jinzd-ai-cli 0.4.324 → 0.4.326

package/README.md

@@ -129,7 +129,7 @@ aicli user delete x          # Delete user
 
 | Provider | Models | Get API Key |
 |----------|--------|-------------|
-| **Claude** | Opus 4, Sonnet 4, Haiku 4 | [console.anthropic.com](https://console.anthropic.com) |
+| **Claude** | Fable 5.1, Opus 5.5, Sonnet 5.5, Haiku 5.5 (+ legacy Opus 4.6, Haiku 4.5) | [console.anthropic.com](https://console.anthropic.com) |
 | **Gemini** | 2.5 Pro, 2.5 Flash | [aistudio.google.com](https://aistudio.google.com) |
 | **DeepSeek** | deepseek-flash (default, multimodal), deepseek-v4-pro | [platform.deepseek.com](https://platform.deepseek.com) |
 | **OpenAI** | GPT-5.4, GPT-5, GPT-4.1, o3, o4-mini | [platform.openai.com](https://platform.openai.com) |

package/README.zh-CN.md

@@ -117,7 +117,7 @@ aicli user delete x          # 删除用户
 
 | Provider | 模型 | 获取 API Key |
 |----------|------|-------------|
-| **Claude** | Opus 4, Sonnet 4, Haiku 4 | [console.anthropic.com](https://console.anthropic.com) |
+| **Claude** | Fable 5.1、Opus 5.5、Sonnet 5.5、Haiku 5.5（另保留旧版 Opus 4.6、Haiku 4.5） | [console.anthropic.com](https://console.anthropic.com) |
 | **Gemini** | 2.5 Pro, 2.5 Flash | [aistudio.google.com](https://aistudio.google.com) |
 | **DeepSeek** | deepseek-flash（默认，多模态）, deepseek-v4-pro | [platform.deepseek.com](https://platform.deepseek.com) |
 | **OpenAI** | GPT-5.4, GPT-5, GPT-4.1, o3, o4-mini | [platform.openai.com](https://platform.openai.com) |

package/dist/{batch-TXSH37GQ.js → batch-D3HG4I6J.js}

@@ -4,12 +4,12 @@ import {
 } from "./chunk-RS7SKYTS.js";
 import {
   ConfigManager
-} from "./chunk-2C7NMQBY.js";
+} from "./chunk-FBT5VPT4.js";
 import "./chunk-TU3L3PW7.js";
 import {
   atomicWriteFileSync
 } from "./chunk-BMLQPSHP.js";
-import "./chunk-IAI2FAX4.js";
+import "./chunk-LCI2GVAU.js";
 
 // src/cli/batch.ts
 import Anthropic from "@anthropic-ai/sdk";
@@ -90,7 +90,7 @@ function getClaudeClient(config) {
 }
 function getDefaultModel(config) {
   const models = config.get("defaultModels");
-  return models?.claude ?? "claude-sonnet-4-5-20250929";
+  return models?.claude ?? "claude-sonnet-5-5";
 }
 async function cmdSubmit(inputFile, options = {}) {
   const config = new ConfigManager();
@@ -110,7 +110,9 @@ async function cmdSubmit(inputFile, options = {}) {
     custom_id: r.customId,
     params: {
       model: r.model ?? defaultModel,
-      max_tokens: r.maxTokens ?? 4096,
+      // 默认模型 Sonnet 5.5 默认就在思考，思考计入 max_tokens；4096 容易被思考吃光、正文为空。
+      // 只按实际产出计费，上限放宽不增加成本（Batches 不受 SDK 非流式 21333 的客户端校验约束）
+      max_tokens: r.maxTokens ?? 16e3,
       messages: r.messages,
       ...r.system ? { system: r.system } : {}
     }

package/dist/{chunk-AYDP3HFZ.js → chunk-2IGHZDPR.js}

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
 import {
   TEST_TIMEOUT
-} from "./chunk-IAI2FAX4.js";
+} from "./chunk-LCI2GVAU.js";
 import {
   getExecutionConfigDir,
   getExecutionSignal

package/dist/{chunk-PIX4ZPFB.js → chunk-5CE64V7X.js}

@@ -13,6 +13,9 @@ import {
 import {
   atomicWriteFileSync
 } from "./chunk-BMLQPSHP.js";
+import {
+  DEFAULT_MAX_TOKENS
+} from "./chunk-LCI2GVAU.js";
 
