merge: sync main and address review feedback

README.md

@@ -26,6 +26,7 @@ PR #1 曾把 reasoning 模型的 `defaultTemperature` / `defaultTopP` 写为 `nu
 | `provider.schema.json` | `compute/providers/*.json`、`compute/coding-plans/*.json` | `defaultTemperature`/`defaultTopP` 必须是 number，禁止 null/string；`additionalProperties: false` |
 | `manifest.schema.json` | `manifest.json` | `presetDataVersion` 必须是递增整数 |
 | `service-map.schema.json` | `compute/service-map.json` | 每条映射须含 `modelName` + `providerId` |
+| `smart-model-catalog.schema.json` | `compute/smart-routing/model-catalog.json` | 智能路由三档量级、接入面、exact model 能力和稳定优先级 |
 | `providers-index.schema.json` | 两个 `_index.json` | `order` 数组无重复 |
 | `pricing.schema.json` | `compute/pricing.json` | `markupRatio` / `usdToCny` 为正数 |
 
@@ -62,6 +63,17 @@ PR #1 曾把 reasoning 模型的 `defaultTemperature` / `defaultTopP` 写为 `nu
 
 否则老客户端会因未知字段校验失败死锁。
 
+### 智能路由目录的边界
+
+`compute/smart-routing/model-catalog.json` 是路由器直接消费的、按接入面核对过的策略快照：
+
+- `tier`、`routingPriority`、`eligibleForAgent` 和 `defaultReference` 属于路由策略；
+- `capabilities`、上下文和 reasoning 是该 `providerId + model` 接入面的可用能力快照；
+- 它不声明 API key、baseUrl、登录状态、用户额度或实时计价；`desirecore-cloud` 的连接与计费状态仍由登录后的 Provider 接口动态下发；
+- Codex、Claude 条目必须能在对应 `compute/providers/*.json` 中按 exact model 找到，测试会阻止已下线模型继续参与路由。
+
+客户端把本文件作为可热更新主数据源，并保留同 Schema 的内置离线兜底。调整 Provider 或 model-spec 的能力事实时，应同步审阅本目录，避免路由快照漂移。
+
 ---
 
 ## 本地校验
@@ -78,7 +90,7 @@ CI（GitHub Actions）会在每个 PR 自动运行 `validate` 和 `test`，不
 
 ## 数据修改流程
 
-1. 编辑 `compute/providers/<name>.json`、`compute/coding-plans/<name>.json` 或 `compute/service-map.json`
+1. 编辑 `compute/providers/<name>.json`、`compute/coding-plans/<name>.json`、`compute/service-map.json` 或 `compute/smart-routing/model-catalog.json`
 2. 编辑 `compute/providers/_index.json` 或 `coding-plans/_index.json`（新增/删除 provider 时）
 3. **必须**递增 `manifest.json#presetDataVersion`，并更新 `updatedAt`
 4. `npm run validate` 本地确认通过
@@ -95,3 +107,4 @@ CI（GitHub Actions）会在每个 PR 自动运行 `validate` 和 `test`，不
 - **构建期同步**：`npm run sync-config-center` 把数据复制到 desirecore 主仓 `lib/agent-service/defaults/`
 - **运行时同步**：客户端启动后后台 git fetch 本仓库，每 30 分钟检查一次远程更新
 - **版本比对**：`presetDataVersion`（递增整数）+ digest（SHA-256）双重校验
+- **智能路由目录**：新客户端按文件 mtime 热加载 `compute/smart-routing/model-catalog.json`；缺失或校验失败时使用随客户端发布的内置 JSON

__tests__/validate.test.mjs

@@ -85,6 +85,49 @@ describe('真实数据全量校验', () => {
     assert.equal(result.ok, true, JSON.stringify(result.errors, null, 2))
   })
 
+  it('智能路由模型目录应通过 smart-model-catalog schema', () => {
+    const result = validateFile(
+      join(ROOT, 'compute', 'smart-routing', 'model-catalog.json'),
+      validators,
+    )
+    assert.equal(result.ok, true, JSON.stringify(result.errors, null, 2))
+
+    const catalog = JSON.parse(readFileSync(
+      join(ROOT, 'compute', 'smart-routing', 'model-catalog.json'),
+      'utf8',
+    ))
+    assert.deepEqual(catalog.tiers.map((tier) => tier.id), [
+      'flagship',
+      'balanced',
+      'lightweight',
+    ])
+    assert.deepEqual(catalog.providers.map((provider) => provider.provider), [
+      'openai-codex',
+      'anthropic-claude',
+      'desirecore-cloud',
+    ])
+    assert.equal(
+      catalog.providers.flatMap((provider) => provider.models).length,
+      37,
+    )
+
+    for (const provider of catalog.providers.filter((item) => item.provider !== 'desirecore-cloud')) {
+      const providerFile = JSON.parse(readFileSync(
+        join(ROOT, 'compute', 'providers', `${provider.provider}.json`),
+        'utf8',
+      ))
+      assert.equal(providerFile.id, provider.providerId)
+      const publishedModels = new Set(providerFile.models.map((model) => model.modelName))
+      for (const model of provider.models) {
+        assert.equal(
+          publishedModels.has(model.model),
+          true,
+          `${provider.providerId}/${model.model} 必须存在于对应接入面 Provider 清单`,
+        )
+      }
+    }
+  })
+
   it('两个 _index.json 应通过 providers-index schema', () => {
     const r1 = validateFile(join(ROOT, 'compute', 'providers', '_index.json'), validators)
     const r2 = validateFile(join(ROOT, 'compute', 'coding-plans', '_index.json'), validators)
@@ -102,6 +145,66 @@ describe('真实数据全量校验', () => {
     assert.equal(result.ok, true, JSON.stringify(result.errors, null, 2))
   })
 
