith(".maas.aliyuncs.com");
     const isInternalOrigin = hostname !== null && (hostname.endsWith(".alibaba-inc.com") || hostname.endsWith(".aliyun-inc.com"));
     const normalizedProxyUrl = DASHSCOPE_PROXY_BASE_URL?.endsWith("/") ? DASHSCOPE_PROXY_BASE_URL.slice(0, -1) : DASHSCOPE_PROXY_BASE_URL;
@@ -7382,9 +7450,1679 @@ var DashScopeOpenAICompatibleProvider = class _DashScopeOpenAICompatibleProvider
   }
 };
 
+// packages/core/src/core/openaiContentGenerator/provider/deepseek.ts
+init_esbuild_shims();
+
+// packages/core/src/core/openaiContentGenerator/provider/utils.ts
+init_esbuild_shims();
+function ensureReasoningContentOnAssistantMessage(message) {
+  if (message.role !== "assistant") {
+    return message;
+  }
+  const assistant = message;
+  if (typeof assistant.reasoning_content === "string") {
+    return message;
+  }
+  return {
+    ...assistant,
+    reasoning_content: ""
+  };
+}
+__name(ensureReasoningContentOnAssistantMessage, "ensureReasoningContentOnAssistantMessage");
+
+// packages/core/src/core/openaiContentGenerator/provider/deepseek.ts
+function isDeepSeekHostname(contentGeneratorConfig) {
+  const baseUrl = contentGeneratorConfig.baseUrl ?? "";
+  if (!baseUrl) return false;
+  try {
+    const hostname = new URL(baseUrl).hostname.toLowerCase();
+    return hostname === "api.deepseek.com" || hostname.endsWith(".api.deepseek.com");
+  } catch {
+    return false;
+  }
+}
+__name(isDeepSeekHostname, "isDeepSeekHostname");
+function isDeepSeekProvider(contentGeneratorConfig) {
+  if (isDeepSeekHostname(contentGeneratorConfig)) return true;
+  const model = contentGeneratorConfig.model ?? "";
+  return model.toLowerCase().includes("deepseek");
+}
+__name(isDeepSeekProvider, "isDeepSeekProvider");
+var DeepSeekOpenAICompatibleProvider = class extends DefaultOpenAICompatibleProvider {
+  static {
+    __name(this, "DeepSeekOpenAICompatibleProvider");
+  }
+  constructor(contentGeneratorConfig, cliConfig) {
+    super(contentGeneratorConfig, cliConfig);
+  }
+  /**
+   * Backward-compatible static delegates for the free `isDeepSeek*`
+   * helpers. New call sites should import the free functions directly to
+   * avoid coupling to this class.
+   */
+  static isDeepSeekProvider = isDeepSeekProvider;
+  static isDeepSeekHostname = isDeepSeekHostname;
+  /**
+   * DeepSeek's API requires message content to be a plain string, not an
+   * array of content parts. Flatten any text-part arrays into joined
+   * strings; non-text parts (image_url, audio, …) are replaced with a
+   * `[Unsupported content type: <type>]` placeholder so the request still
+   * goes through with a textual breadcrumb rather than silently dropping
+   * the part or raising mid-conversation. Also translate the standard
+   * `reasoning.effort` config into DeepSeek's flat `reasoning_effort`
+   * body parameter — but only on actual DeepSeek hostnames, since the
+   * model-name fallback above can match self-hosted/strict OpenAI-compat
+   * backends that don't accept the DeepSeek extension.
+   */
+  buildRequest(request, userPromptId) {
+    const baseRequest = super.buildRequest(request, userPromptId);
+    const reshaped = isDeepSeekHostname(this.contentGeneratorConfig) ? translateReasoningEffort(baseRequest) : baseRequest;
+    if (!reshaped.messages?.length) {
+      return reshaped;
+    }
+    const messages = reshaped.messages.map((message) => {
+      const flattened = flattenContentParts(message);
+      return ensureReasoningContentOnAssistantMessage(flattened);
+    });
+    return {
+      ...reshaped,
+      messages
+    };
+  }
+  getDefaultGenerationConfig() {
+    return {
+      temperature: 0
+    };
+  }
+};
+function flattenContentParts(message) {
+  if (!("content" in message)) {
+    return message;
+  }
+  const { content } = message;
+  if (typeof content === "string" || content === null || content === void 0) {
+    return message;
+  }
+  if (!Array.isArray(content)) {
+    return message;
+  }
+  const text = content.map((part) => {
+    if (typeof part === "string") {
+      return part;
+    }
+    if (part.type === "text") {
+      return part.text ?? "";
+    }
+    return `[Unsupported content type: ${part.type}]`;
+  }).join("\n\n");
+  return {
+    ...message,
+    content: text
+  };
+}
+__name(flattenContentParts, "flattenContentParts");
+function translateReasoningEffort(request) {
+  const r = request;
+  const nested = r["reasoning"];
+  const nestedEffort = nested?.effort;
+  if (typeof nestedEffort !== "string" || !nestedEffort) {
+    return request;
+  }
+  const next = { ...r };
+  if (typeof next["reasoning_effort"] !== "string" || !next["reasoning_effort"]) {
+    let normalized = nestedEffort;
+    if (normalized === "low" || normalized === "medium") normalized = "high";
+    else if (normalized === "xhigh") normalized = "max";
+    next["reasoning_effort"] = normalized;
+  }
+  if (nested && Object.keys(nested).length === 1) {
+    delete next["reasoning"];
+  } else if (nested) {
+    const { effort: _drop, ...rest } = nested;
+    next["reasoning"] = rest;
+  }
+  return next;
+}
+__name(translateReasoningEffort, "translateReasoningEffort");
+
+// packages/core/src/core/openaiContentGenerator/provider/zai.ts
+init_esbuild_shims();
+var debugLogger2 = createDebugLogger("ZAI");
+function isZaiHostname(contentGeneratorConfig) {
+  const baseUrl = contentGeneratorConfig.baseUrl;
+  if (!baseUrl) {
+    return false;
+  }
+  try {
+    const hostname = new URL(baseUrl).hostname.toLowerCase();
+    return hostname === "z.ai" || hostname.endsWith(".z.ai") || hostname === "bigmodel.cn" || hostname.endsWith(".bigmodel.cn");
+  } catch {
+    return false;
+  }
+}
+__name(isZaiHostname, "isZaiHostname");
+function isZaiProvider(contentGeneratorConfig) {
+  if (isZaiHostname(contentGeneratorConfig)) {
+    return true;
+  }
+  const model = contentGeneratorConfig.model ?? "";
+  return model.toLowerCase().startsWith("glm-");
+}
+__name(isZaiProvider, "isZaiProvider");
+var ZaiOpenAICompatibleProvider = class extends DefaultOpenAICompatibleProvider {
+  static {
+    __name(this, "ZaiOpenAICompatibleProvider");
+  }
+  static isZaiProvider = isZaiProvider;
+  static isZaiHostname = isZaiHostname;
+  // Latch so the skipped-flatten warning fires once per provider lifetime.