 // src/providers/claude.ts
 import Anthropic from "@anthropic-ai/sdk";
@@ -41,6 +44,10 @@ var BaseProvider = class {
   get supportsReasoningEffort() {
     return false;
   }
+  /** 该模型能否关掉思考。默认能；Claude Opus 5.5 / Fable 5.x 关不掉（见 claude-models.ts） */
+  canDisableThinking(_model) {
+    return true;
+  }
   /**
    * 将 Message[] 转换为 OpenAI API 格式的消息数组。
    * content 为 string 时直接传递；为 MessageContentPart[] 时保留数组格式（vision 请求）。
@@ -118,6 +125,75 @@ function describeNetworkFailure(err, ctx = {}) {
   return { code, address, transient, message: parts.join(" ") };
 }
 
+// src/providers/claude-models.ts
+var DISABLE_BY_MODEL = [
+  [/^claude-sonnet-5-5/, { type: "between_tools" }],
+  [/^claude-sonnet-5(?!-\d)/, { type: "disabled" }],
+  [/^claude-opus-5(?!-\d)/, { type: "disabled" }],
+  [/^claude-haiku-5-5/, { type: "disabled" }]
+];
+function claudeThinkingStyle(model) {
+  if (/^claude-(?:opus|sonnet|haiku)-4-5|^claude-3/.test(model)) return "manual";
+  if (/^claude-(?:opus-4-[678]|sonnet-4-6)/.test(model)) return "adaptive-optin";
+  return "adaptive-default";
+}
+function claudeAcceptsTemperature(model) {
+  return /^claude-(?:opus|sonnet|haiku)-4-[56]|^claude-3/.test(model);
+}
+function claudeDisableThinking(model) {
+  if (claudeThinkingStyle(model) !== "adaptive-default") return null;
+  for (const [re, param] of DISABLE_BY_MODEL) if (re.test(model)) return param;
+  return null;
+}
+function claudeCanDisableThinking(model) {
+  return claudeThinkingStyle(model) !== "adaptive-default" || claudeDisableThinking(model) !== null;
+}
+var MANUAL_BUDGET = {
+  low: 4096,
+  high: 16e3,
+  max: 32e3
+};
+var THINKING_MAX_TOKENS_STREAM = 32e3;
+var THINKING_MAX_TOKENS_NONSTREAM = 16e3;
+var NONSTREAM_SDK_LIMIT = 21333;
+function buildClaudeRequestParams(request, streaming) {
+  const model = request.model;
+  const style = claudeThinkingStyle(model);
+  const wantThinking = request.thinking === true;
+  let thinking;
+  let outputConfig;
+  let maxTokens = request.maxTokens ?? DEFAULT_MAX_TOKENS;
+  const thinkingActive = wantThinking || style === "adaptive-default" && claudeDisableThinking(model) === null;
+  if (thinkingActive && maxTokens === DEFAULT_MAX_TOKENS) {
+    maxTokens = streaming ? THINKING_MAX_TOKENS_STREAM : THINKING_MAX_TOKENS_NONSTREAM;
+  }
+  if (wantThinking) {
+    if (style === "manual") {
+      let budget = Math.max(
+        request.thinkingBudget ?? (request.thinkingEffort ? MANUAL_BUDGET[request.thinkingEffort] : 1e4),
+        1024
+      );
+      if (budget >= maxTokens) maxTokens = budget + 4096;
+      if (!streaming && maxTokens > NONSTREAM_SDK_LIMIT) {
+        maxTokens = NONSTREAM_SDK_LIMIT;
+        budget = Math.max(1024, Math.min(budget, maxTokens - 4096));
+      }
+      thinking = { type: "enabled", budget_tokens: budget };
+    } else {
+      thinking = { type: "adaptive", display: "summarized" };
+      outputConfig = { effort: request.thinkingEffort ?? "high" };
+    }
+  } else if (style === "adaptive-default") {
+    thinking = claudeDisableThinking(model) ?? void 0;
+  }
+  const temperature = !wantThinking && claudeAcceptsTemperature(model) ? request.temperature : void 0;
+  const out = { max_tokens: maxTokens };
+  if (thinking) out.thinking = thinking;
+  if (outputConfig) out.output_config = outputConfig;
+  if (temperature !== void 0) out.temperature = temperature;
+  return out;
+}
+
 // src/providers/claude.ts
 var CACHE_MIN_SYSTEM_CHARS = 2e3;
 var ClaudeProvider = class extends BaseProvider {
@@ -127,30 +203,28 @@ var ClaudeProvider = class extends BaseProvider {
   info = {
     id: "claude",
     displayName: "Claude (Anthropic)",
-    defaultModel: "claude-sonnet-4-5-20250929",
+    // 2026-10-09 按官方 Models overview / Model deprecations 核实：当前一代 4 个，
+    // 上下文 1M、最大输出 128K；Sonnet 4.5 已弃用（2026-11-30 下线，官方推荐换 Sonnet 5.5），
+    // 从清单移除，旧 config 由 config-manager 自愈迁移。Opus 4.6 / Haiku 4.5 仍为 Active，留作旧版。
+    defaultModel: "claude-sonnet-5-5",
     apiKeyEnvVar: "AICLI_API_KEY_CLAUDE",
     requiresApiKey: true,
     models: [
-      {
-        id: "claude-opus-4-6",
-        displayName: "Claude Opus 4.6",
-        contextWindow: 2e5,
-        supportsStreaming: true
-      },
-      {
-        id: "claude-sonnet-4-5-20250929",
