fix: v0.5.3 模型能力完整性修复 — MCP 工具动态同步 + maxTokens 模型上限钳制

背景：v0.5.2 工具调用失效修复后，对模型全能力矩阵（工具/思考/多模态/流式/ 压缩/摘要/记忆/故障转移/余额）做契约级核查，发现并修复两处同类时序/边界缺陷。 P1 MCP 工具运行中增删不同步引擎: - 症状：运行中添加/启用 MCP server 后，已打开会话拿不到新工具；断开 server 后已有引擎仍持有失效工具定义（模型发起调用才报 Unknown tool） - 根因：setToolsAll 仅在启动期与工具开关时调用，MCP 连接/断开路径缺失（与 v0.5.2 修复的懒创建陷阱同类——"变更点 × 同步路径"未全覆盖） - 修复：MCPManager 新增 setOnToolsChanged 回调，connectServer 注册完成 / disconnectServer 注销完成后触发；main.ts 注入回调同步全部已存在引擎。懒创建引擎由 createEngine 实时拉取（v0.5.2），三条路径（启动/懒创建/ 运行中变更）全覆盖。README"无需重启动态发现"的宣称至此真实成立。 P1 maxTokens 超模型上限直接 400: - 症状：引擎默认 maxTokens=63488，OpenAI gpt-4o(16384)/gpt-4.1(32768)、 Anthropic opus/haiku(32000)、MiMo standard(32768) 每次请求 400，等于不可用 - 修复：五个 adapter（DeepSeek/Agnes/MiMo/OpenAI/Anthropic）统一按 MODEL_INFO.maxOutputTokens 钳制；MiMo 保留 thinking 兜底 32768 语义； Anthropic thinking budget 在钳制后的 max_tokens 内二分，自动跟随测试（224 → 236 用例）: - 新增 maxTokens 钳制契约测试 ×9（max-tokens-clamp.test.ts）：mock fetch 记录真实请求体断言——超限钳制（MiMo standard/OpenAI gpt-4o/Anthropic opus）/ 未超限原样传递（DeepSeek/Agnes/MiMo pro/o3-mini/sonnet）/ 推理模型字段名 / 未配置默认值安全性 - 新增 MCP 动态同步端到端测试 ×3（mcp-tools-sync.test.ts）：mock MCP SDK + 真实 MCPManager/ToolRegistry/AgentEngineManager——先建引擎再连 server，断言同一会话请求的 tools 动态更新 / 断开后移除失效定义 / 回调异常不阻断 MCP 主流程能力矩阵核查结论（无回归确认）: 工具调用主链路 ✓（v0.5.2）/ SubAgent 工具 ✓（delegate 实时 resolveTools）/ thinking 热更新 ✓（baseConfig 合并路径无懒创建陷阱）/ 多模态当轮 ✓ / 压缩与孤立 tool 消息配对 ✓ / 摘要分层 ✓ / 记忆注入 ✓ / 故障转移 ✓ / 余额 ✓（v0.5.2）。已知设计限制：历史轮图片不回传（attachments 仅存缩略图，图片只在发送当轮注入上下文）。验证: lint 0 / typecheck 双工程 0 / test:electron 236 全过 / build 成功
2026-08-21 22:36:52 +08:00
parent e4d693c1c7
commit 3a30e8f5b4
12 changed files with 669 additions and 101 deletions
@@ -0,0 +1,196 @@
+/**
+ * Provider maxTokens 上限钳制契约测试（v0.5.3）
+ *
+ * 背景：引擎默认 maxTokens=63488（engine.ts DEFAULT_CONFIG），超过部分模型
+ * 上限时 API 直接 400 —— OpenAI gpt-4o(16384)/gpt-4.1(32768)、Anthropic
+ * opus/haiku(32000)、MiMo standard(32768) 曾不可用。v0.5.3 各 adapter 按
+ * MODEL_INFO.maxOutputTokens 钳制。
+ *
+ * 测试策略（v0.5.2 教训）：mock fetch 记录真实请求体并断言契约 ——
+ * 不 mock adapter 内部方法，验证"发出的 HTTP 请求体"这个最终事实。
+ */
+
+import { describe, it, expect, vi, beforeEach, afterEach } from 'vitest';
+
+vi.mock('electron-log', () => ({
+  default: { info: vi.fn(), warn: vi.fn(), error: vi.fn(), debug: vi.fn() },
+}));
+
+import { DeepSeekAdapter } from '../deepseek.adapter';
+import { AgnesAdapter } from '../agnes-ai.adapter';
+import { MimoAdapter } from '../mimo.adapter';
+import { OpenAIAdapter } from '../openai.adapter';
+import { AnthropicAdapter } from '../anthropic.adapter';
+import type { MetonaRequest } from '../../types';
+
+const mockFetch = vi.fn();
+vi.stubGlobal('fetch', mockFetch);
+
+beforeEach(() => {
+  mockFetch.mockReset();
+});