+  nonZaiHostnameFlattenWarned = false;
+  buildRequest(request, userPromptId) {
+    const baseRequest = super.buildRequest(request, userPromptId);
+    if (isZaiHostname(this.contentGeneratorConfig)) {
+      return flattenReasoningEffort(baseRequest);
+    }
+    const reasoning = baseRequest["reasoning"];
+    if (reasoning?.effort && !baseRequest["reasoning_effort"] && !this.nonZaiHostnameFlattenWarned) {
+      debugLogger2.warn(
+        `GLM model '${this.contentGeneratorConfig.model ?? "unknown"}' on a non-Z.ai hostname; leaving nested reasoning.effort='${String(
+          reasoning.effort
+        )}' unflattened (reasoning_effort reshape is hostname-gated).`
+      );
+      this.nonZaiHostnameFlattenWarned = true;
+    }
+    return baseRequest;
+  }
+};
+function flattenReasoningEffort(request) {
+  const r = request;
+  const nested = r["reasoning"];
+  const effort = nested?.effort;
+  if (typeof effort !== "string" || !effort) {
+    return request;
+  }
+  const next = { ...r };
+  if (typeof next["reasoning_effort"] !== "string" || !next["reasoning_effort"]) {
+    next["reasoning_effort"] = effort;
+  }
+  if (nested && Object.keys(nested).length === 1) {
+    delete next["reasoning"];
+  } else if (nested) {
+    const { effort: _drop, ...rest } = nested;
+    next["reasoning"] = rest;
+  }
+  return next;
+}
+__name(flattenReasoningEffort, "flattenReasoningEffort");
+
+// packages/core/src/core/openaiContentGenerator/provider/minimax.ts
+init_esbuild_shims();
+var MINIMAX_KNOWN_HOSTS = ["api.minimaxi.com", "api.minimax.io"];
+var MINIMAX_HOST_SUFFIXES = [".minimaxi.com", ".minimax.io"];
+var MiniMaxOpenAICompatibleProvider = class extends DefaultOpenAICompatibleProvider {
+  static {
+    __name(this, "MiniMaxOpenAICompatibleProvider");
+  }
+  static isMiniMaxProvider(config) {
+    if (!config.baseUrl) return false;
+    try {
+      const hostname = new URL(config.baseUrl).hostname.toLowerCase();
+      if (MINIMAX_KNOWN_HOSTS.includes(hostname)) {
+        return true;
+      }
+      return MINIMAX_HOST_SUFFIXES.some((suffix) => hostname.endsWith(suffix));
+    } catch {
+      return false;
+    }
+  }
+  getResponseParsingOptions() {
+    return { taggedThinkingTags: true };
+  }
+};
+
+// packages/core/src/core/openaiContentGenerator/provider/mistral.ts
+init_esbuild_shims();
+var MISTRAL_API_HOST = "api.mistral.ai";
+var MISTRAL_MODEL_MARKERS = [
+  "mistral",
+  "mixtral",
+  "codestral",
+  "ministral",
+  "pixtral",
+  "magistral",
+  "devstral"
+];
+function isMistralHostname(config) {
+  const baseUrl = config.baseUrl ?? "";
+  if (!baseUrl) return false;
+  try {
+    const hostname = new URL(baseUrl).hostname.toLowerCase();
+    return hostname === MISTRAL_API_HOST || hostname.endsWith(`.${MISTRAL_API_HOST}`);
+  } catch {
+    return false;
+  }
+}
+__name(isMistralHostname, "isMistralHostname");
+function isMistralProvider(config) {
+  if (isMistralHostname(config)) return true;
+  const model = config.model?.toLowerCase() ?? "";
+  return MISTRAL_MODEL_MARKERS.some((marker) => model.includes(marker));
+}
+__name(isMistralProvider, "isMistralProvider");
+var MistralOpenAICompatibleProvider = class extends DefaultOpenAICompatibleProvider {
+  static {
+    __name(this, "MistralOpenAICompatibleProvider");
+  }
+  static isMistralProvider = isMistralProvider;
+  buildRequest(request, userPromptId) {
+    const baseRequest = super.buildRequest(request, userPromptId);
+    return {
+      ...baseRequest,
+      messages: baseRequest.messages.map(stripReasoningContent)
+    };
+  }
+};
+function stripReasoningContent(message) {
+  if (!("reasoning_content" in message)) {
+    return message;
+  }
+  const next = { ...message };
+  delete next["reasoning_content"];
+  return next;
+}
+__name(stripReasoningContent, "stripReasoningContent");
+
+// packages/core/src/core/openaiContentGenerator/provider/mimo.ts
+init_esbuild_shims();
+function isMiMoProvider(contentGeneratorConfig) {
+  const baseUrl = contentGeneratorConfig.baseUrl ?? "";
+  if (baseUrl) {
+    try {
+      const hostname = new URL(baseUrl).hostname.toLowerCase();
+      if (hostname === "xiaomimimo.com" || hostname.endsWith(".xiaomimimo.com")) {
+        return true;
+      }
+    } catch {
+    }
+  }
+  const model = contentGeneratorConfig.model ?? "";
+  return model.toLowerCase().startsWith("mimo-");
+}
+__name(isMiMoProvider, "isMiMoProvider");
+var MiMoOpenAICompatibleProvider = class extends DefaultOpenAICompatibleProvider {
+  static {
+    __name(this, "MiMoOpenAICompatibleProvider");
+  }
+  constructor(contentGeneratorConfig, cliConfig) {
+    super(contentGeneratorConfig, cliConfig);
+  }
+  static isMiMoProvider = isMiMoProvider;
+  buildRequest(request, userPromptId) {
+    const baseRequest = super.buildRequest(request, userPromptId);
+    if (!baseRequest.messages?.length) {
+      return baseRequest;
+    }
+    return {
+      ...baseRequest,
+      messages: baseRequest.messages.map(
+        ensureReasoningContentOnAssistantMessage
+      )
+    };
+  }
+  getRequestContextOverrides() {
+    return {
+      splitToolMedia: this.contentGeneratorConfig.splitToolMedia ?? true
+    };
+  }
+};
+
+// packages/core/src/core/openaiContentGenerator/pipeline.ts
+init_esbuild_shims();
+
+// packages/core/src/core/openaiContentGenerator/streamingToolCallParser.ts
+init_esbuild_shims();
+var debugLogger3 = createDebugLogger("STREAMING_TOOL_CALL_PARSER");
+var StreamingToolCallParser = class {
+  static {
+    __name(this, "StreamingToolCallParser");
+  }
+  /** Accumulated buffer containing all received chunks for each tool call index */
+  buffers = /* @__PURE__ */ new Map();
+  /** Current nesting depth in JSON structure for each tool call index */
+  depths = /* @__PURE__ */ new Map();
+  /** Whether we're currently inside a string literal for each tool call index */
+  inStrings = /* @__PURE__ */ new Map();
+  /** Whether the next character should be treated as escaped for each tool call index */
+  escapes = /* @__PURE__ */ new Map();
+  /** Metadata for each tool call index */
+  toolCallMeta = /* @__PURE__ */ new Map();
+  namelessToolCallIndices = /* @__PURE__ */ new Set();
+  /** Map from tool call ID to actual index used for storage */
+  idToIndexMap = /* @__PURE__ */ new Map();
+  /** Remapped slots awaiting a stable ID from a later chunk. */
+  pendingIndexRemaps = /* @__PURE__ */ new Map();
+  /** Counter for generating new indices when collisions occur */
+  nextAvailableIndex = 0;
+  conflictingToolCallIdentity = false;
+  invalidToolCallIndex = false;
+  /**
+   * Processes a new chunk of tool call data and attempts to parse complete JSON objects
+   *
+   * Handles the core problems of streaming tool call parsing:
+   * - Resolves index collisions when the same index is reused for different tool calls
+   * - Routes chunks without IDs to the correct incomplete tool call
+   * - Tracks JSON parsing state (depth, string boundaries, escapes) per tool call
+   * - Attempts parsing only when JSON structure is complete (depth = 0)
+   * - Repairs common issues like unclosed strings
+   *
+   * @param index - Tool call index from streaming response (may collide with existing calls)
+   * @param chunk - String chunk that may be empty, partial JSON, or complete data
+   * @param id - Optional tool call ID for collision detection and chunk routing
+   * @param name - Optional function name stored as metadata
+   * @returns ToolCallParseResult with completion status, parsed value, and repair info
+   */
+  addChunk(index, chunk, id, name) {
+    const validName = name?.trim() || void 0;
+    if (!Number.isSafeInteger(index) || index < 0) {
+      this.conflictingToolCallIdentity = true;
+      this.invalidToolCallIndex = true;
+      return {
