fix: 修复 CoStrict provider 提前停止与 token override 问题

src/costrict/provider/index.test.ts

@@ -0,0 +1,275 @@
+import { afterEach, beforeEach, describe, expect, mock, test } from 'bun:test'
+import type { BetaRawMessageStreamEvent } from '@anthropic-ai/sdk/resources/beta/messages/messages.mjs'
+import type {
+  AssistantMessage,
+  StreamEvent,
+} from '../../types/message.js'
+
+function makeMessageStart(
+  overrides: Record<string, any> = {},
+): BetaRawMessageStreamEvent {
+  return {
+    type: 'message_start',
+    message: {
+      id: 'msg_test',
+      type: 'message',
+      role: 'assistant',
+      content: [],
+      model: 'test-model',
+      stop_reason: null,
+      stop_sequence: null,
+      usage: {
+        input_tokens: 0,
+        output_tokens: 0,
+        cache_creation_input_tokens: 0,
+        cache_read_input_tokens: 0,
+      },
+      ...overrides,
+    },
+  } as any
+}
+
+function makeContentBlockStart(
+  index: number,
+  type: 'text' | 'thinking',
+): BetaRawMessageStreamEvent {
+  return {
+    type: 'content_block_start',
+    index,
+    content_block:
+      type === 'text'
+        ? { type: 'text', text: '' }
+        : { type: 'thinking', thinking: '', signature: '' },
+  } as any
+}
+
+function makeTextDelta(index: number, text: string): BetaRawMessageStreamEvent {
+  return {
+    type: 'content_block_delta',
+    index,
+    delta: { type: 'text_delta', text },
+  } as any
+}
+
+function makeThinkingDelta(
+  index: number,
+  thinking: string,
+): BetaRawMessageStreamEvent {
+  return {
+    type: 'content_block_delta',
+    index,
+    delta: { type: 'thinking_delta', thinking },
+  } as any
+}
+
+function makeContentBlockStop(index: number): BetaRawMessageStreamEvent {
+  return { type: 'content_block_stop', index } as any
+}
+
+function makeMessageDelta(
+  stopReason: string,
+  outputTokens: number,
+): BetaRawMessageStreamEvent {
+  return {
+    type: 'message_delta',
+    delta: { stop_reason: stopReason, stop_sequence: null },
+    usage: { output_tokens: outputTokens },
+  } as any
+}
+
+function makeMessageStop(): BetaRawMessageStreamEvent {
+  return { type: 'message_stop' } as any
+}
+
+async function* eventStream(events: BetaRawMessageStreamEvent[]) {
+  for (const event of events) yield event
+}
+
+let _nextEvents: BetaRawMessageStreamEvent[] = []
+let _lastCreateArgs: Record<string, any> | null = null
+let _mockModelMaxTokens: number | undefined
+
+mock.module('openai', () => ({
+  default: class OpenAI {
+    chat = {
+      completions: {
+        create: async (args: Record<string, any>) => {
+          _lastCreateArgs = args
+          return { [Symbol.asyncIterator]: async function* () {} }
+        },
+      },
+    }
+  },
+}))
+
+mock.module('@ant/model-provider', () => ({
+  anthropicMessagesToOpenAI: () => [],
+  anthropicToolsToOpenAI: () => [],
+  anthropicToolChoiceToOpenAI: () => undefined,
+  adaptOpenAIStreamToAnthropic: () => eventStream(_nextEvents),
+}))
+
+mock.module('../../utils/messages.js', () => ({
+  normalizeMessagesForAPI: (msgs: any) => msgs,
+  normalizeContentFromAPI: (blocks: any[]) => blocks,
+  createAssistantAPIErrorMessage: (opts: any) => ({
+    type: 'assistant',
+    message: {
+      content: [{ type: 'text', text: opts.content }],
+      apiError: opts.apiError,
+    },
+    uuid: 'error-uuid',
+    timestamp: new Date().toISOString(),
+  }),
+}))
+
+mock.module('../../utils/api.js', () => ({
+  toolToAPISchema: async (tool: any) => tool,
+}))
+
+mock.module('../../utils/debug.js', () => ({
+  logForDebugging: () => {},
+}))
+
+mock.module('../../cost-tracker.js', () => ({
+  addToTotalSessionCost: () => {},
+}))
+