+  it('WebSearch 服务端能力仅标记官方直连 Anthropic/OpenAI 模型', () => {
+    const anthropic = JSON.parse(readFileSync(join(ROOT, 'compute', 'providers', 'anthropic.json'), 'utf8'))
+    const openai = JSON.parse(readFileSync(join(ROOT, 'compute', 'providers', 'openai.json'), 'utf8'))
+    const openaiCodex = JSON.parse(readFileSync(join(ROOT, 'compute', 'providers', 'openai-codex.json'), 'utf8'))
+
+    const enabled = (provider) => provider.models
+      .filter((model) => model.extra?.serverSideWebSearch?.enabled === true)
+      .map((model) => model.modelName)
+
+    assert.deepEqual(enabled(anthropic), [
+      'claude-fable-5',
+      'claude-opus-5',
+      'claude-sonnet-5',
+    ])
+    assert.deepEqual(enabled(openai), [
+      'gpt-5.5',
+      'gpt-5.5-pro',
+      'gpt-5.4',
+      'gpt-5.4-pro',
+      'gpt-5.4-mini',
+      'gpt-5.4-nano',
+      'gpt-5',
+      'gpt-4.1',
+      'gpt-4.1-mini',
+      'o4-mini',
+    ])
+    assert.deepEqual(enabled(openaiCodex), [])
+
+    for (const model of anthropic.models.filter((item) => enabled(anthropic).includes(item.modelName))) {
+      assert.equal(model.extra.serverSideWebSearch.dialect, 'anthropic-messages')
+      assert.equal(model.extra.serverSideWebSearch.toolType, 'web_search_20260318')
+      assert.equal(model.extra.serverSideWebSearch.searchRequestPriceUsd, 0.01)
+      assert.equal('maxUses' in model.extra.serverSideWebSearch, false)
+    }
+    for (const model of openai.models.filter((item) => enabled(openai).includes(item.modelName))) {
+      assert.equal(model.extra.serverSideWebSearch.dialect, 'openai-responses')
+      assert.equal(model.extra.serverSideWebSearch.toolType, 'web_search')
+      assert.equal(model.extra.serverSideWebSearch.searchRequestPriceUsd, 0.01)
+      assert.equal('maxUses' in model.extra.serverSideWebSearch, false)
+    }
+  })
+
+  it('Brave Search API 使用显式密钥声明且默认关闭', () => {
+    const brave = JSON.parse(readFileSync(join(ROOT, 'api-providers', 'web-search', 'brave.json'), 'utf8'))
+    const index = JSON.parse(readFileSync(join(ROOT, 'api-providers', 'web-search', '_index.json'), 'utf8'))
+
+    assert.equal(brave.enabled, false)
+    assert.equal(brave.endpoint, 'https://api.search.brave.com/res/v1/web/search')
+    assert.deepEqual(brave.auth, {
+      type: 'header',
+      headerName: 'X-Subscription-Token',
+      apiKeyRef: 'brave',
+    })
+    assert.deepEqual(index.order, ['tavily', 'brave', 'serper'])
+    assert.equal(
+      validateFile(join(ROOT, 'api-providers', 'web-search', 'brave.json'), validators).ok,
+      true,
+    )
+  })
+
   it('DeepSeek V4 Pro/Flash 应与官方的 1M/384K reasoning profile 一致', () => {
     const provider = JSON.parse(readFileSync(join(ROOT, 'compute', 'providers', 'deepseek.json'), 'utf8'))
     const specFile = JSON.parse(readFileSync(join(ROOT, 'compute', 'model-specs', 'deepseek.json'), 'utf8'))
@@ -135,6 +238,28 @@ describe('真实数据全量校验', () => {
     }
   })
 
+  it('新增模型应保留官方输出上限和工作流分类', () => {
+    const loadSpecs = (provider) => JSON.parse(readFileSync(
+      join(ROOT, 'compute', 'model-specs', `${provider}.json`),
+      'utf8',
+    )).specs
+
+    const gemini = loadSpecs('google').find((item) => item.id === 'gemini-3.1-flash-lite-image')
+    assert.ok(gemini)
+    assert.equal(gemini.spec.maxOutputTokens, 4096)
+    assert.ok(gemini.spec.capabilities.includes('image_editing'))
+
+    const sonnet = loadSpecs('anthropic').filter((item) => item.id === 'claude-sonnet-5')
+    assert.equal(sonnet.length, 1, 'claude-sonnet-5 应且仅应有一条规格')
+    assert.ok(sonnet[0].spec.capabilities.includes('computer_use'))
+    assert.ok(sonnet[0].spec.serviceType.includes('computer_use'))
+
+    const grok = loadSpecs('xai').find((item) => item.id === 'grok-4.5')
+    assert.ok(grok)
+    assert.ok(grok.spec.capabilities.includes('reasoning'))
+    assert.ok(grok.spec.serviceType.includes('reasoning'))
+  })
+
   it('MiMo V2.5 ASR 应有独立精确规格，避免回落到 MiMo V2.5 family', () => {
     const specFile = JSON.parse(readFileSync(join(ROOT, 'compute', 'model-specs', 'xiaomi.json'), 'utf8'))
     const specs = specFile.specs.filter((item) => item.id === 'mimo-v2.5-asr')
@@ -145,9 +270,61 @@ describe('真实数据全量校验', () => {
     assert.ok(specs[0].spec.capabilities.includes('asr'))
   })
 