+afterEach(() => {
+  mockFetch.mockReset();
+});
+
+/** 引擎默认形态的请求（maxTokens=63488，与 engine DEFAULT_CONFIG 一致） */
+function makeRequest(overrides: Partial<MetonaRequest> = {}): MetonaRequest {
+  return {
+    meta: {
+      sessionId: 's',
+      iteration: 1,
+      requestId: 'r',
+      timestamp: Date.now(),
+      agentVersion: '1',
+    },
+    systemPrompt: { roleDefinition: '', outputConstraints: '', safetyGuidelines: '' },
+    messages: [{ role: 'user', content: 'hi', timestamp: Date.now() }],
+    params: {
+      maxTokens: 63_488,
+      temperature: 0,
+      stream: false,
+      thinkingEnabled: false,
+    },
+    ...overrides,
+  };
+}
+
+/** OpenAI 兼容系成功响应 */
+function openAIResponse(): Response {
+  return {
+    ok: true,
+    status: 200,
+    json: async () => ({ choices: [{ message: { content: 'ok' } }], usage: {} }),
+  } as unknown as Response;
+}
+
+/** Anthropic 成功响应 */
+function anthropicResponse(): Response {
+  return {
+    ok: true,
+    status: 200,
+    json: async () => ({ content: [{ type: 'text', text: 'ok' }], usage: {} }),
+  } as unknown as Response;
+}
+
+/** 取 mock fetch 收到的请求体 */
+function requestBody(): Record<string, unknown> {
+  const [, init] = mockFetch.mock.calls[0] as [string, RequestInit];
+  return JSON.parse(init.body as string) as Record<string, unknown>;
+}
+
+describe('maxTokens 模型上限钳制（引擎默认 63488 场景）', () => {
+  it('DeepSeek v4 pro（上限 384K）：63488 未超限，原样传递', async () => {
+    mockFetch.mockResolvedValue(openAIResponse());
+    const adapter = new DeepSeekAdapter({
+      provider: 'deepseek',
+      baseURL: 'https://api.deepseek.com',
+      apiKey: 'sk',
+      defaultModel: 'deepseek-v4-pro',
+    });
+    await adapter.send(makeRequest());
+    expect(requestBody().max_tokens).toBe(63_488);
+  });
+
+  it('Agnes flash（上限 65536）：63488 未超限，原样传递', async () => {
+    mockFetch.mockResolvedValue(openAIResponse());
+    const adapter = new AgnesAdapter({
+      provider: 'agnes',
+      baseURL: 'https://api.agnes.com/v1',
+      apiKey: 'k',
+      defaultModel: 'agnes-2.0-flash',
+    });
+    await adapter.send(makeRequest());
+    expect(requestBody().max_tokens).toBe(63_488);
+  });
+
+  it('MiMo standard（上限 32768）：钳制到 32768（原为 400 错误场景）', async () => {
+    mockFetch.mockResolvedValue(openAIResponse());
+    const adapter = new MimoAdapter({
+      provider: 'mimo',
+      baseURL: 'https://api.mimo.com/v1',
+      apiKey: 'k',
+      defaultModel: 'mimo-v2.5',
+    });
+    await adapter.send(makeRequest());
+    expect(requestBody().max_completion_tokens).toBe(32_768);
+  });
+
+  it('MiMo pro（上限 131072）：63488 未超限，原样传递', async () => {
+    mockFetch.mockResolvedValue(openAIResponse());