+mock.module('../../utils/modelCost.js', () => ({
+  calculateUSDCost: () => 0,
+}))
+
+mock.module('../../utils/proxy.js', () => ({
+  getProxyFetchOptions: () => ({}),
+}))
+
+mock.module('../../utils/http.js', () => ({
+  getUserAgent: () => 'test-agent',
+}))
+
+mock.module('./fetch.js', () => ({
+  createCoStrictFetch: () => fetch,
+}))
+
+mock.module('./modelMapping.js', () => ({
+  resolveCoStrictModel: (model: string) => model,
+}))
+
+mock.module('./auth.js', () => ({
+  getCoStrictBaseURL: () => 'https://example.test',
+}))
+
+mock.module('./credentials.js', () => ({
+  loadCoStrictCredentials: async () => ({
+    access_token: 'token',
+    base_url: 'https://example.test',
+  }),
+}))
+
+mock.module('./models.js', () => ({
+  fetchCoStrictModels: async () => [
+    {
+      id: 'test-model',
+      maxTokens: _mockModelMaxTokens,
+      maxTokensKey: 'max_completion_tokens',
+    },
+  ],
+}))
+
+mock.module('../../services/api/openai/requestBody.js', () => ({
+  isOpenAIThinkingEnabled: (model: string) => model.includes('deepseek'),
+  resolveOpenAIMaxTokens: (
+    upperLimit: number,
+    maxOutputTokensOverride?: number,
+  ) => maxOutputTokensOverride ?? upperLimit,
+}))
+
+mock.module('../../bootstrap/state.js', () => ({
+  getMainThreadAgentType: () => null,
+  getActiveSkillName: () => null,
+}))
+
+mock.module('../../utils/context.js', () => ({
+  getModelMaxOutputTokens: () => ({ upperLimit: 8192, default: 8192 }),
+}))
+
+async function runQueryModel(
+  events: BetaRawMessageStreamEvent[],
+  optionsOverrides: Record<string, unknown> = {},
+) {
+  _nextEvents = events
+  const { queryModelCoStrict } = await import('./index.js')
+  const assistantMessages: AssistantMessage[] = []
+  const streamEvents: StreamEvent[] = []
+
+  const options: any = {
+    model: 'test-model',
+    tools: [],
+    agents: [],
+    querySource: 'main_loop',
+    ...optionsOverrides,
+  }
+
+  for await (const item of queryModelCoStrict(
+    [],
+    { type: 'text', text: '' } as any,
+    [],
+    new AbortController().signal,
+    options,
+  )) {
+    if (item.type === 'assistant') {
+      assistantMessages.push(item as AssistantMessage)
+    } else if (item.type === 'stream_event') {
+      streamEvents.push(item as StreamEvent)
+    }
+  }
+
+  return { assistantMessages, streamEvents }
+}
+
+beforeEach(() => {
+  _nextEvents = []
+  _lastCreateArgs = null
+  _mockModelMaxTokens = undefined
+})
+
+afterEach(() => {
+  _nextEvents = []
+  _mockModelMaxTokens = undefined
+})
+
+describe('queryModelCoStrict', () => {
+  test('yields exactly one AssistantMessage for thinking + text content', async () => {
+    const events = [
+      makeMessageStart(),
+      makeContentBlockStart(0, 'thinking'),
+      makeThinkingDelta(0, 'let me think'),
+      makeContentBlockStop(0),
+      makeContentBlockStart(1, 'text'),
+      makeTextDelta(1, 'answer'),
+      makeContentBlockStop(1),
+      makeMessageDelta('end_turn', 12),
+      makeMessageStop(),
+    ]
+
+    const { assistantMessages } = await runQueryModel(events)
+
+    expect(assistantMessages).toHaveLength(1)
+    expect(assistantMessages[0]!.message.stop_reason).toBe('end_turn')
+    expect(
+      (assistantMessages[0]!.message.content as any[]).map(
+        block => block.type,
+      ),
+    ).toEqual(['thinking', 'text'])
+  })
+
+  test('preserves explicit max token override over model metadata default', async () => {
+    _mockModelMaxTokens = 16384
+    const events = [makeMessageStart(), makeMessageStop()]
+
+    await runQueryModel(events, { maxOutputTokensOverride: 2048 })
+
+    expect(_lastCreateArgs).not.toBeNull()
+    expect(_lastCreateArgs!.max_completion_tokens).toBe(2048)
+  })
+})