+  it('MiMo V2.5 仅非 Pro 型号应声明多模态能力', () => {
+    const provider = JSON.parse(readFileSync(join(ROOT, 'compute', 'providers', 'xiaomi.json'), 'utf8'))
+    const specFile = JSON.parse(readFileSync(join(ROOT, 'compute', 'model-specs', 'xiaomi.json'), 'utf8'))
+    const specFor = (modelId) => specFile.specs.find((item) => item.id === modelId)
+
+    const mimoV25 = specFor('mimo-v2.5')
+    assert.ok(mimoV25, 'model-specs 缺少 mimo-v2.5')
+    assert.ok(mimoV25.spec.capabilities.includes('vision'))
+    assert.ok(mimoV25.spec.serviceType.includes('vision'))
+
+    for (const modelId of ['mimo-v2.5-pro', 'mimo-v2-pro']) {
+      const modelSpec = specFor(modelId)
+      assert.ok(modelSpec, `model-specs 缺少 ${modelId}`)
+      assert.equal(modelSpec.spec.capabilities.includes('vision'), false)
+      assert.equal(modelSpec.spec.serviceType.includes('vision'), false)
+    }
+
+    const pro = provider.models.find((item) => item.modelName === 'mimo-v2.5-pro')
+    assert.ok(pro, 'provider 缺少 mimo-v2.5-pro')
+    assert.equal(pro.capabilities.includes('vision'), false)
+    assert.equal(pro.serviceType.includes('vision'), false)
+  })
+
+  it('Qwen3.8 Max Preview 应在 Token Plan 中提供完整的推理与视觉规格', () => {
+    const tokenPlan = JSON.parse(readFileSync(join(ROOT, 'compute', 'coding-plans', 'dashscope-token-plan.json'), 'utf8'))
+    const specFile = JSON.parse(readFileSync(join(ROOT, 'compute', 'model-specs', 'qwen.json'), 'utf8'))
+    const modelId = 'qwen3.8-max-preview'
+    const model = tokenPlan.models.find((item) => item.modelName === modelId)
+
+    assert.ok(model, `Token Plan 缺少 ${modelId}`)
+    assert.equal(model.contextWindow, 983616)
+    assert.equal(model.defaultTemperature, 0.6)
+    assert.ok(model.capabilities.includes('reasoning'))
+    assert.ok(model.capabilities.includes('vision'))
+    assert.ok(model.serviceType.includes('reasoning'))
+    assert.ok(model.serviceType.includes('vision'))
+    assert.deepEqual(model.extra.reasoning.supportedEfforts, ['low', 'high', 'xhigh'])
+    assert.equal(model.extra.reasoning.defaultEffort, 'xhigh')
+    assert.equal(model.extra.thinkingOnly, true)
+    assert.equal(model.extra.thinkingMaxTokens, 262144)
+    assert.equal(model.extra.preserveThinkingDefault, true)
+    assert.equal(model.extra.supportsParallelToolCalls, false)
+
+    const specs = specFile.specs.filter((item) => item.id === modelId)
+    assert.equal(specs.length, 1, `model-specs 中 ${modelId} 应且仅应有一条规格`)
+    assert.equal(specs[0].spec.contextWindow, 1000000)
+    assert.equal(specs[0].spec.defaultTemperature, 0.6)
+    assert.equal(specs[0].spec.supportsReasoning, true)
+    assert.ok(specs[0].spec.capabilities.includes('vision'))
+  })
+
   it('所有 Provider 应按供应商归属计价，模型来源不覆盖供应商币种', () => {
     const expectedCurrencies = {
       anthropic: 'USD',
+      'anthropic-claude': 'USD',
       baichuan: 'CNY',
       baidu: 'CNY',
       cohere: 'USD',
@@ -220,6 +397,28 @@ describe('真实数据全量校验', () => {
   })
 })
 
+describe('智能路由模型目录 schema 反例', () => {
+  const validate = compile('smart-model-catalog')
+
+  it('拒绝未声明的策略字段，避免新旧客户端静默分叉', () => {
+    const data = JSON.parse(readFileSync(
+      join(ROOT, 'compute', 'smart-routing', 'model-catalog.json'),
+      'utf8',
+    ))
+    data.providers[0].models[0].unknownRoutingPolicy = true
+    assert.equal(validate(data), false)
+  })
+
+  it('拒绝当前客户端未声明支持的目录版本', () => {
+    const data = JSON.parse(readFileSync(
+      join(ROOT, 'compute', 'smart-routing', 'model-catalog.json'),
+      'utf8',
+    ))
+    data.version = 2
+    assert.equal(validate(data), false)
+  })
+})
+
 // ==================== Runtime 清单反例 ====================
 
 describe('runtime-recommended schema 反例', () => {

api-providers/web-search/_index.json

@@ -1,4 +1,4 @@
 {
   "description": "web_search 工具后端加载顺序索引",
-  "order": ["tavily", "serper"]
+  "order": ["tavily", "brave", "serper"]
 }

api-providers/web-search/brave.json

@@ -0,0 +1,35 @@
+{
+  "id": "brave",
+  "name": "Brave Search API",
+  "capability": "web_search",
+  "enabled": false,
+  "builtin": true,
+  "priority": 20,
+  "endpoint": "https://api.search.brave.com/res/v1/web/search",
+  "method": "GET",
+  "auth": {
+    "type": "header",
+    "headerName": "X-Subscription-Token",
+    "apiKeyRef": "brave"
+  },
+  "request": {
+    "headers": {
+      "Accept": "application/json",
+      "Accept-Encoding": "gzip"
+    },
+    "queryParams": {
+      "q": "${query}",
+      "count": "${maxResults}"
+    }
+  },
+  "response": {
+    "resultsPath": "web.results",
+    "item": {
+      "titlePath": "title",
+      "urlPath": "url",
+      "snippetPath": "description",
+      "pageAgePath": "page_age"
+    }
+  },
+  "timeoutMs": 10000
+}

api-providers/web-search/serper.json

@@ -4,7 +4,7 @@
   "capability": "web_search",
   "enabled": false,
   "builtin": true,
-  "priority": 20,
+  "priority": 30,
   "endpoint": "https://google.serper.dev/search",
   "method": "POST",
   "auth": { "type": "header", "headerName": "X-API-KEY", "apiKeyRef": "serper" },

compute/coding-plans/dashscope-token-plan.json

@@ -26,6 +26,42 @@
     }
   },
   "models": [
+    {
+      "modelName": "qwen3.8-max-preview",
+      "displayName": "Qwen3.8 Max Preview (Token Plan)",
+      "serviceType": [
+        "chat",
+        "reasoning",
+        "vision"
+      ],
+      "description": "百炼 Token Plan，通义千问 3.8 Max 预览版，支持文本与视觉输入，始终开启深度思考",
+      "contextWindow": 983616,
+      "maxOutputTokens": 65536,
+      "capabilities": [
+        "chat",
+        "reasoning",
+        "deep_thinking",
+        "code",
+        "vision",
+        "tool_use",
+        "agent",
+        "long_context"
+      ],
+      "defaultTemperature": 0.6,
+      "extra": {
+        "reasoning": {
+          "supportedEfforts": ["low", "high", "xhigh"],
+          "defaultEffort": "xhigh"
+        },
+        "supportsThinking": true,
+        "thinkingDefault": true,
+        "thinkingOnly": true,
+        "thinkingMaxTokens": 262144,
+        "preserveThinkingDefault": true,
+        "supportsParallelToolCalls": false
+      },
+      "source": "preset"
+    },
     {
       "modelName": "qwen3.7-max",
       "displayName": "Qwen3.7 Max (Token Plan)",