+    const adapter = new MimoAdapter({
+      provider: 'mimo',
+      baseURL: 'https://api.mimo.com/v1',
+      apiKey: 'k',
+      defaultModel: 'mimo-v2.5-pro',
+    });
+    await adapter.send(makeRequest());
+    expect(requestBody().max_completion_tokens).toBe(63_488);
+  });
+
+  it('OpenAI gpt-4o（上限 16384）：钳制到 16384（原为 400 错误场景）', async () => {
+    mockFetch.mockResolvedValue(openAIResponse());
+    const adapter = new OpenAIAdapter({
+      provider: 'openai',
+      baseURL: 'https://api.openai.com/v1',
+      apiKey: 'k',
+      defaultModel: 'gpt-4o',
+    });
+    await adapter.send(makeRequest());
+    expect(requestBody().max_tokens).toBe(16_384);
+  });
+
+  it('OpenAI o3-mini（上限 100K）：63488 未超限，推理模型字段名正确', async () => {
+    mockFetch.mockResolvedValue(openAIResponse());
+    const adapter = new OpenAIAdapter({
+      provider: 'openai',
+      baseURL: 'https://api.openai.com/v1',
+      apiKey: 'k',
+      defaultModel: 'o3-mini',
+    });
+    await adapter.send(makeRequest());
+    expect(requestBody().max_completion_tokens).toBe(63_488);
+    expect(requestBody().max_tokens).toBeUndefined();
+  });
+
+  it('Anthropic opus（上限 32000）：钳制到 32000（原为 400 错误场景）', async () => {
+    mockFetch.mockResolvedValue(anthropicResponse());
+    const adapter = new AnthropicAdapter({
+      provider: 'anthropic',
+      baseURL: 'https://api.anthropic.com',
+      apiKey: 'k',
+      defaultModel: 'claude-opus-4-1',
+    });
+    await adapter.send(makeRequest());
+    expect(requestBody().max_tokens).toBe(32_000);
+  });
+
+  it('Anthropic sonnet（上限 64000）：63488 未超限', async () => {
+    mockFetch.mockResolvedValue(anthropicResponse());
+    const adapter = new AnthropicAdapter({
+      provider: 'anthropic',
+      baseURL: 'https://api.anthropic.com',
+      apiKey: 'k',
+      defaultModel: 'claude-sonnet-4-5',
+    });
+    await adapter.send(makeRequest());
+    expect(requestBody().max_tokens).toBe(63_488);
+  });
+
+  it('未配置 maxTokens 时各 adapter 使用安全默认值（不超过模型上限）', async () => {
+    mockFetch.mockResolvedValue(openAIResponse());
+    const request = makeRequest({
+      params: { temperature: 0, stream: false } as MetonaRequest['params'],
+    });
+
+    // MiMo standard + thinking 默认 → 兜底 32768（= 上限，安全）
+    const mimo = new MimoAdapter({
+      provider: 'mimo',
+      baseURL: 'https://api.mimo.com/v1',
+      apiKey: 'k',
+      defaultModel: 'mimo-v2.5',
+    });
+    await mimo.send(request);
+    expect(requestBody().max_completion_tokens).toBe(32_768);
+  });
+});
@@ -48,21 +48,25 @@ export class AgnesAdapter extends BaseAdapter {
    const body = this.toNativeRequest(request, false);