+        complete: false,
+        error: new Error(`Invalid tool call index: ${index}`)
+      };
+    }
+    if (!id && !validName && !chunk.trim()) {
+      const depth2 = this.depths.get(index) ?? 0;
+      const inString2 = this.inStrings.get(index) ?? false;
+      if (!this.buffers.has(index) || depth2 === 0 && !inString2) {
+        return { complete: false };
+      }
+    }
+    let actualIndex = index;
+    const isKnownId = Boolean(id && this.idToIndexMap.has(id));
+    const existingName = this.toolCallMeta.get(index)?.name;
+    const isNameOnlyDelta = Boolean(
+      validName && chunk.length === 0 && (!existingName || existingName === validName)
+    );
+    if (id) {
+      if (this.idToIndexMap.has(id)) {
+        actualIndex = this.idToIndexMap.get(id);
+      } else if (this.pendingIndexRemaps.has(index)) {
+        actualIndex = this.pendingIndexRemaps.get(index);
+        this.pendingIndexRemaps.delete(index);
+        this.idToIndexMap.set(id, actualIndex);
+      } else {
+        if (this.buffers.has(index)) {
+          const existingBuffer = this.buffers.get(index);
+          const existingDepth = this.depths.get(index);
+          const existingMeta = this.toolCallMeta.get(index);
+          if (existingMeta?.id && existingMeta.id !== id) {
+            let existingComplete = existingDepth === 0;
+            if (existingComplete && existingBuffer.trim()) {
+              try {
+                JSON.parse(existingBuffer);
+              } catch {
+                existingComplete = false;
+              }
+            }
+            if (existingComplete) {
+              actualIndex = 0;
+              while (this.buffers.has(actualIndex)) actualIndex += 1;
+              if (!existingMeta.name) {
+                this.conflictingToolCallIdentity = true;
+              }
+            } else {
+              this.conflictingToolCallIdentity = true;
+            }
+          }
+        }
+        this.idToIndexMap.set(id, actualIndex);
+      }
+    } else if (!isNameOnlyDelta) {
+      if (this.pendingIndexRemaps.has(index)) {
+        actualIndex = this.pendingIndexRemaps.get(index);
+        const existingBuffer = this.buffers.get(actualIndex);
+        const existingDepth = this.depths.get(actualIndex);
+        if (existingDepth === 0 && existingBuffer.trim()) {
+          try {
+            JSON.parse(existingBuffer);
+            actualIndex = this.findMostRecentIncompleteIndex();
+          } catch {
+          }
+        }
+      } else if (this.buffers.has(index)) {
+        const existingBuffer = this.buffers.get(index);
+        const existingDepth = this.depths.get(index);
+        if (existingDepth > 0 || !existingBuffer.trim()) {
+          actualIndex = index;
+        } else {
+          try {
+            JSON.parse(existingBuffer);
+            actualIndex = this.findMostRecentIncompleteIndex();
+          } catch {
+            actualIndex = index;
+          }
+        }
+      }
+    }
+    if (!this.buffers.has(actualIndex)) {
+      this.buffers.set(actualIndex, "");
+      this.depths.set(actualIndex, 0);
+      this.inStrings.set(actualIndex, false);
+      this.escapes.set(actualIndex, false);
+      this.toolCallMeta.set(actualIndex, {});
+    }
+    const currentBuffer = this.buffers.get(actualIndex);
+    const currentDepth = this.depths.get(actualIndex);
+    const meta = this.toolCallMeta.get(actualIndex);
+    if (chunk.length === 0 && (id || validName)) {
+      if (id) meta.id = id;
+      if (validName && !meta.name) meta.name = validName;
+      if (!meta.name && meta.id) {
+        this.namelessToolCallIndices.add(actualIndex);
+      } else {
+        this.namelessToolCallIndices.delete(actualIndex);
+      }
+      if (!meta.id && actualIndex !== index) {
+        this.pendingIndexRemaps.set(index, actualIndex);
+      }
+      return { actualIndex, complete: false };
+    }
+    if (isKnownId && currentDepth === 0) {
+      if (currentBuffer.trim()) {
+        try {
+          JSON.parse(currentBuffer);
+          debugLogger3.debug(
+            `Ignoring replay chunk for completed toolCall id=${id}`
+          );
+          return { actualIndex, complete: false };
+        } catch {
+        }
+      }
+    }
+    const identityChanged = Boolean(id && meta.id && meta.id !== id);
+    if (id) meta.id = id;
+    if (validName) {
+      if (!identityChanged && meta.name && meta.name !== validName) {
+        this.conflictingToolCallIdentity = true;
+      } else {
+        meta.name = validName;
+      }
+    }
+    if (!meta.id && actualIndex !== index) {
+      this.pendingIndexRemaps.set(index, actualIndex);
+    }
+    const currentInString = this.inStrings.get(actualIndex);
+    const currentEscape = this.escapes.get(actualIndex);
+    const newBuffer = currentBuffer + chunk;
+    this.buffers.set(actualIndex, newBuffer);
+    if (!meta.name && (meta.id || /\S/.test(newBuffer))) {
+      this.namelessToolCallIndices.add(actualIndex);
+    } else {
+      this.namelessToolCallIndices.delete(actualIndex);
+    }
+    let depth = currentDepth;
+    let inString = currentInString;
+    let escape2 = currentEscape;
+    for (const char of chunk) {
+      if (!inString) {
+        if (char === "{" || char === "[") depth++;
+        else if (char === "}" || char === "]") depth--;
+      }
+      if (char === '"' && !escape2) {
+        inString = !inString;
+      }
+      escape2 = char === "\\" && !escape2;
+    }
+    this.depths.set(actualIndex, depth);
+    this.inStrings.set(actualIndex, inString);
+    this.escapes.set(actualIndex, escape2);
+    if (depth === 0 && newBuffer.trim().length > 0) {
+      try {
+        const parsed = JSON.parse(newBuffer);
+        return { actualIndex, complete: true, value: parsed };
+      } catch (e) {
+        if (inString) {
+          try {
+            const repaired = JSON.parse(newBuffer + '"');
+            return {
+              actualIndex,
+              complete: true,
+              value: repaired,
+              repaired: true
+            };
+          } catch {
+          }
+        }
+        return {
+          actualIndex,
+          complete: false,
+          error: e instanceof Error ? e : new Error(String(e))
+        };
+      }
+    }
+    return { actualIndex, complete: false };
+  }
+  /**
+   * Gets the current tool call metadata for a specific index
+   *
+   * @param index - The tool call index
+   * @returns Object containing id and name if available
+   */
+  getToolCallMeta(index) {
+    return this.toolCallMeta.get(index) || {};
+  }
+  hasNamelessToolCall() {
+    return this.namelessToolCallIndices.size > 0;
+  }
+  hasConflictingToolCallIdentity() {
+    return this.conflictingToolCallIdentity;
+  }
+  hasInvalidToolCallIndex() {
+    return this.invalidToolCallIndex;
+  }
+  hasInvalidToolCallArguments() {
+    for (const [index, buffer] of this.buffers.entries()) {
+      if (!this.toolCallMeta.get(index)?.name || buffer.length === 0) continue;
+      try {
+        const args = JSON.parse(buffer);
+        if (typeof args !== "object" || args === null || Array.isArray(args)) {
+          return true;
+        }
+      } catch {
+        return true;
+      }
+    }
+    return false;
+  }
+  /**
+   * Gets all completed tool calls that are ready to be emitted
+   *
+   * Attempts to parse accumulated buffers using multiple strategies:
+   * 1. Standard JSON.parse()
+   * 2. Auto-close unclosed strings and retry
+   * 3. Fallback to safeJsonParse for malformed data
+   *
+   * Only returns tool calls with name metadata. An empty buffer yields
+   * empty arguments ({}), matching the non-streaming path for no-argument
+   * tools. Should be called when streaming is complete (finish_reason is
+   * present).