compute/model-specs/anthropic.json

@@ -1,20 +1,6 @@
 {
   "description": "Anthropic Claude 系列模型规格。参数来源：config-center compute/providers/anthropic.json。",
   "specs": [
-    {
-      "id": "claude-sonnet-5",
-      "displayName": "Claude Sonnet 5",
-      "family": "claude-sonnet-5",
-      "match": { "exact": ["claude-sonnet-5"], "patterns": ["claude-sonnet-5*"] },
-      "spec": {
-        "contextWindow": 1000000,
-        "maxOutputTokens": 128000,
-        "capabilities": ["chat", "reasoning", "code", "tool_use", "agent", "vision", "long_context"],
-        "serviceType": ["chat"],
-        "supportsReasoning": true,
-        "description": "Anthropic 最强 Sonnet 级模型，支持自适应思考（effort: low/medium/high/max）"
-      }
-    },
     {
       "id": "claude-fable-5",
       "displayName": "Claude Fable 5",
@@ -29,6 +15,20 @@
         "description": "Anthropic 最强模型，基于 Mythos 架构，支持自主知识工作"
       }
     },
+    {
+      "id": "claude-opus-5",
+      "displayName": "Claude Opus 5",
+      "family": "claude-opus",
+      "match": { "patterns": ["claude-opus-5*"] },
+      "spec": {
+        "contextWindow": 1000000,
+        "maxOutputTokens": 128000,
+        "capabilities": ["chat", "reasoning", "code", "vision", "tool_use", "long_context"],
+        "serviceType": ["chat"],
+        "supportsReasoning": true,
+        "description": "Claude Opus 5，Anthropic Opus 系列当前旗舰；思考默认开启，effort 支持到 max"
+      }
+    },
     {
       "id": "claude-opus-4-8",
       "displayName": "Claude Opus 4.8",
@@ -58,6 +58,20 @@
         "description": "Anthropic 当前最强通用模型，适合复杂推理和智能体编码任务"
       }
     },
+    {
+      "id": "claude-sonnet-5",
+      "displayName": "Claude Sonnet 5",
+      "family": "claude-sonnet",
+      "match": { "patterns": ["claude-sonnet-5*"] },
+      "spec": {
+        "contextWindow": 1000000,
+        "maxOutputTokens": 128000,
+        "capabilities": ["chat", "reasoning", "code", "vision", "tool_use", "computer_use", "long_context"],
+        "serviceType": ["chat", "computer_use"],
+        "supportsReasoning": true,
+        "description": "Claude Sonnet 5，速度与智能平衡；思考默认开启，effort 支持到 max"
+      }
+    },
     {
       "id": "claude-sonnet-4-6",
       "displayName": "Claude Sonnet 4.6",

compute/model-specs/google.json

@@ -8,8 +8,8 @@
       "match": { "exact": ["gemini-3.1-flash-lite-image"], "patterns": ["gemini-3.1-flash-lite-image*"] },
       "spec": {
         "contextWindow": 65536,
-        "maxOutputTokens": 65536,
+        "maxOutputTokens": 4096,
-        "capabilities": ["chat", "vision", "image_generation"],
+        "capabilities": ["chat", "vision", "image_generation", "image_editing"],
         "serviceType": ["image_gen"],
         "description": "Nano Banana 2 Lite，Google 最快的文生图模型"
       }

compute/model-specs/qwen.json

@@ -1,6 +1,22 @@
 {
   "description": "阿里通义千问 Qwen 系列模型规格。参数来源：config-center compute/providers/dashscope.json。",
   "specs": [
+    {
+      "id": "qwen3.8-max-preview",
+      "displayName": "Qwen3.8 Max Preview",
+      "family": "qwen3.8",
+      "match": { "exact": ["qwen3.8-max-preview"], "patterns": ["qwen3.8-max-preview*"] },
+      "spec": {
+        "contextWindow": 1000000,
+        "maxOutputTokens": 65536,
+        "capabilities": ["chat", "reasoning", "code", "multilingual", "tool_use", "long_context", "agent", "vision"],
+        "serviceType": ["chat"],
+        "defaultTemperature": 0.6,
+        "defaultTopP": 0.95,
+        "supportsReasoning": true,
+        "description": "通义千问 Qwen3.8 Max 预览版，100 万上下文，支持视觉理解"
+      }
+    },
     {
       "id": "qwen3.7-max",
       "displayName": "Qwen3.7 Max",

compute/model-specs/xai.json

@@ -10,7 +10,7 @@
         "contextWindow": 500000,
         "maxOutputTokens": 128000,
         "capabilities": ["chat", "reasoning", "code", "tool_use", "vision", "long_context"],
-        "serviceType": ["chat"],
+        "serviceType": ["chat", "reasoning"],
         "supportsReasoning": true,
         "description": "xAI 最新旗舰模型，编程和 STEM 领域前沿"
       }

compute/model-specs/xiaomi.json

@@ -18,8 +18,7 @@
           "chat",
           "reasoning",
           "tool_use",
-          "code",
+          "code"
-          "vision"
         ],
         "serviceType": [
           "chat",
@@ -27,7 +26,7 @@
         ],
         "defaultTemperature": 1,
         "supportsReasoning": true,
-        "description": "小米 MiMo 旗舰推理模型，支持文本/图像/工具调用"
+        "description": "小米 MiMo 旗舰推理模型，支持文本/工具调用"
       }
     },
     {
@@ -78,8 +77,7 @@
           "chat",
           "reasoning",
           "tool_use",
-          "code",
+          "code"
-          "vision"
         ],
         "serviceType": [
           "chat",

compute/providers/_index.json

@@ -4,6 +4,7 @@
     "openai",
     "openai-codex",
     "anthropic",
+    "anthropic-claude",
     "deepseek",
     "dashscope",
     "volcengine",