src/costrict/provider/index.ts

@@ -17,7 +17,12 @@ import type { Options } from '../../services/api/claude.js'
 import OpenAI from 'openai'
 import { getProxyFetchOptions } from '../../utils/proxy.js'
 import { getUserAgent } from '../../utils/http.js'
-import { anthropicMessagesToOpenAI, anthropicToolsToOpenAI, anthropicToolChoiceToOpenAI, adaptOpenAIStreamToAnthropic } from '@ant/model-provider'
+import {
+  anthropicMessagesToOpenAI,
+  anthropicToolsToOpenAI,
+  anthropicToolChoiceToOpenAI,
+  adaptOpenAIStreamToAnthropic,
+} from '@ant/model-provider'
 import { normalizeMessagesForAPI } from '../../utils/messages.js'
 import { toolToAPISchema } from '../../utils/api.js'
 import { logForDebugging } from '../../utils/debug.js'
@@ -32,9 +37,16 @@ import { createCoStrictFetch } from './fetch.js'
 import { resolveCoStrictModel } from './modelMapping.js'
 import { getCoStrictBaseURL } from './auth.js'
 import { loadCoStrictCredentials } from './credentials.js'
-import { isOpenAIThinkingEnabled } from '../../services/api/openai/requestBody.js'
+import {
+  isOpenAIThinkingEnabled,
+  resolveOpenAIMaxTokens,
+} from '../../services/api/openai/requestBody.js'
 import { fetchCoStrictModels } from './models.js'
-import { getMainThreadAgentType, getActiveSkillName } from '../../bootstrap/state.js'
+import {
+  getMainThreadAgentType,
+  getActiveSkillName,
+} from '../../bootstrap/state.js'
+import { getModelMaxOutputTokens } from '../../utils/context.js'
 
 /**
  * CoStrict 查询路径
@@ -60,8 +72,8 @@ export async function* queryModelCoStrict(
     const chatBaseURL = `${baseUrl}/chat-rag/api/v1`
 
     // 3. 从模型列表获取 maxTokens 相关参数
+    let defaultMaxTokens = getModelMaxOutputTokens(costrictModel).upperLimit
     let maxTokensParamKey: string = 'max_tokens'
-    let maxTokensValue: number | undefined
     if (creds?.access_token) {
       try {
         const modelList = await fetchCoStrictModels(baseUrl, creds.access_token)
@@ -69,13 +81,17 @@ export async function* queryModelCoStrict(
         if (modelInfo) {
           maxTokensParamKey = modelInfo.maxTokensKey || 'max_tokens'
           if (modelInfo.maxTokens != null) {
-            maxTokensValue = modelInfo.maxTokens
+            defaultMaxTokens = modelInfo.maxTokens
           }
         }
       } catch {
         // 获取模型列表失败，使用默认值
       }
     }
+    const maxTokensValue = resolveOpenAIMaxTokens(
+      defaultMaxTokens,
+      options.maxOutputTokensOverride,
+    )
 
     // 4. 规范化消息
     const messagesForAPI = normalizeMessagesForAPI(messages, tools)
@@ -107,7 +123,7 @@ export async function* queryModelCoStrict(
     const openaiMessages = anthropicMessagesToOpenAI(
       messagesForAPI,
       systemPrompt,
-      { enableThinking }
+      { enableThinking },
     )
     const openaiTools = anthropicToolsToOpenAI(standardTools)
     const openaiToolChoice = anthropicToolChoiceToOpenAI(options.toolChoice)
@@ -146,12 +162,16 @@ export async function* queryModelCoStrict(
       }),
       stream: true,
       stream_options: { include_usage: true },
-      ...(options.temperatureOverride !== undefined && {
-        temperature: options.temperatureOverride,
-      }),
-      ...(maxTokensValue != null && {
-        [maxTokensParamKey]: maxTokensValue,
+      [maxTokensParamKey]: maxTokensValue,
+      ...(enableThinking && {
+        thinking: { type: 'enabled' },
+        enable_thinking: true,
+        chat_template_kwargs: { thinking: true },
       }),
+      ...(!enableThinking &&
+        options.temperatureOverride !== undefined && {
+          temperature: options.temperatureOverride,
+        }),
     }
     const stream = await client.chat.completions.create(
       requestBody as unknown as OpenAI.Chat.Completions.ChatCompletionCreateParamsStreaming,
@@ -162,8 +182,6 @@ export async function* queryModelCoStrict(
     const adaptedStream = adaptOpenAIStreamToAnthropic(stream, costrictModel)
 