+   *
+   * @returns Array of completed tool calls with their metadata and parsed arguments
+   */
+  getCompletedToolCalls() {
+    const completed = [];
+    const emittedIds = /* @__PURE__ */ new Set();
+    for (const [index, buffer] of this.buffers.entries()) {
+      const meta = this.toolCallMeta.get(index);
+      if (meta?.name) {
+        if (meta.id) {
+          if (emittedIds.has(meta.id)) {
+            continue;
+          }
+          emittedIds.add(meta.id);
+        }
+        let args = {};
+        if (buffer.trim()) {
+          try {
+            args = JSON.parse(buffer);
+          } catch {
+            const inString = this.inStrings.get(index);
+            if (inString) {
+              try {
+                args = JSON.parse(buffer + '"');
+              } catch {
+                args = safeJsonParse(buffer, {});
+              }
+            } else {
+              args = safeJsonParse(buffer, {});
+            }
+          }
+          if (typeof args !== "object" || args === null || Array.isArray(args)) {
+            debugLogger3.debug(
+              `Collapsing non-object arguments for tool call ${meta.name} (id=${meta.id}) at index ${index}; buffer likely polluted by a misrouted fragment`
+            );
+            args = {};
+          }
+        } else {
+          debugLogger3.debug(
+            `Emitting no-argument tool call ${meta.name} (id=${meta.id}) at index ${index} with empty buffer`
+          );
+        }
+        completed.push({
+          id: meta.id,
+          name: meta.name,
+          args,
+          index
+        });
+      }
+    }
+    return completed;
+  }
+  /**
+   * Finds the next available index for a new tool call
+   *
+   * Scans indices starting from nextAvailableIndex to find one that's safe to use.
+   * Reuses indices never claimed by a named tool call or with incomplete
+   * parsing states. Skips indices with complete tool call data — including
+   * no-argument calls with empty buffers — to prevent overwriting.
+   *
+   * @returns The next available index safe for storing a new tool call
+   */
+  findNextAvailableIndex() {
+    while (this.buffers.has(this.nextAvailableIndex)) {
+      const buffer = this.buffers.get(this.nextAvailableIndex);
+      const depth = this.depths.get(this.nextAvailableIndex);
+      const meta = this.toolCallMeta.get(this.nextAvailableIndex);
+      if (!meta?.name || depth > 0 || !meta?.id) {
+        return this.nextAvailableIndex;
+      }
+      if (buffer.trim()) {
+        try {
+          JSON.parse(buffer);
+        } catch {
+          return this.nextAvailableIndex;
+        }
+      }
+      this.nextAvailableIndex++;
+    }
+    return this.nextAvailableIndex++;
+  }
+  /**
+   * Finds the most recent incomplete tool call index
+   *
+   * Used when continuation chunks arrive without IDs. Scans existing tool calls
+   * to find the highest index with incomplete parsing state (depth > 0, empty
+   * buffer with no name metadata yet, or unparseable JSON). Falls back to
+   * creating a new index if none found.
+   *
+   * @returns The index of the most recent incomplete tool call, or a new available index
+   */
+  findMostRecentIncompleteIndex() {
+    let maxIndex = -1;
+    for (const [index, buffer] of this.buffers.entries()) {
+      const depth = this.depths.get(index);
+      const meta = this.toolCallMeta.get(index);
+      if (meta?.id && (depth > 0 || !buffer.trim() && !meta?.name)) {
+        maxIndex = Math.max(maxIndex, index);
+      } else if (buffer.trim()) {
+        try {
+          JSON.parse(buffer);
+        } catch {
+          maxIndex = Math.max(maxIndex, index);
+        }
+      }
+    }
+    return maxIndex >= 0 ? maxIndex : this.findNextAvailableIndex();
+  }
+  /**
+   * Resets the parser state for a specific tool call index
+   *
+   * @param index - The tool call index to reset
+   */
+  resetIndex(index) {
+    this.buffers.set(index, "");
+    this.depths.set(index, 0);
+    this.inStrings.set(index, false);
+    this.escapes.set(index, false);
+    this.toolCallMeta.set(index, {});
+    this.namelessToolCallIndices.delete(index);
+    for (const [providerIndex, actualIndex] of this.pendingIndexRemaps) {
+      if (providerIndex === index || actualIndex === index) {
+        this.pendingIndexRemaps.delete(providerIndex);
+      }
+    }
+  }
+  /**
+   * Resets the entire parser state for processing a new stream
+   *
+   * Clears all accumulated buffers, parsing states, metadata, and counters.
+   * Allows the parser to be reused for multiple independent streams without
+   * data leakage between sessions.
+   */
+  reset() {
+    this.buffers.clear();
+    this.depths.clear();
+    this.inStrings.clear();
+    this.escapes.clear();
+    this.toolCallMeta.clear();
+    this.namelessToolCallIndices.clear();
+    this.idToIndexMap.clear();
+    this.pendingIndexRemaps.clear();
+    this.nextAvailableIndex = 0;
+    this.conflictingToolCallIdentity = false;
+    this.invalidToolCallIndex = false;
+  }
+  /**
+   * Gets the current accumulated buffer content for a specific index
+   *
+   * @param index - The tool call index to retrieve buffer for
+   * @returns The current buffer content for the specified index (empty string if not found)
+   */
+  getBuffer(index) {
+    return this.buffers.get(index) || "";
+  }
+  /**
+   * Gets the current parsing state information for a specific index
+   *
+   * @param index - The tool call index to get state information for
+   * @returns Object containing current parsing state (depth, inString, escape)
+   */
+  getState(index) {
+    return {
+      depth: this.depths.get(index) || 0,
+      inString: this.inStrings.get(index) || false,
+      escape: this.escapes.get(index) || false
+    };
+  }
+  /**
+   * Checks whether any buffered tool call has incomplete JSON at stream end.
+   *
+   * A tool call is considered incomplete when its JSON parsing state indicates
+   * the buffer was truncated mid-stream:
+   * - depth > 0: unclosed braces/brackets remain
+   * - inString === true: still inside a string literal
+   *
+   * This is critical for detecting output truncation that the LLM provider
+   * may not report correctly via finish_reason (e.g. reporting "stop" or
+   * "tool_calls" instead of "length" when output was actually cut off).
+   *
+   * @returns true if at least one tool call buffer has incomplete JSON
+   */
+  hasIncompleteToolCalls() {
+    for (const [index] of this.buffers.entries()) {
+      const meta = this.toolCallMeta.get(index);
+      if (!meta?.name) continue;
+      const depth = this.depths.get(index) || 0;
+      const inString = this.inStrings.get(index) || false;
+      if (depth > 0 || inString) {
+        return true;
+      }
+    }
+    return false;
+  }
+};
+
+// packages/core/src/core/openaiContentGenerator/pipeline.ts
+var debugLogger4 = createDebugLogger("OPENAI_PIPELINE");
+var StreamContentError = class extends Error {
+  static {
+    __name(this, "StreamContentError");
+  }
+  constructor(message) {
+    super(message);
+    this.name = "StreamContentError";
+  }
+};
+var StreamInactivityTimeoutError = class extends Error {
+  constructor(idleMs, chunksReceived, streamLifetimeMs) {
+    super(
+      `No stream activity for ${idleMs}ms after ${chunksReceived} chunks (stream lifetime: ${streamLifetimeMs}ms). Set ${QWEN_STREAM_IDLE_TIMEOUT_MS_ENV} to increase this window (or 0 to disable it).`
+    );
+    this.idleMs = idleMs;
+    this.chunksReceived = chunksReceived;
+    this.streamLifetimeMs = streamLifetimeMs;
+    this.name = "StreamInactivityTimeoutError";
+  }
+  static {
+    __name(this, "StreamInactivityTimeoutError");
+  }
+  code = "ETIMEDOUT";
+};
+var NON_SSE_BODY_PREFIX_LIMIT = 512;
+function isSSECompatibleContentType(contentType) {
+  if (!contentType) return true;
+  const mediaType = (contentType.split(";")[0] ?? "").trim().toLowerCase();
+  return mediaType === "text/event-stream" || mediaType === "application/x-ndjson" || mediaType === "application/stream+json";
+}
+__name(isSSECompatibleContentType, "isSSECompatibleContentType");
+var NonSSEResponseError = class extends Error {
+  constructor(contentType, httpStatus, bodyPrefix, requestId) {
+    const preview = bodyPrefix.length > 0 ? ` Body prefix: ${bodyPrefix}` : "";
+    super(
+      `Streaming request received a non-SSE response (HTTP ${httpStatus}, Content-Type: ${contentType || "unknown"}).${preview}`
+    );
+    this.contentType = contentType;
+    this.httpStatus = httpStatus;
+    this.bodyPrefix = bodyPrefix;
+    this.requestId = requestId;
+    this.name = "NonSSEResponseError";
+    this.status = httpStatus;
+    this.request_id = requestId;
+  }
+  static {
+    __name(this, "NonSSEResponseError");
+  }
+  status;
+  request_id;
+};
+var PROVIDER_OUTPUT_BUDGET_KEYS = ["max_completion_tokens", "max_new_tokens"];
+function hasProviderOutputBudgetKey(samplingParams) {
+  return PROVIDER_OUTPUT_BUDGET_KEYS.some(
+    (key) => samplingParams[key] !== void 0
+  );