    // #24 修复: 使用 fetchWithTimeout 替代 getFetchSignal + fetch，确保 timer 清理
-    const response = await this.fetchWithTimeout(`${this.config.baseURL}/chat/completions`, {
-      method: 'POST',
-      headers: {
-        'Content-Type': 'application/json',
-        Authorization: `Bearer ${this.config.apiKey}`,
-        ...this.config.headers,
+    const response = await this.fetchWithTimeout(
+      `${this.config.baseURL}/chat/completions`,
+      {
+        method: 'POST',
+        headers: {
+          'Content-Type': 'application/json',
+          Authorization: `Bearer ${this.config.apiKey}`,
+          ...this.config.headers,
+        },
+        body: JSON.stringify(body),
      },
-      body: JSON.stringify(body),
-    }, this.config.timeoutMs ?? 300_000);
+      this.config.timeoutMs ?? 300_000,
+    );

    if (!response.ok) {
      await this.throwHttpError(response, 'Agnes AI API error');
    }

-    const data = await response.json() as Record<string, unknown>;
+    const data = (await response.json()) as Record<string, unknown>;
    const parsed = parseOpenAICompatibleResponse(data);

    return {
@@ -88,15 +92,19 @@ export class AgnesAdapter extends BaseAdapter {
    const body = this.toNativeRequest(request, true);

    // #24 修复: 使用 fetchWithTimeout 替代 getFetchSignal + fetch，确保 timer 清理
-    const response = await this.fetchWithTimeout(`${this.config.baseURL}/chat/completions`, {
-      method: 'POST',
-      headers: {
-        'Content-Type': 'application/json',
-        Authorization: `Bearer ${this.config.apiKey}`,
-        ...this.config.headers,
+    const response = await this.fetchWithTimeout(
+      `${this.config.baseURL}/chat/completions`,
+      {
+        method: 'POST',
+        headers: {
+          'Content-Type': 'application/json',
+          Authorization: `Bearer ${this.config.apiKey}`,
+          ...this.config.headers,
+        },
+        body: JSON.stringify(body),
      },
-      body: JSON.stringify(body),
-    }, this.config.timeoutMs ?? 300_000);
+      this.config.timeoutMs ?? 300_000,
+    );

    if (!response.ok || !response.body) {
      await this.throwHttpError(response, 'Agnes AI stream error');
@@ -170,15 +178,22 @@ export class AgnesAdapter extends BaseAdapter {
    }

    if (imageCount > 0) {
-      const firstUrl = request.messages.find(m => m.images?.length)?.images?.[0]?.url ?? '';
-      log.info(`[Agnes] Processing ${imageCount} image(s), first URL prefix: ${firstUrl.slice(0, 50)}`);
+      const firstUrl = request.messages.find((m) => m.images?.length)?.images?.[0]?.url ?? '';
+      log.info(
+        `[Agnes] Processing ${imageCount} image(s), first URL prefix: ${firstUrl.slice(0, 50)}`,
+      );
    }

+    // v0.5.3: max_tokens 按模型上限钳制（agnes-2.0-flash 上限 65536）
+    const modelInfo = AgnesAdapter.MODEL_INFO[this.config.defaultModel];
+    const maxOutput = modelInfo?.maxOutputTokens ?? 65_536;
+    const maxTokens = Math.min(request.params.maxTokens ?? maxOutput, maxOutput);
+
    const body: Record<string, unknown> = {
      model: this.config.defaultModel,
      messages,
      temperature: request.params.temperature,
-      max_tokens: request.params.maxTokens ?? 65536,
+      max_tokens: maxTokens,
      stream,
    };

@@ -72,17 +72,21 @@ export class AnthropicAdapter extends BaseAdapter {

  async send(request: MetonaRequest): Promise<MetonaResponse> {
    const body = await this.toNativeRequest(request, false);
-    const response = await this.fetchWithTimeout(`${this.config.baseURL}/v1/messages`, {
-      method: 'POST',
-      headers: this.buildHeaders(),
-      body: JSON.stringify(body),
-    }, this.config.timeoutMs ?? 120_000);
+    const response = await this.fetchWithTimeout(
+      `${this.config.baseURL}/v1/messages`,
+      {
+        method: 'POST',
+        headers: this.buildHeaders(),
+        body: JSON.stringify(body),
+      },
+      this.config.timeoutMs ?? 120_000,
+    );

    if (!response.ok) {
      await this.throwHttpError(response, 'Anthropic API error');
    }

-    const data = await response.json() as Record<string, unknown>;
+    const data = (await response.json()) as Record<string, unknown>;
    return this.toMetonaResponse(data, request.meta.requestId);
  }

@@ -90,11 +94,15 @@ export class AnthropicAdapter extends BaseAdapter {

  async *sendStream(request: MetonaRequest): AsyncIterable<MetonaStreamEvent> {
    const body = await this.toNativeRequest(request, true);
-    const response = await this.fetchWithTimeout(`${this.config.baseURL}/v1/messages`, {
-      method: 'POST',
-      headers: this.buildHeaders(),
-      body: JSON.stringify(body),
-    }, this.config.timeoutMs ?? 300_000);
+    const response = await this.fetchWithTimeout(
+      `${this.config.baseURL}/v1/messages`,
+      {
+        method: 'POST',
+        headers: this.buildHeaders(),
+        body: JSON.stringify(body),
+      },
+      this.config.timeoutMs ?? 300_000,
+    );

