feat: Grok 适配完善 — 防御性 usage 合并 + thinking 自动检测 (#1234)

* feat: Grok 适配完善 — 防御性 usage 合并 + thinking 自动检测

1. 提取 updateOpenAIUsage 到共享模块 openaiShared.ts，供 OpenAI 和
   Grok 两条路径复用，消除 Grok 中重复的 spread 漏洞。

2. 在 requestBody.ts 的 isOpenAIThinkingEnabled() 中增加 Grok 模型
   自动检测（模型名含 "grok"），与 DeepSeek/MiMo 并列。

3. messaging 层的 reasoning_content 回传（openaiConvertMessages.ts）
   和流解析（openaiStreamAdapter.ts）无需修改，Grok 与 DeepSeek/MiMo
   共用相同的 reasoning_content 字段协议。

Co-Authored-By: deepseek-v4-pro[1m] <deepseek-ai@claude-code-best.win>

* fix: 回退 Grok 从 isOpenAIThinkingEnabled 的自动检测

Grok 推理模型（如 grok-4.20-reasoning）自动进行推理，不需要
thinking/enable_thinking 请求参数。发送这些参数虽大概率被忽略
（OpenAI SDK 透传 unknown keys），但属于不正确行为。

Co-Authored-By: deepseek-v4-pro[1m] <deepseek-ai@claude-code-best.win>

---------

Co-authored-by: deepseek-v4-pro[1m] <deepseek-ai@claude-code-best.win>

src/services/api/grok/index.ts

@@ -12,6 +12,7 @@ import type {
   ChatCompletionCreateParamsStreaming,
 } from 'openai/resources/chat/completions/completions.mjs'
 import { getGrokClient } from './client.js'
+import { updateOpenAIUsage } from '../openai/openaiShared.js'
 import {
   anthropicMessagesToOpenAI,
   anthropicToolsToOpenAI,
@@ -136,7 +137,7 @@ export async function* queryModelGrok(
           partialMessage = (event as any).message
           ttftMs = Date.now() - start
           if ((event as any).message?.usage) {
-            usage = { ...usage, ...(event as any).message.usage }
+            usage = updateOpenAIUsage(usage, (event as any).message.usage)
           }
           break
         }
@@ -192,7 +193,7 @@ export async function* queryModelGrok(
         case 'message_delta': {
           const deltaUsage = (event as any).usage
           if (deltaUsage) {
-            usage = { ...usage, ...deltaUsage }
+            usage = updateOpenAIUsage(usage, deltaUsage)
           }
           break
         }

src/services/api/openai/index.ts

@@ -10,6 +10,7 @@ import type {
 import type { AgentId } from '../../../types/ids.js'
 import type { Tools } from '../../../Tool.js'
 import { getOpenAIClient } from './client.js'
+import { updateOpenAIUsage } from './openaiShared.js'
 import {
   anthropicMessagesToOpenAI,
   resolveOpenAIModel,
@@ -449,7 +450,7 @@ export async function* queryModelOpenAI(
         case 'message_delta': {
           const deltaUsage = (event as any).usage
           if (deltaUsage) {
-            usage = { ...usage, ...deltaUsage }
+            usage = updateOpenAIUsage(usage, deltaUsage)
           }
           if ((event as any).delta?.stop_reason != null) {
             stopReason = (event as any).delta.stop_reason

src/services/api/openai/openaiShared.ts

@@ -0,0 +1,46 @@
+/**
+ * Shared utilities for OpenAI-compatible API paths.
+ *
+ * Both the OpenAI path (queryModelOpenAI) and Grok path (queryModelGrok) use
+ * the same adapters (openaiStreamAdapter, openaiConvertMessages), so the event
+ * processing logic should be shared rather than duplicated.
+ */
+
+/**
+ * Merge a delta usage into the accumulated usage, preserving cache-related
+ * fields from previous values when the delta carries explicit zeroes or
+ * undefined values.
+ *
+ * Mirrors updateUsage() in claude.ts: a future adapter change that omits
+ * cache fields from certain streaming events should not silently zero the
+ * accumulated counters.
+ */
+export function updateOpenAIUsage(
+  current: {
+    input_tokens: number
+    output_tokens: number
+    cache_creation_input_tokens: number
+    cache_read_input_tokens: number
+  },
+  delta: {
+    input_tokens?: number
+    output_tokens?: number
+    cache_creation_input_tokens?: number
+    cache_read_input_tokens?: number
+  },
+): typeof current {
+  return {
+    input_tokens: delta.input_tokens ?? current.input_tokens,
+    output_tokens: delta.output_tokens ?? current.output_tokens,
+    cache_creation_input_tokens:
+      delta.cache_creation_input_tokens !== undefined &&
+      delta.cache_creation_input_tokens > 0
+        ? delta.cache_creation_input_tokens
+        : current.cache_creation_input_tokens,
+    cache_read_input_tokens:
+      delta.cache_read_input_tokens !== undefined &&
+      delta.cache_read_input_tokens > 0
+        ? delta.cache_read_input_tokens
+        : current.cache_read_input_tokens,
+  }
+}

src/services/api/openai/requestBody.ts

@@ -23,7 +23,9 @@ export function isOpenAIThinkingEnabled(model: string): boolean {
   if (isEnvDefinedFalsy(process.env.OPENAI_ENABLE_THINKING)) return false
   // Explicit enable
   if (isEnvTruthy(process.env.OPENAI_ENABLE_THINKING)) return true
-  // Auto-detect from model name (DeepSeek and MiMo models support thinking mode)
+  // Auto-detect from model name (DeepSeek and MiMo models support thinking mode).
+  // Grok is intentionally excluded — Grok reasoning models reason automatically
+  // and do NOT require thinking/enable_thinking request body parameters.
   const modelLower = model.toLowerCase()
   return modelLower.includes('deepseek') || modelLower.includes('mimo')
 }