-        displayName: "Claude Sonnet 4.5",
-        contextWindow: 2e5,
-        supportsStreaming: true
-      },
-      {
-        id: "claude-haiku-4-5-20251001",
-        displayName: "Claude Haiku 4.5",
-        contextWindow: 2e5,
-        supportsStreaming: true
-      }
+      { id: "claude-fable-5-1", displayName: "Claude Fable 5.1", contextWindow: 1e6, supportsStreaming: true, supportsThinking: true },
+      { id: "claude-opus-5-5", displayName: "Claude Opus 5.5", contextWindow: 1e6, supportsStreaming: true, supportsThinking: true },
+      { id: "claude-sonnet-5-5", displayName: "Claude Sonnet 5.5", contextWindow: 1e6, supportsStreaming: true, supportsThinking: true },
+      { id: "claude-haiku-5-5", displayName: "Claude Haiku 5.5", contextWindow: 1e6, supportsStreaming: true, supportsThinking: true },
+      { id: "claude-opus-4-6", displayName: "Claude Opus 4.6 (legacy)", contextWindow: 1e6, supportsStreaming: true, supportsThinking: true },
+      { id: "claude-haiku-4-5-20251001", displayName: "Claude Haiku 4.5 (legacy)", contextWindow: 2e5, supportsStreaming: true, supportsThinking: true }
     ]
   };
+  /** 4.6 起用 `output_config.effort` 控制思考深浅；4.5 系列按档位换算成思考预算 */
+  get supportsReasoningEffort() {
+    return true;
+  }
+  canDisableThinking(model) {
+    return claudeCanDisableThinking(model);
+  }
   async initialize(apiKey, options) {
     const clientOptions = {
       apiKey,
@@ -258,24 +332,6 @@ var ClaudeProvider = class extends BaseProvider {
     }
     return usage;
   }
-  /**
-   * 构建 Extended Thinking 参数。
-   * - thinking 启用时 temperature 必须为 1 或不设置（Anthropic API 要求）
-   * - budget_tokens 最小 1024，默认 10000
-   */
-  buildThinkingParams(request) {
-    if (!request.thinking) {
-      return { temperature: request.temperature };
-    }
-    return {
-      thinking: {
-        type: "enabled",
-        budget_tokens: Math.max(request.thinkingBudget ?? 1e4, 1024)
-      },
-      // thinking 启用时 temperature 必须不设置（API 强制 temperature=1）
-      temperature: void 0
-    };
-  }
   /**
    * 从 response.content 中提取文本内容。
    * 当 thinking 启用时，将 ThinkingBlock 内容用 <think>...</think> 标签包裹，
@@ -286,7 +342,8 @@ var ClaudeProvider = class extends BaseProvider {
     const textParts = [];
     for (const block of blocks) {
       if (block.type === "thinking") {
-        thinkingParts.push(block.thinking);
+        const text = block.thinking;
+        if (text) thinkingParts.push(text);
       } else if (block.type === "text") {
         textParts.push(block.text);
       }
@@ -301,14 +358,11 @@ var ClaudeProvider = class extends BaseProvider {
   async chat(request) {
     try {
       const messages = this.toClaudeMessagesWithToolHistory(request.messages);
-      const { thinking, temperature } = this.buildThinkingParams(request);
       const response = await this.client.messages.create({
         model: request.model,
         messages,
         system: this.buildSystemParam(request.systemPrompt, request.systemPromptVolatile),
-        max_tokens: request.maxTokens ?? 8192,
-        temperature,
-        thinking
+        ...buildClaudeRequestParams(request, false)
         // 非流式路径必须收 signal，否则 Ctrl+C 期间请求不可中断（同 openai-compatible 的同类遗漏）
       }, { signal: request.signal });
       const content = this.extractContent(response.content);
@@ -324,33 +378,31 @@ var ClaudeProvider = class extends BaseProvider {
   async *chatStream(request) {
     try {
       const messages = this.toClaudeMessagesWithToolHistory(request.messages);
-      const { thinking, temperature } = this.buildThinkingParams(request);
       const stream = this.client.messages.stream({
         model: request.model,
         messages,
         system: this.buildSystemParam(request.systemPrompt, request.systemPromptVolatile),
-        max_tokens: request.maxTokens ?? 8192,
-        temperature,
-        thinking
+        ...buildClaudeRequestParams(request, true)
       }, { signal: request.signal });
-      let currentBlockType = null;
+      let thinkOpen = false;
       for await (const event of stream) {
-        if (event.type === "content_block_start") {
-          currentBlockType = event.content_block.type;
-          if (currentBlockType === "thinking") {