compute/providers/anthropic-claude.json

@@ -0,0 +1,149 @@
+{
+  "id": "provider-anthropic-claude-plan-001",
+  "provider": "anthropic-claude",
+  "brandGroup": "anthropic",
+  "label": "Claude 订阅 (Claude Code)",
+  "baseUrl": "https://api.anthropic.com",
+  "apiFormat": "anthropic-messages",
+  "apiKeyRef": "anthropic-claude-oauth-token",
+  "apiKeyVerified": false,
+  "enabled": false,
+  "status": "unconfigured",
+  "priceCurrency": "USD",
+  "accessMode": "coding-plan",
+  "credentialSource": "claude-oauth",
+  "requiredClientVersion": "10.0.83",
+  "codingPlan": {
+    "quotas": {},
+    "usageTracking": {
+      "method": "none",
+      "consoleUrl": "https://claude.ai/settings/usage"
+    }
+  },
+  "services": [
+    "chat",
+    "reasoning"
+  ],
+  "models": [
+    {
+      "modelName": "claude-fable-5",
+      "displayName": "Claude Fable 5",
+      "serviceType": [
+        "chat"
+      ],
+      "description": "Claude 订阅额度计费，Anthropic 最高能力公开模型，面向长程智能体与高难推理，1M 上下文",
+      "contextWindow": 1000000,
+      "maxOutputTokens": 128000,
+      "capabilities": [
+        "chat",
+        "reasoning",
+        "code",
+        "vision",
+        "tool_use",
+        "agent",
+        "long_context"
+      ],
+      "extra": {
+        "adaptiveThinking": true,
+        "reasoning": {
+          "supportedEfforts": [
+            "low",
+            "medium",
+            "high",
+            "xhigh",
+            "max"
+          ],
+          "defaultEffort": "xhigh"
+        }
+      },
+      "source": "preset"
+    },
+    {
+      "modelName": "claude-opus-5",
+      "displayName": "Claude Opus 5",
+      "serviceType": [
+        "chat"
+      ],
+      "description": "Claude 订阅额度计费，Opus 系列当前旗舰，复杂智能体编码与长程任务首选，1M 上下文",
+      "contextWindow": 1000000,
+      "maxOutputTokens": 128000,
+      "capabilities": [
+        "chat",
+        "reasoning",
+        "code",
+        "vision",
+        "tool_use",
+        "agent",
+        "long_context"
+      ],
+      "extra": {
+        "adaptiveThinking": true,
+        "thinkingOnByDefault": true,
+        "reasoning": {
+          "supportedEfforts": [
+            "low",
+            "medium",
+            "high",
+            "xhigh",
+            "max"
+          ],
+          "defaultEffort": "xhigh"
+        }
+      },
+      "source": "preset"
+    },
+    {
+      "modelName": "claude-sonnet-5",
+      "displayName": "Claude Sonnet 5",
+      "serviceType": [
+        "chat"
+      ],
+      "description": "Claude 订阅额度计费，速度与智能平衡模型，支持长上下文智能体任务",
+      "contextWindow": 1000000,
+      "maxOutputTokens": 128000,
+      "capabilities": [
+        "chat",
+        "reasoning",
+        "code",
+        "vision",
+        "tool_use",
+        "agent",
+        "long_context"
+      ],
+      "extra": {
+        "adaptiveThinking": true,
+        "reasoning": {
+          "supportedEfforts": [
+            "low",
+            "medium",
+            "high",
+            "xhigh",
+            "max"
+          ],
+          "defaultEffort": "xhigh"
+        }
+      },
+      "source": "preset"
+    },
+    {
+      "modelName": "claude-haiku-4-5",
+      "displayName": "Claude Haiku 4.5",
+      "serviceType": [
+        "chat"
+      ],
+      "description": "Claude 订阅额度计费，最快模型，具备接近前沿模型的智能水平，200K 上下文",
+      "contextWindow": 200000,
+      "maxOutputTokens": 64000,
+      "capabilities": [
+        "chat",
+        "code",
+        "vision",
+        "tool_use"
+      ],
+      "source": "preset"
+    }
+  ],
+  "tombstones": [
+    "claude-opus-4-8"
+  ]
+}