    if (!response.ok || !response.body) {
      await this.throwHttpError(response, 'Anthropic stream error');
@@ -140,7 +148,11 @@ export class AnthropicAdapter extends BaseAdapter {
          if (delta?.type === 'text_delta' && typeof delta.text === 'string') {
            events.push({ type: MetonaStreamEventType.TEXT_DELTA, ...base(), delta: delta.text });
          } else if (delta?.type === 'thinking_delta' && typeof delta.thinking === 'string') {
-            events.push({ type: MetonaStreamEventType.REASONING_DELTA, ...base(), delta: delta.thinking });
+            events.push({
+              type: MetonaStreamEventType.REASONING_DELTA,
+              ...base(),
+              delta: delta.thinking,
+            });
          } else if (delta?.type === 'input_json_delta' && typeof delta.partial_json === 'string') {
            const block = toolBlocks.get(index);
            if (block) {
@@ -189,7 +201,8 @@ export class AnthropicAdapter extends BaseAdapter {
              usage: {
                inputTokens: (this.lastInputTokens as number) ?? 0,
                outputTokens: (usage.output_tokens as number) ?? 0,
-                totalTokens: ((this.lastInputTokens as number) ?? 0) + ((usage.output_tokens as number) ?? 0),
+                totalTokens:
+                  ((this.lastInputTokens as number) ?? 0) + ((usage.output_tokens as number) ?? 0),
              },
            });
          }
@@ -250,7 +263,10 @@ export class AnthropicAdapter extends BaseAdapter {
            yield ev;
          }
        } catch (parseErr) {
-          log.warn(`[Anthropic] Failed to parse SSE line: ${(parseErr as Error).message}`, trimmed.slice(0, 200));
+          log.warn(
+            `[Anthropic] Failed to parse SSE line: ${(parseErr as Error).message}`,
+            trimmed.slice(0, 200),
+          );
        }
      }
    }
@@ -291,7 +307,10 @@ export class AnthropicAdapter extends BaseAdapter {
   * 4. 连续同角色消息合并（API 要求严格交替）
   * 5. 首条消息必须为 user（历史以 assistant 开头时补占位）
   */
-  private async toNativeRequest(request: MetonaRequest, stream: boolean): Promise<Record<string, unknown>> {
+  private async toNativeRequest(
+    request: MetonaRequest,
+    stream: boolean,
+  ): Promise<Record<string, unknown>> {
    // System Prompt 拼接（Anthropic 使用顶层 system 字段）
    const system = [
      request.systemPrompt.roleDefinition,
@@ -303,7 +322,10 @@ export class AnthropicAdapter extends BaseAdapter {
      .join('\n\n');

    // 转换消息（非 system）
-    const converted: Array<{ role: 'user' | 'assistant'; content: Array<Record<string, unknown>> }> = [];
+    const converted: Array<{
+      role: 'user' | 'assistant';
+      content: Array<Record<string, unknown>>;
+    }> = [];
    for (const m of request.messages) {
      if (m.role === 'system') continue;

@@ -316,7 +338,9 @@ export class AnthropicAdapter extends BaseAdapter {
            : JSON.stringify(m.toolResult.result);
        converted.push({
          role: 'user',
-          content: [{ type: 'tool_result', tool_use_id: m.toolResult.toolCallId, content: contentStr }],
+          content: [
+            { type: 'tool_result', tool_use_id: m.toolResult.toolCallId, content: contentStr },
+          ],
        });
        continue;
      }
@@ -345,7 +369,8 @@ export class AnthropicAdapter extends BaseAdapter {
    }

    // 合并连续同角色消息（Anthropic 要求 user/assistant 交替）
-    const merged: Array<{ role: 'user' | 'assistant'; content: Array<Record<string, unknown>> }> = [];
+    const merged: Array<{ role: 'user' | 'assistant'; content: Array<Record<string, unknown>> }> =
+      [];
    for (const msg of converted) {
      const last = merged[merged.length - 1];
      if (last && last.role === msg.role) {
@@ -357,12 +382,20 @@ export class AnthropicAdapter extends BaseAdapter {