+}
+__name(hasProviderOutputBudgetKey, "hasProviderOutputBudgetKey");
+function clampProviderOutputBudgetKeys(samplingParams, requestMaxTokens) {
+  if (typeof requestMaxTokens !== "number") return samplingParams;
+  for (const key of PROVIDER_OUTPUT_BUDGET_KEYS) {
+    const value = samplingParams[key];
+    if (typeof value === "number" && value > requestMaxTokens) {
+      samplingParams[key] = requestMaxTokens;
+    }
+  }
+  return samplingParams;
+}
+__name(clampProviderOutputBudgetKeys, "clampProviderOutputBudgetKeys");
+function resolveStreamIdleTimeoutMs(config) {
+  const fromConfig = config.streamIdleTimeoutMs;
+  if (typeof fromConfig === "number") {
+    if (Number.isInteger(fromConfig) && fromConfig <= MAX_STREAM_IDLE_TIMEOUT_MS) {
+      return fromConfig;
+    }
+    console.warn(
+      `[qwen-code] Ignoring out-of-range streamIdleTimeoutMs=${fromConfig} (expected an integer in (-\u221E, ${MAX_STREAM_IDLE_TIMEOUT_MS}]); falling back to ${QWEN_STREAM_IDLE_TIMEOUT_MS_ENV}/default.`
+    );
+  }
+  const raw = process.env[QWEN_STREAM_IDLE_TIMEOUT_MS_ENV];
+  const trimmed = raw?.trim();
+  if (trimmed) {
+    if (/^\d+$/.test(trimmed)) {
+      const parsed = Number(trimmed);
+      if (parsed <= MAX_STREAM_IDLE_TIMEOUT_MS) {
+        return parsed;
+      }
+    }
+    console.warn(
+      `[qwen-code] Ignoring invalid ${QWEN_STREAM_IDLE_TIMEOUT_MS_ENV}="${raw}" (expected an integer of milliseconds in [0, ${MAX_STREAM_IDLE_TIMEOUT_MS}]); using default ${DEFAULT_STREAM_IDLE_TIMEOUT_MS}ms.`
+    );
+  }
+  return DEFAULT_STREAM_IDLE_TIMEOUT_MS;
+}
+__name(resolveStreamIdleTimeoutMs, "resolveStreamIdleTimeoutMs");
+async function* withStreamInactivityTimeout(source, idleMs, abortRequest, parentSignal) {
+  const it = source[Symbol.asyncIterator]();
+  const streamStartedAt = Date.now();
+  let chunksReceived = 0;
+  try {
+    while (true) {
+      const nextPromise = it.next();
+      let timer;
+      const timeout = new Promise((_resolve, reject) => {
+        timer = setTimeout(() => {
+          if (parentSignal?.aborted) {
+            const abortErr = new Error("Aborted");
+            abortErr.name = "AbortError";
+            reject(abortErr);
+          } else {
+            abortRequest();
+            reject(
+              new StreamInactivityTimeoutError(
+                idleMs,
+                chunksReceived,
+                Date.now() - streamStartedAt
+              )
+            );
+          }
+        }, idleMs);
+        timer.unref?.();
+      });
+      let result;
+      try {
+        result = await Promise.race([nextPromise, timeout]);
+      } catch (err) {
+        void Promise.resolve(nextPromise).catch(() => {
+        });
+        throw err;
+      } finally {
+        if (timer !== void 0) clearTimeout(timer);
+      }
+      if (result.done) return;
+      chunksReceived += 1;
+      yield result.value;
+    }
+  } finally {
+    abortRequest();
+    try {
+      await it.return?.();
+    } catch {
+    }
+  }
+}
+__name(withStreamInactivityTimeout, "withStreamInactivityTimeout");
+var ContentGenerationPipeline = class {
+  constructor(config) {
+    this.config = config;
+    this.contentGeneratorConfig = config.contentGeneratorConfig;
+    this.client = this.config.provider.buildClient();
+    this.streamIdleTimeoutMs = resolveStreamIdleTimeoutMs(
+      this.contentGeneratorConfig
+    );
+  }
+  static {
+    __name(this, "ContentGenerationPipeline");
+  }
+  client;
+  contentGeneratorConfig;
+  // Resolved once (config field > env > default) so the env read + any
+  // invalid-value warning happen per pipeline, not per streaming request.
+  streamIdleTimeoutMs;
+  async execute(request, userPromptId) {
+    return this.executeWithErrorHandling(
+      request,
+      userPromptId,
+      false,
+      async (openaiRequest, context) => {
+        const parentSignal = request.config?.abortSignal;
+        const perRequestAc = parentSignal ? createChildAbortController(parentSignal) : void 0;
+        try {
+          const openaiResponse = await this.client.chat.completions.create(
+            openaiRequest,
+            {
+              signal: perRequestAc?.signal
+            }
+          );
+          const geminiResponse = OpenAIContentConverter.convertOpenAIResponseToGemini(
+            openaiResponse,
+            context
+          );
+          return geminiResponse;
+        } finally {
+          perRequestAc?.abort();
+        }
+      }
+    );
+  }
+  async executeStream(request, userPromptId) {
+    return this.executeWithErrorHandling(
+      request,
+      userPromptId,
+      true,
+      async (openaiRequest, context) => {
+        const parentSignal = request.config?.abortSignal;
+        const perRequestAc = createChildAbortController(parentSignal);
+        let stream;
+        try {
+          const createPromise = this.client.chat.completions.create(
+            openaiRequest,
+            { signal: perRequestAc.signal }
+          );
+          if (typeof createPromise.withResponse === "function") {
+            const {
+              data,
+              response: httpResponse,
+              request_id
+            } = await createPromise.withResponse();
+            stream = data;
+            const contentType = httpResponse.headers.get("content-type") ?? null;
+            if (!isSSECompatibleContentType(contentType)) {
+              let bodyPrefix = "";
+              try {
+                if (httpResponse.body) {
+                  const reader = httpResponse.body.getReader();
+                  const { value } = await reader.read();
+                  reader.releaseLock();
+                  if (value) {
+                    bodyPrefix = new TextDecoder().decode(value).slice(0, NON_SSE_BODY_PREFIX_LIMIT);
+                  }
+                }
+              } catch {
+              }
+              throw new NonSSEResponseError(
+                contentType,
+                httpResponse.status,
+                bodyPrefix,
+                request_id
+              );
+            }
+          } else {
+            stream = await createPromise;
+          }
+        } catch (e) {
+          perRequestAc.abort();
+          throw e;
+        }
+        const idleMs = this.streamIdleTimeoutMs;
+        const guarded = idleMs > 0 ? withStreamInactivityTimeout(
+          stream,
+          idleMs,
+          () => perRequestAc.abort(),
+          parentSignal
+        ) : stream;
+        const innerStream = this.processStreamWithLogging(
+          guarded,
+          context,
+          request,
+          userPromptId
+        );
+        async function* drainThenCleanup() {
+          try {
+            yield* innerStream;
+          } finally {
+            perRequestAc.abort();
+          }
+        }
+        __name(drainThenCleanup, "drainThenCleanup");
+        return drainThenCleanup();
+      }
+    );
+  }
+  /**
+   * Stage 2: Process OpenAI stream with conversion and logging
+   * This method handles the complete stream processing pipeline:
+   * 1. Convert OpenAI chunks to Gemini format while preserving original chunks
+   * 2. Filter empty responses
+   * 3. Handle chunk merging for providers that send finishReason and usageMetadata separately
+   * 4. Handle success/error logging
+   */
+  async *processStreamWithLogging(stream, context, request, userPromptId) {
+    let pendingFinishResponse = null;
+    let finishYielded = false;
+    let pendingFinishProtocolTagSanitized;
+    const logPendingProtocolTagSanitized = /* @__PURE__ */ __name((response, sanitization) => {
+      if (!sanitization) return;
+      const event = new ProtocolTagSanitizedEvent({
+        model: context.model,
+        promptId: userPromptId,
+        responseId: response.responseId,
+        tagName: sanitization.tagName,
+        toolCallCount: sanitization.toolCallCount
+      });
+      debugLogger4.warn("Sanitized a model protocol tag", {
+        model: event.model,
+        promptId: event.prompt_id,
+        responseId: event.response_id,
+        tagName: event.tag_name,
+        toolCallCount: event.tool_call_count
+      });
+      logProtocolTagSanitized(this.config.cliConfig, event);
+    }, "logPendingProtocolTagSanitized");
+    try {
+      for await (const chunk of stream) {
+        if (chunk.choices?.[0]?.finish_reason === "error_finish") {
+          const errorContent = chunk.choices?.[0]?.delta?.content?.trim() || "Unknown stream error";
+          throw new StreamContentError(errorContent);
+        }
+        const response = OpenAIContentConverter.convertOpenAIChunkToGemini(
+          chunk,
+          context
+        );
+        const sanitization = context.protocolTagSanitized;
+        if (sanitization) {
+          context.protocolTagSanitized = void 0;
+        }
+        if (response.candidates?.[0]?.content?.parts?.length === 0 && !response.candidates?.[0]?.finishReason && !response.usageMetadata && // Preparation-only responses must reach ACP before arguments complete.