compute/providers/anthropic.json

@@ -35,23 +35,40 @@
       "inputPrice": 10,
       "outputPrice": 50,
       "extra": {
+        "serverSideWebSearch": {
+          "enabled": true,
+          "dialect": "anthropic-messages",
+          "fallbackPriority": 1,
+          "searchRequestPriceUsd": 0.01,
+          "toolType": "web_search_20260318"
+        },
         "cachePricing": {
           "write5m": 12.5,
           "write1h": 20,
           "read": 1
         },
         "adaptiveThinking": true,
+        "reasoning": {
+          "supportedEfforts": [
+            "low",
+            "medium",
+            "high",
+            "xhigh",
+            "max"
+          ],
+          "defaultEffort": "xhigh"
+        },
         "samplingParametersDeprecated": true,
         "pricingNotes": "Prices are per 1M tokens. Full 1M context is billed at standard pricing."
       }
     },
     {
-      "modelName": "claude-opus-4-8",
+      "modelName": "claude-opus-5",
-      "displayName": "Claude Opus 4.8",
+      "displayName": "Claude Opus 5",
       "serviceType": [
         "chat"
       ],
-      "description": "Anthropic Opus 级复杂智能体编码与企业任务模型，1M 上下文",
+      "description": "Anthropic Opus 系列当前旗舰，复杂智能体编码与长程任务能力显著强于 Opus 4.8，价格不变，1M 上下文",
       "contextWindow": 1000000,
       "maxOutputTokens": 128000,
       "capabilities": [
@@ -66,13 +83,32 @@
       "inputPrice": 5,
       "outputPrice": 25,
       "extra": {
+        "serverSideWebSearch": {
+          "enabled": true,
+          "dialect": "anthropic-messages",
+          "fallbackPriority": 2,
+          "searchRequestPriceUsd": 0.01,
+          "toolType": "web_search_20260318"
+        },
         "cachePricing": {
           "write5m": 6.25,
           "write1h": 10,
           "read": 0.5
         },
+        "promptCacheMinTokens": 512,
         "adaptiveThinking": true,
-        "defaultEffort": "high",
+        "thinkingOnByDefault": true,
+        "reasoning": {
+          "supportedEfforts": [
+            "low",
+            "medium",
+            "high",
+            "xhigh",
+            "max"
+          ],
+          "defaultEffort": "xhigh"
+        },
+        "defaultEffort": "xhigh",
         "samplingParametersDeprecated": true,
         "pricingNotes": "Prices are per 1M tokens. Full 1M context is billed at standard pricing."
       }
@@ -100,6 +136,13 @@
       "inputPrice": 2,
       "outputPrice": 10,
       "extra": {
+        "serverSideWebSearch": {
+          "enabled": true,
+          "dialect": "anthropic-messages",
+          "fallbackPriority": 3,
+          "searchRequestPriceUsd": 0.01,
+          "toolType": "web_search_20260318"
+        },
         "cachePricing": {
           "write5m": 2.5,
           "write1h": 4,
@@ -110,67 +153,20 @@
         "standardInputPrice": 3,
         "standardOutputPrice": 15,
         "adaptiveThinking": true,
-        "defaultEffort": "high",
+        "reasoning": {
+          "supportedEfforts": [
+            "low",
+            "medium",
+            "high",
+            "xhigh",
+            "max"
+          ],
+          "defaultEffort": "xhigh"
+        },
+        "defaultEffort": "xhigh",
         "samplingParametersDeprecated": true
       }
     },
-    {
-      "modelName": "claude-opus-4-7",
-      "displayName": "Claude Opus 4.7",
-      "serviceType": [
-        "chat"
-      ],
-      "description": "Anthropic 当前最强通用模型，适合复杂推理和智能体编码任务",
-      "contextWindow": 1000000,
-      "maxOutputTokens": 128000,
-      "capabilities": [
-        "chat",
-        "reasoning",
-        "code",
-        "vision",
-        "tool_use"
-      ],
-      "inputPrice": 5,
-      "outputPrice": 25,
-      "extra": {
-        "cachePricing": {
-          "write5m": 6.25,
-          "write1h": 10,
-          "read": 0.5
-        },
-        "pricingNotes": "Prices are per 1M tokens. Opus 4.7 includes the full 1M context window at standard pricing."
-      }
-    },
-    {
-      "modelName": "claude-sonnet-4-6",
-      "displayName": "Claude Sonnet 4.6",
-      "serviceType": [
-        "chat",
-        "computer_use"
-      ],
-      "description": "Anthropic 高智能高速度模型，适合编码、工具使用和智能体任务",
-      "contextWindow": 1000000,
-      "maxOutputTokens": 64000,
-      "capabilities": [
-        "chat",
-        "reasoning",
-        "code",
-        "vision",
-        "tool_use",
-        "computer_use"
-      ],
-      "inputPrice": 3,
-      "outputPrice": 15,
-      "defaultTemperature": 1,
-      "extra": {
-        "cachePricing": {
-          "write5m": 3.75,
-          "write1h": 6,
-          "read": 0.3
-        },
-        "pricingNotes": "Prices are per 1M tokens. Sonnet 4.6 includes the full 1M context window at standard pricing."
-      }
-    },
     {
       "modelName": "claude-haiku-4-5",
       "displayName": "Claude Haiku 4.5",
@@ -200,5 +196,10 @@
         "pricingNotes": "Prices are per 1M tokens."
       }
     }
+  ],
+  "tombstones": [
+    "claude-opus-4-8",
+    "claude-opus-4-7",
+    "claude-sonnet-4-6"
   ]
 }

compute/providers/dashscope.json

@@ -33,6 +33,32 @@
     "qwen3-max-trans"
   ],
   "models": [
+    {
+      "modelName": "qwen3.8-max-preview",
+      "displayName": "阿里云 Qwen3.8-Max (Preview)",
+      "serviceType": [
+        "chat",
+        "vision"
+      ],
+      "description": "通义千问3.8 Max 预览版，面向复杂推理与长程 Agent 任务的旗舰模型，100万上下文，支持视觉理解、内置工具和 Function Calling",
+      "contextWindow": 1000000,
+      "maxOutputTokens": 65536,
+      "capabilities": [
+        "chat",
+        "reasoning",
+        "code",
+        "multilingual",
+        "long_context",
+        "tool_use",
+        "agent",
+        "vision"
+      ],
+      "inputPrice": 12,
+      "outputPrice": 36,
+      "defaultTemperature": 0.6,
+      "defaultTopP": 0.95,
+      "extra": {}
+    },
     {
       "modelName": "qwen3.7-max",
       "displayName": "阿里云 Qwen3.7-Max",

compute/providers/openai.json

@@ -46,6 +46,13 @@
       "inputPrice": 5,
       "outputPrice": 30,
       "extra": {
+        "serverSideWebSearch": {
+          "enabled": true,
+          "dialect": "openai-responses",
+          "fallbackPriority": 1,
+          "searchRequestPriceUsd": 0.01,
+          "toolType": "web_search"
+        },
         "cachedInputPrice": 0.5,
         "reasoningEffort": [
           "none",
@@ -79,6 +86,13 @@
       "inputPrice": 30,
       "outputPrice": 180,
       "extra": {
+        "serverSideWebSearch": {
+          "enabled": true,
+          "dialect": "openai-responses",
+          "fallbackPriority": 2,
+          "searchRequestPriceUsd": 0.01,
+          "toolType": "web_search"
+        },
         "responsesOnly": true,
         "reasoningEffort": [
           "medium",
@@ -110,6 +124,13 @@
       "inputPrice": 2.5,
       "outputPrice": 15,
       "extra": {
+        "serverSideWebSearch": {
+          "enabled": true,
+          "dialect": "openai-responses",
+          "fallbackPriority": 3,
+          "searchRequestPriceUsd": 0.01,
+          "toolType": "web_search"
+        },
         "cachedInputPrice": 0.25,
         "reasoningEffort": [
           "none",
@@ -140,6 +161,13 @@
       "inputPrice": 30,
       "outputPrice": 180,
       "extra": {
+        "serverSideWebSearch": {
+          "enabled": true,
+          "dialect": "openai-responses",
+          "fallbackPriority": 4,
+          "searchRequestPriceUsd": 0.01,
+          "toolType": "web_search"
+        },
         "responsesOnly": true,
         "reasoningEffort": [
           "medium",
@@ -172,6 +200,13 @@
       "inputPrice": 0.75,
       "outputPrice": 4.5,
       "extra": {
+        "serverSideWebSearch": {
+          "enabled": true,
+          "dialect": "openai-responses",
+          "fallbackPriority": 5,
+          "searchRequestPriceUsd": 0.01,
+          "toolType": "web_search"
+        },
         "cachedInputPrice": 0.075,
         "reasoningEffort": [
           "none",
@@ -201,6 +236,13 @@
       "inputPrice": 0.2,
       "outputPrice": 1.25,
       "extra": {
+        "serverSideWebSearch": {
+          "enabled": true,
+          "dialect": "openai-responses",
+          "fallbackPriority": 6,
+          "searchRequestPriceUsd": 0.01,
+          "toolType": "web_search"
+        },
         "cachedInputPrice": 0.02,
         "reasoningEffort": [
           "none",
@@ -395,7 +437,15 @@
       ],
       "inputPrice": 1.25,
       "outputPrice": 10,
-      "extra": {}
+      "extra": {
+        "serverSideWebSearch": {
+          "enabled": true,
+          "dialect": "openai-responses",
+          "fallbackPriority": 7,
+          "searchRequestPriceUsd": 0.01,
+          "toolType": "web_search"
+        }
+      }
     },
     {
       "modelName": "gpt-5-pro",
@@ -477,7 +527,15 @@
       "outputPrice": 8,
       "defaultTemperature": 1,
       "defaultTopP": 1,
-      "extra": {}
+      "extra": {
+        "serverSideWebSearch": {
+          "enabled": true,
+          "dialect": "openai-responses",
+          "fallbackPriority": 8,
+          "searchRequestPriceUsd": 0.01,
+          "toolType": "web_search"
+        }
+      }
     },
     {
       "modelName": "gpt-4.1-mini",
@@ -499,7 +557,15 @@
       "outputPrice": 1.6,
       "defaultTemperature": 1,
       "defaultTopP": 1,
-      "extra": {}
+      "extra": {
+        "serverSideWebSearch": {
+          "enabled": true,
+          "dialect": "openai-responses",
+          "fallbackPriority": 9,
+          "searchRequestPriceUsd": 0.01,
+          "toolType": "web_search"
+        }
+      }
     },
     {
       "modelName": "gpt-4.1-nano",
@@ -754,7 +820,15 @@
       ],
       "inputPrice": 1.1,
       "outputPrice": 4.4,
-      "extra": {}
+      "extra": {
+        "serverSideWebSearch": {
+          "enabled": true,
+          "dialect": "openai-responses",
+          "fallbackPriority": 10,
+          "searchRequestPriceUsd": 0.01,
+          "toolType": "web_search"
+        }
+      }
     }
   ],
   "tombstones": [