    // 首条消息必须为 user
    if (merged.length === 0 || merged[0].role !== 'user') {
-      merged.unshift({ role: 'user', content: [{ type: 'text', text: '[Conversation history follows]' }] });
+      merged.unshift({
+        role: 'user',
+        content: [{ type: 'text', text: '[Conversation history follows]' }],
+      });
    }

+    // v0.5.3: max_tokens 按模型上限钳制（sonnet 64000 / opus 32000 / haiku 32000）—
+    // 引擎默认 63488 超过 opus/haiku 上限时 API 直接 400；thinking budget 已在此值内二分
+    const anthropicMaxOutput =
+      AnthropicAdapter.MODEL_INFO[this.config.defaultModel]?.maxOutputTokens ?? 64_000;
+
    const body: Record<string, unknown> = {
      model: this.config.defaultModel,
-      max_tokens: request.params.maxTokens ?? 8192,
+      max_tokens: Math.min(request.params.maxTokens ?? 8192, anthropicMaxOutput),
      system,
      messages: merged,
      stream,
@@ -379,7 +412,12 @@ export class AnthropicAdapter extends BaseAdapter {

    // Thinking 模式：budget_tokens（必须小于 max_tokens，此处钳制到一半）
    if (request.params.thinkingEnabled) {
-      const budgetMap: Record<string, number> = { low: 1024, medium: 4096, high: 16384, max: 32768 };
+      const budgetMap: Record<string, number> = {
+        low: 1024,
+        medium: 4096,
+        high: 16384,
+        max: 32768,
+      };
      const budget = Math.min(
        budgetMap[request.params.thinkingEffort ?? 'high'] ?? 16384,
        Math.floor((body.max_tokens as number) / 2),
@@ -435,7 +473,8 @@ export class AnthropicAdapter extends BaseAdapter {

    for (const block of contentBlocks) {
      if (block.type === 'text') text += (block.text as string) ?? '';
-      else if (block.type === 'thinking') reasoningContent = (block.thinking as string) ?? undefined;
+      else if (block.type === 'thinking')
+        reasoningContent = (block.thinking as string) ?? undefined;
      else if (block.type === 'tool_use') {
        let args: Record<string, unknown> = {};
        const rawInput = block.input;
@@ -453,9 +492,11 @@ export class AnthropicAdapter extends BaseAdapter {
    const usage = (data.usage as Record<string, number>) ?? {};
    const stopReason = (data.stop_reason as string) ?? 'end_turn';
    const finishReason: MetonaFinishReason =
-      stopReason === 'tool_use' ? MetonaFinishReason.TOOL_CALLS
-      : stopReason === 'max_tokens' ? MetonaFinishReason.LENGTH
-      : MetonaFinishReason.STOP;
+      stopReason === 'tool_use'
+        ? MetonaFinishReason.TOOL_CALLS
+        : stopReason === 'max_tokens'
+          ? MetonaFinishReason.LENGTH
+          : MetonaFinishReason.STOP;

    return {
      meta: {
@@ -239,11 +239,16 @@ export class DeepSeekAdapter extends BaseAdapter {
    const messages = buildOpenAICompatibleMessages(request);
    const tools = buildOpenAICompatibleTools(request.tools);

+    // v0.5.3: max_tokens 按模型上限钳制 — 引擎默认 63488 超过部分模型上限时 API 直接 400
+    const modelInfo = DeepSeekAdapter.MODEL_INFO[this.config.defaultModel];
+    const maxOutput = modelInfo?.maxOutputTokens ?? 384_000;
+    const maxTokens = Math.min(request.params.maxTokens ?? maxOutput, maxOutput);
+
    const body: Record<string, unknown> = {
      model: this.config.defaultModel,
      messages,
      temperature: request.params.temperature,
-      max_tokens: request.params.maxTokens,
+      max_tokens: maxTokens,
      stream,
    };

@@ -59,21 +59,25 @@ export class MimoAdapter extends BaseAdapter {
    const body = this.toNativeRequest(request, false);