+        getToolCallPreparations(response).length === 0) {
+          continue;
+        }
+        if (pendingFinishProtocolTagSanitized && pendingFinishResponse && !response.candidates?.[0]?.finishReason && response.candidates?.some(
+          (candidate) => (candidate.content?.parts?.length ?? 0) > 0
+        )) {
+          throw new InvalidStreamError(
+            "Model response continued after a finish reason.",
+            "PROTOCOL_TAG_LEAK"
+          );
+        }
+        if (finishYielded) {
+          if (response.usageMetadata) {
+            const pending = pendingFinishResponse;
+            if (pending) {
+              pending.usageMetadata = response.usageMetadata;
+            }
+          }
+          continue;
+        }
+        if (!pendingFinishResponse && response.candidates?.[0]?.finishReason && sanitization) {
+          pendingFinishProtocolTagSanitized = sanitization;
+        }
+        const shouldYield = this.handleChunkMerging(
+          response,
+          pendingFinishResponse,
+          (mergedResponse) => {
+            pendingFinishResponse = mergedResponse;
+          }
+        );
+        if (shouldYield) {
+          if (pendingFinishResponse) {
+            logPendingProtocolTagSanitized(
+              pendingFinishResponse,
+              pendingFinishProtocolTagSanitized
+            );
+            yield pendingFinishResponse;
+            finishYielded = true;
+          } else {
+            logPendingProtocolTagSanitized(response, sanitization);
+            yield response;
+          }
+        }
+      }
+      if (context.pendingThinkingTagCandidate && !context.pendingThinkingTagCandidate.closingTagName && !/\S/.test(context.pendingThinkingTagCandidate.text)) {
+        const pendingParts = context.pendingUntrustedResponseParts;
+        context.pendingThinkingTagCandidate = void 0;
+        context.pendingUntrustedResponseParts = void 0;
+        if (pendingParts?.length) {
+          const response = new GenerateContentResponse();
+          response.candidates = [
+            {
+              content: { parts: pendingParts, role: "model" },
+              index: 0
+            }
+          ];
+          yield response;
+        }
+      } else if (context.pendingThinkingTagCandidate) {
+        throw new InvalidStreamError(
+          "Model response leaked thinking tags.",
+          "PROTOCOL_TAG_LEAK"
+        );
+      }
+      if (pendingFinishResponse && !finishYielded) {
+        logPendingProtocolTagSanitized(
+          pendingFinishResponse,
+          pendingFinishProtocolTagSanitized
+        );
+        yield pendingFinishResponse;
+      }
+    } catch (error) {
+      if (error instanceof InvalidStreamError) {
+        throw error;
+      }
+      if (error instanceof StreamContentError) {
+        throw redactProxyError(error);
+      }
+      if (context.pendingThinkingTagCandidate?.closingTagName && request.config?.abortSignal?.aborted !== true) {
+        context.pendingThinkingTagCandidate = void 0;
+        context.pendingUntrustedResponseParts = void 0;
+        throw new InvalidStreamError(
+          "Model response leaked thinking tags.",
+          "PROTOCOL_TAG_LEAK"
+        );
+      }
+      if (error instanceof StreamInactivityTimeoutError) {
+        debugLogger4.warn("OpenAI stream inactivity timeout", {
+          idleMs: error.idleMs,
+          chunksReceived: error.chunksReceived,
+          streamLifetimeMs: error.streamLifetimeMs
+        });
+        throw redactProxyError(error);
+      }
+      await this.handleError(error, context, request);
+    }
+  }
+  /**
+   * Handle chunk merging for providers that send finishReason and usageMetadata separately.
+   *
+   * Strategy: When we encounter a finishReason chunk, we hold it and merge all subsequent
+   * chunks into it until the stream ends. This ensures the final chunk contains both
+   * finishReason and the most up-to-date usage information from any provider pattern.
+   *
+   * @param response Current Gemini response
+   * @param pendingFinishResponse Finish response currently held for merging
+   * @param setPendingFinish Callback to set pending finish response
+   * @returns true if the response should be yielded, false if it should be held for merging
+   */
+  handleChunkMerging(response, pendingFinishResponse, setPendingFinish) {
+    const isFinishChunk = response.candidates?.[0]?.finishReason;
+    if (isFinishChunk) {
+      if (pendingFinishResponse) {
+        if (response.usageMetadata) {
+          pendingFinishResponse.usageMetadata = response.usageMetadata;
+        }
+        setPendingFinish(pendingFinishResponse);
+      } else {
+        setPendingFinish(response);
+      }
+      return false;
+    } else if (pendingFinishResponse) {
+      const mergedResponse = new GenerateContentResponse();
+      mergedResponse.candidates = pendingFinishResponse.candidates;
+      if (response.usageMetadata) {
+        mergedResponse.usageMetadata = response.usageMetadata;
+      } else {
+        mergedResponse.usageMetadata = pendingFinishResponse.usageMetadata;
+      }
+      mergedResponse.responseId = response.responseId;
+      mergedResponse.createTime = response.createTime;
+      mergedResponse.modelVersion = response.modelVersion;
+      mergedResponse.promptFeedback = response.promptFeedback;
+      setPendingFinish(mergedResponse);
+      return true;
+    }
+    return true;
+  }
+  async buildRequest(request, userPromptId, context, isStreaming) {
+    const messages = OpenAIContentConverter.convertGeminiRequestToOpenAI(
+      request,
+      context
+    );
+    const baseRequest = {
+      model: context.model,
+      messages,
+      ...this.buildGenerateContentConfig(request)
+    };
+    if (isStreaming) {
+      baseRequest.stream = true;
+      baseRequest.stream_options = { include_usage: true };
+    } else {
+      baseRequest.stream = false;
+    }
+    if (request.config?.tools && request.config.tools.length > 0) {
+      baseRequest.tools = await OpenAIContentConverter.convertGeminiToolsToOpenAI(
+        request.config.tools,
+        this.contentGeneratorConfig.schemaCompliance ?? "auto"
+      );
+      const fcMode = request.config?.toolConfig?.functionCallingConfig?.mode;
+      if (fcMode === "ANY") {
+        baseRequest["tool_choice"] = "required";