compute/providers/xiaomi.json

@@ -31,8 +31,7 @@
         "chat",
         "reasoning",
         "tool_use",
-        "code",
+        "code"
-        "vision"
       ],
       "extra": {}
     },

manifest.json

@@ -1,6 +1,6 @@
 {
   "version": "1.0.0",
-  "presetDataVersion": 80,
+  "presetDataVersion": 90,
-  "updatedAt": "2026-07-20",
+  "updatedAt": "2026-08-10",
   "description": "DesireCore 官方配置中心"
 }

schemas/provider.schema.json

@@ -83,8 +83,8 @@
     },
     "credentialSource": {
       "type": "string",
-      "enum": ["codex-cli"],
+      "enum": ["codex-cli", "claude-oauth"],
-      "description": "凭证托管来源。codex-cli：密钥由客户端本地 Codex CLI 凭证检测器托管（读取 ~/.codex/auth.json 自动刷新写回，fresh token 同步 secrets.json）。需 requiredClientVersion ≥ 引入此字段的客户端版本——老客户端不识别本字段，必须先发版铺开 compute.json 韧性（容忍未知字段）再推送本数据"
+      "description": "凭证托管来源。codex-cli：密钥由客户端本地 Codex CLI 凭证检测器托管（读取 ~/.codex/auth.json 自动刷新写回，fresh token 同步 secrets.json）。claude-oauth：Claude 订阅接入，由客户端 claude-auth 检测器托管——首选复用本机 Claude Code 登录（Agent SDK 自读，不写不刷新），或应用内 OAuth / setup-token 兜底；模型调用经 compat-proxy 内 claude-agent-sdk 后端履约，不直连。需 requiredClientVersion ≥ 引入此字段的客户端版本——老客户端不识别本字段，必须先发版铺开 compute.json 韧性（容忍未知字段）再推送本数据"
     },
     "codingPlan": {
       "type": "object",