    // #24 修复: 使用 fetchWithTimeout 替代 getFetchSignal + fetch，确保 timer 清理
-    const response = await this.fetchWithTimeout(`${this.config.baseURL}/chat/completions`, {
-      method: 'POST',
-      headers: {
-        'Content-Type': 'application/json',
-        Authorization: `Bearer ${this.config.apiKey}`,
-        ...this.config.headers,
+    const response = await this.fetchWithTimeout(
+      `${this.config.baseURL}/chat/completions`,
+      {
+        method: 'POST',
+        headers: {
+          'Content-Type': 'application/json',
+          Authorization: `Bearer ${this.config.apiKey}`,
+          ...this.config.headers,
+        },
+        body: JSON.stringify(body),
      },
-      body: JSON.stringify(body),
-    }, this.config.timeoutMs ?? 120_000);
+      this.config.timeoutMs ?? 120_000,
+    );

    if (!response.ok) {
      await this.throwHttpError(response, 'MiMo API error');
    }

-    const data = await response.json() as Record<string, unknown>;
+    const data = (await response.json()) as Record<string, unknown>;
    const parsed = parseOpenAICompatibleResponse(data);

    return {
@@ -98,15 +102,19 @@ export class MimoAdapter extends BaseAdapter {
    const body = this.toNativeRequest(request, true);

    // #24 修复: 使用 fetchWithTimeout 替代 getFetchSignal + fetch，确保 timer 清理
-    const response = await this.fetchWithTimeout(`${this.config.baseURL}/chat/completions`, {
-      method: 'POST',
-      headers: {
-        'Content-Type': 'application/json',
-        Authorization: `Bearer ${this.config.apiKey}`,
-        ...this.config.headers,
+    const response = await this.fetchWithTimeout(
+      `${this.config.baseURL}/chat/completions`,
+      {
+        method: 'POST',
+        headers: {
+          'Content-Type': 'application/json',
+          Authorization: `Bearer ${this.config.apiKey}`,
+          ...this.config.headers,
+        },
+        body: JSON.stringify(body),
      },
-      body: JSON.stringify(body),
-    }, this.config.timeoutMs ?? 300_000);
+      this.config.timeoutMs ?? 300_000,
+    );

    if (!response.ok || !response.body) {
      await this.throwHttpError(response, 'MiMo stream error');
@@ -187,16 +195,18 @@ export class MimoAdapter extends BaseAdapter {
      messages[i].content = contentParts;
    }

+    // MiMo 使用 max_completion_tokens（非 max_tokens）
+    // #41 修复: thinking 模式下未配置时兜底 32768（thinking 占用 token 配额，API 默认值过小会截断输出）
+    // v0.5.3: 按模型上限钳制（pro 131072 / standard 32768）—
+    // 引擎默认 63488 超过 standard 上限时 API 直接 400
+    const mimoMaxOutput =
+      MimoAdapter.MODEL_INFO[this.config.defaultModel]?.maxOutputTokens ?? 131_072;
+    const mimoDefault = request.params.thinkingEnabled !== false ? 32_768 : mimoMaxOutput;
+
    const body: Record<string, unknown> = {
      model: this.config.defaultModel,
      messages,
-      // MiMo 使用 max_completion_tokens（非 max_tokens）
-      // #41 修复: thinking 模式下 max_completion_tokens 未配置时使用兜底默认值（32768）
-      // thinking 占用 token 配额，未配置时 API 默认值可能过小导致输出被截断
-      // thinkingEnabled !== false 包含 true 和 undefined（MiMo 默认 enabled）两种情况
-      // 审查修复: 兜底值从 63488 改为 32768，避免超出标准版上限导致 API 400
-      max_completion_tokens: request.params.maxTokens
-        ?? (request.params.thinkingEnabled !== false ? 32768 : undefined),
+      max_completion_tokens: Math.min(request.params.maxTokens ?? mimoDefault, mimoMaxOutput),
      stream,
    };

@@ -69,21 +69,25 @@ export class OpenAIAdapter extends BaseAdapter {
  async send(request: MetonaRequest): Promise<MetonaResponse> {
    const body = this.toNativeRequest(request, false);