     const contentBlocks: Record<number, any> = {}
-    // 跟踪已 yield 的 assistant messages，用于 message_delta 时回写 usage
-    const yieldedMessages: AssistantMessage[] = []
     let partialMessage: any
     let stopReason: string | null = null
     let usage = {
@@ -174,6 +192,48 @@ export async function* queryModelCoStrict(
     }
     let ttftMs = 0
     const start = Date.now()
+    const assembleFinalAssistantOutputs = (): (
+      | AssistantMessage
+      | SystemAPIErrorMessage
+    )[] => {
+      const outputs: (AssistantMessage | SystemAPIErrorMessage)[] = []
+      if (!partialMessage) return outputs
+
+      const allBlocks = Object.keys(contentBlocks)
+        .sort((a, b) => Number(a) - Number(b))
+        .map(k => contentBlocks[Number(k)])
+        .filter(Boolean)
+
+      if (allBlocks.length > 0) {
+        outputs.push({
+          message: {
+            ...partialMessage,
+            content: normalizeContentFromAPI(allBlocks, tools, options.agentId),
+            usage,
+            stop_reason: stopReason,
+            stop_sequence: null,
+          },
+          requestId: undefined,
+          type: 'assistant',
+          uuid: randomUUID(),
+          timestamp: new Date().toISOString(),
+        } as AssistantMessage)
+      }
+
+      if (stopReason === 'max_tokens') {
+        outputs.push(
+          createAssistantAPIErrorMessage({
+            content:
+              `Output truncated: response exceeded the ${maxTokensValue} token limit. ` +
+              `Set OPENAI_MAX_TOKENS or CLAUDE_CODE_MAX_OUTPUT_TOKENS to override.`,
+            apiError: 'max_output_tokens',
+            error: 'max_output_tokens',
+          }),
+        )
+      }
+
+      return outputs
+    }
 
     for await (const event of adaptedStream) {
       switch (event.type) {
@@ -216,44 +276,27 @@ export async function* queryModelCoStrict(
           break
         }
         case 'content_block_stop': {
-          const idx = (event as any).index
-          const block = contentBlocks[idx]
-          if (!block || !partialMessage) break
-          const m: AssistantMessage = {
-            message: {
-              ...partialMessage,
-              content: normalizeContentFromAPI([block], tools, options.agentId),
-              usage,
-            },
-            requestId: undefined,
-            type: 'assistant',
-            uuid: randomUUID(),
-            timestamp: new Date().toISOString(),
-          }
-          yieldedMessages.push(m)
-          yield m
+          // Block accumulation is complete; emit one AssistantMessage at
+          // message_stop so reasoning/text/tool blocks stay in a single turn.
           break
         }
         case 'message_delta': {
           const deltaUsage = (event as any).usage
           if (deltaUsage) usage = { ...usage, ...deltaUsage }
-          // 回写 usage 到已 yield 的 assistant messages
-          // 与 Anthropic 原生路径 claude.ts:2298 保持一致
-          for (const msg of yieldedMessages) {
-            msg.message.usage = usage
-          }
-          // 记录 stop_reason，回写到最后的 message
           if ((event as any).delta?.stop_reason != null) {
             stopReason = (event as any).delta.stop_reason
-            const lastMsg = yieldedMessages[yieldedMessages.length - 1]
-            if (lastMsg) {
-              lastMsg.message.stop_reason = stopReason
-            }
           }
           break
         }
-        case 'message_stop':
+        case 'message_stop': {
+          if (partialMessage) {
+            for (const output of assembleFinalAssistantOutputs()) {
+              yield output
+            }
+            partialMessage = null
+          }
           break
+        }
       }
 
       if (
@@ -270,6 +313,12 @@ export async function* queryModelCoStrict(
         ...(event.type === 'message_start' ? { ttftMs } : undefined),
       } as StreamEvent
     }
+
+    if (partialMessage) {
+      for (const output of assembleFinalAssistantOutputs()) {
+        yield output
+      }
+    }
   } catch (error) {
     const errorMsg = error instanceof Error ? error.message : String(error)
     logForDebugging(`[CoStrict] Error: ${errorMsg}`, { level: 'error' })