+      } else if (fcMode === "NONE") {
+        baseRequest["tool_choice"] = "none";
+      }
+    }
+    const providerRequest = this.config.provider.buildRequest(
+      baseRequest,
+      userPromptId
+    );
+    const reasoningDisabled = request.config?.thinkingConfig?.includeThoughts === false || this.contentGeneratorConfig.reasoning === false;
+    if (reasoningDisabled) {
+      const typed = providerRequest;
+      const model = (context.model ?? "").toLowerCase();
+      if (model.startsWith("qwen") || model === "coder-model") {
+        if (DashScopeOpenAICompatibleProvider.isDashScopeProvider(
+          this.contentGeneratorConfig
+        )) {
+          typed["enable_thinking"] = false;
+        } else {
+          delete typed["enable_thinking"];
+          const existing = typed["chat_template_kwargs"] ?? {};
+          typed["chat_template_kwargs"] = {
+            ...existing,
+            enable_thinking: false
+          };
+        }
+      }
+      if ("reasoning" in typed) {
+        delete typed["reasoning"];
+      }
+      if ("reasoning_effort" in typed) {
+        delete typed["reasoning_effort"];
+      }
+      if (isDeepSeekHostname(this.contentGeneratorConfig)) {
+        typed["thinking"] = { type: "disabled" };
+      }
+    }
+    return providerRequest;
+  }
+  buildGenerateContentConfig(request) {
+    const defaultSamplingParams = this.config.provider.getDefaultGenerationConfig();
+    const configSamplingParams = this.contentGeneratorConfig.samplingParams;
+    const getParameterValue = /* @__PURE__ */ __name((configKey, requestKey) => {
+      const configValue = configSamplingParams?.[configKey];
+      const requestValue = requestKey ? request.config?.[requestKey] : void 0;
+      const defaultValue = requestKey ? defaultSamplingParams[requestKey] : void 0;
+      if (configValue !== void 0) return configValue;
+      if (requestValue !== void 0) return requestValue;
+      return defaultValue;
+    }, "getParameterValue");
+    const addParameterIfDefined = /* @__PURE__ */ __name((key, configKey, requestKey) => {
+      const value = getParameterValue(configKey, requestKey);
+      return value !== void 0 ? { [key]: value } : {};
+    }, "addParameterIfDefined");
+    if (configSamplingParams !== void 0) {
+      const requestMaxTokens = request.config?.maxOutputTokens;
+      const maxTokens = reconcileMaxTokens(configSamplingParams.max_tokens, requestMaxTokens) ?? configSamplingParams.max_tokens ?? (hasProviderOutputBudgetKey(configSamplingParams) ? void 0 : requestMaxTokens);
+      return clampProviderOutputBudgetKeys(
+        maxTokens !== void 0 ? { ...configSamplingParams, max_tokens: maxTokens } : { ...configSamplingParams },
+        requestMaxTokens
+      );
+    }
+    const params = {
+      // Parameters with request fallback but no defaults
+      ...addParameterIfDefined("temperature", "temperature", "temperature"),
+      ...addParameterIfDefined("top_p", "top_p", "topP"),
+      // Max tokens (special case: different property names)
+      ...addParameterIfDefined("max_tokens", "max_tokens", "maxOutputTokens"),
+      // Config-only parameters (no request fallback)
+      ...addParameterIfDefined("top_k", "top_k", "topK"),
+      ...addParameterIfDefined("repetition_penalty", "repetition_penalty"),
+      ...addParameterIfDefined(
+        "presence_penalty",
+        "presence_penalty",
+        "presencePenalty"
+      ),
+      ...addParameterIfDefined(
+        "frequency_penalty",
+        "frequency_penalty",
+        "frequencyPenalty"
+      ),
+      ...this.buildReasoningConfig(request)
+    };
+    return params;
+  }
+  buildReasoningConfig(request) {
+    if (request.config?.thinkingConfig?.includeThoughts === false) {
+      return {};
+    }
+    const reasoning = this.contentGeneratorConfig.reasoning;
+    if (reasoning === false || reasoning === void 0) {
+      return {};
+    }
+    return { reasoning };
+  }
+  /**
+   * Common error handling wrapper for execute methods
+   */
+  async executeWithErrorHandling(request, userPromptId, isStreaming, executor) {
+    const context = this.createRequestContext(request, isStreaming);
+    try {
+      const openaiRequest = await this.buildRequest(
+        request,
+        userPromptId,
+        context,
+        isStreaming
+      );
+      openaiRequestCaptureContext.getStore()?.(openaiRequest);
+      runtimeDiagnostics.recordOpenAIWireRequest(openaiRequest);
+      const result = await executor(openaiRequest, context);
+      return result;
+    } catch (error) {
+      return await this.handleError(error, context, request);
+    }
+  }
+  /**
+   * Shared error handling logic for both executeWithErrorHandling and processStreamWithLogging
+   * This centralizes the common error processing steps to avoid duplication
+   */
+  async handleError(error, context, request) {
+    this.config.errorHandler.handle(redactProxyError(error), context, request);
+  }
+  /**
+   * Create request context with common properties
+   */
+  createRequestContext(request, isStreaming) {
+    const effectiveModel = request.model || this.contentGeneratorConfig.model;
+    const providerOverrides = this.config.provider.getRequestContextOverrides?.() ?? {};
+    const toolCallParser = isStreaming ? new StreamingToolCallParser() : void 0;
+    const responseParsingOptions = this.config.provider.getResponseParsingOptions?.();
+    const taggedThinkingParser = isStreaming && responseParsingOptions?.taggedThinkingTags ? new TaggedThinkingParser() : void 0;
+    return {
+      model: effectiveModel,
+      modalities: this.contentGeneratorConfig.modalities ?? {},
+      startTime: Date.now(),
+      splitToolMedia: providerOverrides.splitToolMedia ?? this.contentGeneratorConfig.splitToolMedia ?? // Default true: the OpenAI Chat Completions spec only permits text on
+      // `role: "tool"` messages, so tool-returned media (e.g. an image read
+      // by read_file) embedded there is silently dropped or rejected by
+      // strict providers (doubao / new-api / LM Studio) and the model never
+      // sees it (QwenLM/qwen-code#4876). Splitting it into a follow-up user
+      // message is spec-compliant and safe for permissive providers too.
+      // Opt out via generationConfig.splitToolMedia = false.