-    const response = await this.fetchWithTimeout(`${this.config.baseURL}/chat/completions`, {
-      method: 'POST',
-      headers: {
-        'Content-Type': 'application/json',
-        Authorization: `Bearer ${this.config.apiKey}`,
-        ...this.config.headers,
+    const response = await this.fetchWithTimeout(
+      `${this.config.baseURL}/chat/completions`,
+      {
+        method: 'POST',
+        headers: {
+          'Content-Type': 'application/json',
+          Authorization: `Bearer ${this.config.apiKey}`,
+          ...this.config.headers,
+        },
+        body: JSON.stringify(body),
      },
-      body: JSON.stringify(body),
-    }, this.config.timeoutMs ?? 120_000);
+      this.config.timeoutMs ?? 120_000,
+    );

    if (!response.ok) {
      await this.throwHttpError(response, 'OpenAI API error');
    }

-    const data = await response.json() as Record<string, unknown>;
+    const data = (await response.json()) as Record<string, unknown>;
    const parsed = parseOpenAICompatibleResponse(data);

    return {
@@ -107,15 +111,19 @@ export class OpenAIAdapter extends BaseAdapter {
  async *sendStream(request: MetonaRequest): AsyncIterable<MetonaStreamEvent> {
    const body = this.toNativeRequest(request, true);

-    const response = await this.fetchWithTimeout(`${this.config.baseURL}/chat/completions`, {
-      method: 'POST',
-      headers: {
-        'Content-Type': 'application/json',
-        Authorization: `Bearer ${this.config.apiKey}`,
-        ...this.config.headers,
+    const response = await this.fetchWithTimeout(
+      `${this.config.baseURL}/chat/completions`,
+      {
+        method: 'POST',
+        headers: {
+          'Content-Type': 'application/json',
+          Authorization: `Bearer ${this.config.apiKey}`,
+          ...this.config.headers,
+        },
+        body: JSON.stringify(body),
      },
-      body: JSON.stringify(body),
-    }, this.config.timeoutMs ?? 300_000);
+      this.config.timeoutMs ?? 300_000,
+    );

    if (!response.ok || !response.body) {
      await this.throwHttpError(response, 'OpenAI stream error');
@@ -139,7 +147,7 @@ export class OpenAIAdapter extends BaseAdapter {
        signal: AbortSignal.timeout(10_000),
      });
      if (response.ok) {
-        const data = await response.json() as { data?: Array<{ id: string }> };
+        const data = (await response.json()) as { data?: Array<{ id: string }> };
        if (data.data?.length) {
          return data.data.map((m) => OpenAIAdapter.MODEL_INFO[m.id] ?? { id: m.id });
        }
@@ -208,15 +216,19 @@ export class OpenAIAdapter extends BaseAdapter {
    };

    // Token 上限参数：o 系列/gpt-5 使用 max_completion_tokens
-    if (request.params.maxTokens) {
+    // v0.5.3: 按模型上限钳制（gpt-4o 16384 / gpt-4.1 32768 / o3-mini 100000）—
+    // 引擎默认 63488 超过 gpt-4o/gpt-4.1 上限时 API 直接 400
+    const oaMaxOutput = OpenAIAdapter.MODEL_INFO[model]?.maxOutputTokens ?? 128_000;
+    const oaMaxTokens = Math.min(
+      request.params.maxTokens ?? (isReasoningModel ? 32_768 : oaMaxOutput),
+      oaMaxOutput,
+    );
+    if (oaMaxTokens) {
      if (isReasoningModel) {
-        body.max_completion_tokens = request.params.maxTokens;
+        body.max_completion_tokens = oaMaxTokens;
      } else {
-        body.max_tokens = request.params.maxTokens;
+        body.max_tokens = oaMaxTokens;
      }
-    } else if (isReasoningModel) {
-      // 推理模型未配置时使用兜底值（thinking 占用 token 配额，默认值过小会被截断）
-      body.max_completion_tokens = 32_768;
    }

    if (stream) {
@@ -229,7 +241,12 @@ export class OpenAIAdapter extends BaseAdapter {

    // Thinking 模式：推理模型映射 reasoning_effort；非推理模型忽略
    if (request.params.thinkingEnabled && isReasoningModel) {
-      const effortMap: Record<string, string> = { low: 'low', medium: 'medium', high: 'high', max: 'high' };
+      const effortMap: Record<string, string> = {
+        low: 'low',
+        medium: 'medium',
+        high: 'high',
+        max: 'high',
+      };
      body.reasoning_effort = effortMap[request.params.thinkingEffort ?? 'high'] ?? 'high';
    } else if (!isReasoningModel) {
      // 非推理模型使用温度控制