schemas/smart-model-catalog.schema.json

@@ -0,0 +1,273 @@
+{
+  "$schema": "http://json-schema.org/draft-07/schema#",
+  "$id": "https://desirecore.net/schemas/smart-model-catalog.json",
+  "title": "SmartModelCatalog",
+  "description": "智能路由量级、Provider 和模型能力目录；Config Center、内置兜底与未来接口共用此契约。",
+  "type": "object",
+  "required": [
+    "version",
+    "updatedAt",
+    "source",
+    "tiers",
+    "providers"
+  ],
+  "properties": {
+    "$schema": {
+      "type": "string",
+      "description": "本数据对应的 JSON Schema 标识。"
+    },
+    "version": {
+      "type": "integer",
+      "minimum": 1,
+      "maximum": 1,
+      "description": "目录结构版本；当前客户端只接受 v1，不兼容的新语义必须递增版本并由新客户端接入。"
+    },
+    "updatedAt": {
+      "type": "string",
+      "format": "date-time",
+      "description": "本目录最近一次人工/自动核对时间。"
+    },
+    "source": {
+      "type": "string",
+      "enum": [
+        "builtin",
+        "config-center",
+        "provider-api"
+      ],
+      "description": "目录来源：config-center=配置中心下发，builtin=客户端离线兜底，provider-api=未来接口实时数据。"
+    },
+    "tiers": {
+      "type": "array",
+      "minItems": 3,
+      "maxItems": 3,
+      "items": {
+        "type": "object",
+        "required": [
+          "id",
+          "label",
+          "labelEn",
+          "description",
+          "defaultReasoning"
+        ],
+        "properties": {
+          "id": {
+            "type": "string",
+            "enum": [
+              "flagship",
+              "balanced",
+              "lightweight"
+            ],
+            "description": "智能路由量级：flagship=最高能力，balanced=质量与成本平衡，lightweight=最快最省。"
+          },
+          "label": {
+            "type": "string",
+            "minLength": 1,
+            "description": "中文量级标签。"
+          },
+          "labelEn": {
+            "type": "string",
+            "minLength": 1,
+            "description": "英文量级标签。"
+          },
+          "description": {
+            "type": "string",
+            "minLength": 1,
+            "description": "该量级的任务选择目标。"
+          },
+          "defaultReasoning": {
+            "type": "string",
+            "enum": [
+              "auto",
+              "off",
+              "minimal",
+              "low",
+              "medium",
+              "high",
+              "xhigh",
+              "max"
+            ],
+            "description": "该量级在模型支持时优先采用的默认思考深度。"
+          }
+        },
+        "additionalProperties": false
+      },
+      "description": "三档量级的自描述定义。"
+    },
+    "providers": {
+      "type": "array",
+      "minItems": 1,
+      "items": {
+        "type": "object",
+        "required": [
+          "providerId",
+          "provider",
+          "displayName",
+          "models"
+        ],
+        "properties": {
+          "providerId": {
+            "type": "string",
+            "minLength": 1,
+            "maxLength": 200,
+            "description": "compute.json 中的 Provider 唯一 ID。"
+          },
+          "provider": {
+            "type": "string",
+            "minLength": 1,
+            "maxLength": 120,
+            "description": "Provider 协议标识。"
+          },
+          "displayName": {
+            "type": "string",
+            "minLength": 1,
+            "maxLength": 200,
+            "description": "Provider 展示名称。"
+          },
+          "models": {
+            "type": "array",
+            "minItems": 1,
+            "items": {
+              "type": "object",
+              "required": [
+                "model",
+                "displayName",
+                "tier",
+                "routingPriority",
+                "serviceTypes",
+                "capabilities",
+                "contextWindow",
+                "maxOutputTokens",
+                "supportedReasoning",
+                "defaultReasoning",
+                "eligibleForAgent",
+                "defaultReference"
+              ],
+              "properties": {
+                "model": {
+                  "type": "string",
+                  "minLength": 1,
+                  "maxLength": 300,
+                  "description": "传给对应 Provider 的真实 modelName；禁止使用 smart 等虚拟名称。"
+                },
+                "displayName": {
+                  "type": "string",
+                  "minLength": 1,
+                  "maxLength": 200,
+                  "description": "客户端展示名称。"
+                },
+                "tier": {
+                  "type": "string",
+                  "enum": [
+                    "flagship",
+                    "balanced",
+                    "lightweight"
+                  ],
+                  "description": "此模型在智能路由中的能力量级。"
+                },
+                "routingPriority": {
+                  "type": "integer",
+                  "minimum": 0,
+                  "maximum": 10000,
+                  "description": "同一量级内的稳定选择顺序，数值越小越优先；凭据与硬能力校验仍先于此排序。"
+                },
+                "serviceTypes": {
+                  "type": "array",
+                  "minItems": 1,
+                  "maxItems": 16,
+                  "uniqueItems": true,
+                  "items": {
+                    "type": "string",
+                    "minLength": 1,
+                    "maxLength": 64,
+                    "pattern": "^[a-z0-9_]+$",
+                    "description": "可用于任务硬约束匹配的规范能力标签，例如 vision、code、tool_use、long_context。"
+                  },
+                  "description": "模型服务类型快照，例如 chat、reasoning、image_gen、video_gen。"
+                },
+                "capabilities": {
+                  "type": "array",
+                  "maxItems": 64,
+                  "uniqueItems": true,
+                  "items": {
+                    "type": "string",
+                    "minLength": 1,
+                    "maxLength": 64,
+                    "pattern": "^[a-z0-9_]+$",
+                    "description": "可用于任务硬约束匹配的规范能力标签，例如 vision、code、tool_use、long_context。"
+                  },
+                  "description": "模型能力快照；任务 requiredCapabilities 必须全部命中。"
+                },
+                "contextWindow": {
+                  "type": [
+                    "integer",
+                    "null"
+                  ],
+                  "minimum": 1,
+                  "maximum": 10000000,
+                  "description": "上下文窗口 token 数；非文本模型或未知时为 null。"
+                },
+                "maxOutputTokens": {
+                  "type": [
+                    "integer",
+                    "null"
+                  ],
+                  "minimum": 1,
+                  "maximum": 1000000,
+                  "description": "最大输出 token 数；未知或不适用时为 null。"
+                },
+                "supportedReasoning": {
+                  "type": "array",
+                  "minItems": 1,
+                  "maxItems": 8,
+                  "uniqueItems": true,
+                  "items": {
+                    "type": "string",
+                    "enum": [
+                      "auto",
+                      "off",
+                      "minimal",
+                      "low",
+                      "medium",
+                      "high",
+                      "xhigh",
+                      "max"
+                    ],
+                    "description": "本次 Run 或当前 assignment 的思考深度；最终值仍受目标 Provider × Model 能力矩阵约束。"
+                  },
+                  "description": "该 Provider × Model 接入面声明的可用 reasoning 档位。"
+                },
+                "defaultReasoning": {
+                  "type": "string",
+                  "enum": [
+                    "auto",
+                    "off",
+                    "minimal",
+                    "low",
+                    "medium",
+                    "high",
+                    "xhigh",
+                    "max"
+                  ],
+                  "description": "未显式指定时的模型参考 reasoning；路由还会结合量级默认值并做能力协商。"
+                },
+                "eligibleForAgent": {
+                  "type": "boolean",
+                  "description": "是否可作为 Agent 主模型。图像/视频生成、ASR/TTS 等专用模型为 false。"
+                },
+                "defaultReference": {
+                  "type": "boolean",
+                  "description": "是否为 Provider/defaultServiceMap 给出的默认参考。它只影响同级排序，不是强制答案。"
+                }
+              },
+              "additionalProperties": false
+            },
+            "description": "该接入面当前已核对的模型目录。"
+          }
+        },
+        "additionalProperties": false
+      },
+      "description": "参与智能路由和能力匹配的 Provider 目录。"
+    }
+  },
+  "additionalProperties": false
+}

scripts/validate.mjs

@@ -46,6 +46,7 @@ function pickSchemaKey(absPath) {
   if (rel === 'manifest.json') return 'manifest'
   if (rel === 'compute/pricing.json') return 'pricing'
   if (rel === 'compute/service-map.json') return 'service-map'
+  if (rel === 'compute/smart-routing/model-catalog.json') return 'smart-model-catalog'
   if (rel === 'compute/providers/_index.json') return 'providers-index'
   if (rel === 'runtimes/recommended.json') return 'runtime-recommended'
   if (rel === 'runtimes/versions-fallback.json') return 'runtime-versions-fallback'