+      true,
+      toolResultContentFormat: providerOverrides.toolResultContentFormat ?? this.contentGeneratorConfig.toolResultContentFormat ?? "parts",
+      ...toolCallParser ? { toolCallParser } : {},
+      ...responseParsingOptions ? { responseParsingOptions } : {},
+      ...taggedThinkingParser ? { taggedThinkingParser } : {}
+    };
+  }
+};
+
+// packages/core/src/core/openaiContentGenerator/errorHandler.ts
+init_esbuild_shims();
+var debugLogger5 = createDebugLogger("OPENAI_ERROR");
+var EnhancedErrorHandler = class {
+  constructor(shouldSuppressLogging = () => false) {
+    this.shouldSuppressLogging = shouldSuppressLogging;
+  }
+  static {
+    __name(this, "EnhancedErrorHandler");
+  }
+  handle(error, context, request) {
+    const redactedError = redactProxyError(error);
+    const isTimeoutError = this.isTimeoutError(redactedError);
+    const errorMessage = this.buildErrorMessage(
+      redactedError,
+      context,
+      isTimeoutError
+    );
+    if (!this.shouldSuppressErrorLogging(redactedError, request)) {
+      debugLogger5.error(
+        "OpenAI API Error:",
+        errorMessage,
+        this.buildDiagnostics(redactedError, context)
+      );
+    }
+    if (isTimeoutError) {
+      throw new Error(
+        `${errorMessage}
+
+${this.getTimeoutTroubleshootingTips()}`
+      );
+    }
+    throw redactedError;
+  }
+  shouldSuppressErrorLogging(error, request) {
+    return this.shouldSuppressLogging(error, request);
+  }
+  isTimeoutError(error) {
+    if (!error) return false;
+    const errorMessage = error instanceof Error ? error.message.toLowerCase() : String(error).toLowerCase();
+    const errorCode = error?.code;
+    const errorType = error?.type;
+    return errorMessage.includes("timeout") || errorMessage.includes("timed out") || errorMessage.includes("connection timeout") || errorMessage.includes("request timeout") || errorMessage.includes("read timeout") || errorMessage.includes("etimedout") || errorMessage.includes("esockettimedout") || errorCode === "ETIMEDOUT" || errorCode === "ESOCKETTIMEDOUT" || errorType === "timeout" || errorMessage.includes("request timed out") || errorMessage.includes("deadline exceeded");
+  }
+  buildErrorMessage(error, context, isTimeoutError) {
+    const durationSeconds = Math.round((Date.now() - context.startTime) / 1e3);
+    if (isTimeoutError) {
+      return `Request timeout after ${durationSeconds}s. Try reducing input length or increasing timeout in config.`;
+    }
+    return error instanceof Error ? getErrorMessage(error) : String(error);
+  }
+  buildDiagnostics(error, context) {
+    const details = getRateLimitErrorDetails(error);
+    const requestId = this.getRequestId(error) ?? details.requestId;
+    const statusCode = getErrorStatus(error);
+    return {
+      model: context.model,
+      durationMs: Date.now() - context.startTime,
+      errorType: getErrorType(error),
+      ...statusCode !== void 0 ? { statusCode } : {},
+      ...details.providerCode !== void 0 ? { providerCode: details.providerCode } : {},
+      ...details.providerMessage !== void 0 ? { providerMessage: details.providerMessage } : {},
+      ...requestId !== void 0 ? { requestId } : {},
+      ...details.transport !== "unknown" ? { transport: details.transport } : {}
+    };
+  }
+  getRequestId(error) {
+    if (!error || typeof error !== "object") return void 0;
+    const source = error;
+    for (const value of [
+      source.requestID,
+      source.request_id,
+      source.response_id
+    ]) {
+      if (typeof value === "string" && value) {
+        return value;
+      }
+    }
+    return void 0;
+  }
+  getTimeoutTroubleshootingTips() {
+    const tips = [
+      "- Reduce input length or complexity",
+      "- Increase timeout in config: contentGenerator.timeout",
+      "- Check network connectivity"
+    ];
+    return `Troubleshooting tips:
+${tips.join("\n")}`;
+  }
+};
+
+// packages/core/src/core/openaiContentGenerator/openaiContentGenerator.ts
+var debugLogger6 = createDebugLogger("OPENAI");
+var OpenAIContentGenerator = class {
+  static {
+    __name(this, "OpenAIContentGenerator");
+  }
+  pipeline;
+  constructor(contentGeneratorConfig, cliConfig, provider) {
+    const pipelineConfig = {
+      cliConfig,
+      provider,
+      contentGeneratorConfig,
+      errorHandler: new EnhancedErrorHandler(
+        (error, request) => this.shouldSuppressErrorLogging(error, request)
+      )
+    };
+    this.pipeline = new ContentGenerationPipeline(pipelineConfig);
+  }
+  /**
+   * Hook for subclasses to customize error handling behavior
+   * @param error The error that occurred
+   * @param request The original request
+   * @returns true if error logging should be suppressed, false otherwise
+   */
+  shouldSuppressErrorLogging(error, request) {
+    if (isAbortError(error) && request.config?.abortSignal?.aborted) {
+      return true;
+    }
+    return false;
+  }
+  async generateContent(request, userPromptId) {
+    return this.pipeline.execute(request, userPromptId);
+  }
+  async generateContentStream(request, userPromptId) {
+    return this.pipeline.executeStream(request, userPromptId);
+  }
+  async countTokens(request) {
+    try {
+      const estimator = new RequestTokenizer();
+      const result = await estimator.calculateTokens(request);
+      return {
+        totalTokens: result.totalTokens
+      };
+    } catch (error) {
+      debugLogger6.warn(
+        "Failed to calculate tokens with new tokenizer, falling back to simple method:",
+        error
+      );
+      const content = JSON.stringify(request.contents);
+      const totalTokens = Math.ceil(content.length / 4);
+      return {
+        totalTokens
+      };
+    }
+  }
+  async embedContent(request) {
+    let text = "";
+    if (Array.isArray(request.contents)) {
+      text = request.contents.map((content) => {
+        if (typeof content === "string") return content;
+        if ("parts" in content && content.parts) {
+          return content.parts.map(
+            (part) => typeof part === "string" ? part : "text" in part ? part.text || "" : ""
+          ).join(" ");
+        }
+        return "";
+      }).join(" ");
+    } else if (request.contents) {
+      if (typeof request.contents === "string") {
+        text = request.contents;
+      } else if ("parts" in request.contents && request.contents.parts) {
+        text = request.contents.parts.map(
+          (part) => typeof part === "string" ? part : "text" in part ? part.text : ""
+        ).join(" ");
+      }
+    }
+    try {
+      const embedding = await this.pipeline.client.embeddings.create({
+        model: "text-embedding-ada-002",
+        // Default embedding model
+        input: text
+      });
+      return {
+        embeddings: [
+          {
+            values: embedding.data[0].embedding
+          }
+        ]
+      };
+    } catch (error) {
+      const redactedError = redactProxyError(error);
+      debugLogger6.error("OpenAI API Embedding Error:", redactedError);
+      throw new Error(
+        `OpenAI API error: ${redactedError instanceof Error ? redactedError.message : String(redactedError)}`
+      );
+    }
+  }
+  useSummarizedThinking() {
+    return false;
+  }
+};
+
+// packages/core/src/core/openaiContentGenerator/index.ts
+function createOpenAIContentGenerator(contentGeneratorConfig, cliConfig) {
+  const provider = determineProvider(contentGeneratorConfig, cliConfig);
+  return new OpenAIContentGenerator(
+    contentGeneratorConfig,
+    cliConfig,
+    provider
+  );
+}
+__name(createOpenAIContentGenerator, "createOpenAIContentGenerator");
+function determineProvider(contentGeneratorConfig, cliConfig) {
+  const config = contentGeneratorConfig || cliConfig.getContentGeneratorConfig();
+  if (DashScopeOpenAICompatibleProvider.isDashScopeProvider(config)) {
+    return new DashScopeOpenAICompatibleProvider(
+      contentGeneratorConfig,
+      cliConfig
+    );
+  }
+  if (DeepSeekOpenAICompatibleProvider.isDeepSeekProvider(config)) {
+    return new DeepSeekOpenAICompatibleProvider(
+      contentGeneratorConfig,
+      cliConfig
+    );
+  }
+  if (ZaiOpenAICompatibleProvider.isZaiProvider(config)) {
+    return new ZaiOpenAICompatibleProvider(contentGeneratorConfig, cliConfig);
+  }
+  if (MiMoOpenAICompatibleProvider.isMiMoProvider(config)) {
+    return new MiMoOpenAICompatibleProvider(contentGeneratorConfig, cliConfig);
+  }
+  if (ModelScopeOpenAICompatibleProvider.isModelScopeProvider(config)) {
+    return new ModelScopeOpenAICompatibleProvider(
+      contentGeneratorConfig,
+      cliConfig
+    );
+  }
+  if (MiniMaxOpenAICompatibleProvider.isMiniMaxProvider(config)) {
+    return new MiniMaxOpenAICompatibleProvider(
+      contentGeneratorConfig,
+      cliConfig
+    );
+  }
+  if (MistralOpenAICompatibleProvider.isMistralProvider(config)) {
+    return new MistralOpenAICompatibleProvider(
+      contentGeneratorConfig,
+      cliConfig
+    );
+  }
+  return new DefaultOpenAICompatibleProvider(contentGeneratorConfig, cliConfig);
+}
+__name(determineProvider, "determineProvider");
+
 export {
-  OpenAI,
-  DefaultOpenAICompatibleProvider,
-  DASHSCOPE_REGIONAL_HOSTS,
-  DashScopeOpenAICompatibleProvider
+  DashScopeOpenAICompatibleProvider,
+  DeepSeekOpenAICompatibleProvider,
+  MiniMaxOpenAICompatibleProvider,
+  MistralOpenAICompatibleProvider,
+  MiMoOpenAICompatibleProvider,
+  ContentGenerationPipeline,
+  EnhancedErrorHandler,
+  OpenAIContentGenerator,
+  createOpenAIContentGenerator,
+  determineProvider
 };
+/**
+ * @license
+ * Copyright 2025 Qwen
+ * SPDX-License-Identifier: Apache-2.0
+ */