-            yield { delta: "<think>", done: false };
-          }
-        } else if (event.type === "content_block_delta") {
+        if (event.type === "content_block_delta") {
           if (event.delta.type === "text_delta") {
             yield { delta: event.delta.text, done: false };
           } else if (event.delta.type === "thinking_delta") {
-            yield { delta: event.delta.thinking, done: false };
+            const text = event.delta.thinking;
+            if (!text) continue;
+            if (!thinkOpen) {
+              yield { delta: "<think>", done: false };
+              thinkOpen = true;
+            }
+            yield { delta: text, done: false };
           }
         } else if (event.type === "content_block_stop") {
-          if (currentBlockType === "thinking") {
+          if (thinkOpen) {
             yield { delta: "</think>", done: false };
+            thinkOpen = false;
           }
-          currentBlockType = null;
         } else if (event.type === "message_stop") {
           yield { delta: "", done: true };
         }
@@ -380,15 +432,12 @@ var ClaudeProvider = class extends BaseProvider {
       const baseMessages = this.toClaudeMessagesWithToolHistory(request.messages);
       const extraMessages = request._extraMessages ?? [];
       const allMessages = [...baseMessages, ...extraMessages];
-      const { thinking, temperature } = this.buildThinkingParams(request);
       const response = await this.client.messages.create({
         model: request.model,
         messages: allMessages,
         tools: anthropicTools,
         system: this.buildSystemParam(request.systemPrompt, request.systemPromptVolatile),
-        max_tokens: request.maxTokens ?? 8192,
-        temperature,
-        thinking
+        ...buildClaudeRequestParams(request, false)
         // 轮次耗尽的总结轮对所有 provider 都走这条非流式路径——漏 signal 则 Ctrl+C 无效
       }, { signal: request.signal });
       const usage = this.extractUsage(response.usage);
@@ -434,7 +483,6 @@ var ClaudeProvider = class extends BaseProvider {
     const baseMessages = this.toClaudeMessagesWithToolHistory(request.messages);
     const extraMessages = request._extraMessages ?? [];
     const allMessages = [...baseMessages, ...extraMessages];
-    const { thinking, temperature } = this.buildThinkingParams(request);
     let doneEmitted = false;
     const rawContentBlocks = [];
     try {
@@ -443,12 +491,11 @@ var ClaudeProvider = class extends BaseProvider {
         messages: allMessages,
         tools: anthropicTools,
         system: this.buildSystemParam(request.systemPrompt, request.systemPromptVolatile),
-        max_tokens: request.maxTokens ?? 8192,
-        temperature,
-        thinking
+        ...buildClaudeRequestParams(request, true)
       }, { signal: request.signal });
       let currentBlockType = null;
       let currentToolIndex = 0;
+      let thinkingShown = false;
       let currentBlockData = {};
       let startUsage = null;
       for await (const event of stream) {
@@ -462,7 +509,7 @@ var ClaudeProvider = class extends BaseProvider {
           currentBlockType = block.type;
           currentBlockData = { type: block.type };
           if (block.type === "thinking") {
-            yield { type: "thinking_start" };
+            thinkingShown = false;
             currentBlockData = { type: "thinking", thinking: "", signature: "" };
           } else if (block.type === "text") {
             currentBlockData = { type: "text", text: "" };
@@ -484,7 +531,13 @@ var ClaudeProvider = class extends BaseProvider {
             currentBlockData.text += event.delta.text;
           } else if (event.delta.type === "thinking_delta") {
             const thinkingDelta = event.delta.thinking;
-            yield { type: "thinking_delta", delta: thinkingDelta };
+            if (thinkingDelta) {
+              if (!thinkingShown) {
+                yield { type: "thinking_start" };
+                thinkingShown = true;
+              }
+              yield { type: "thinking_delta", delta: thinkingDelta };
+            }
             currentBlockData.thinking += thinkingDelta;
           } else if (event.delta.type === "input_json_delta") {
             const jsonDelta = event.delta.partial_json;
@@ -499,7 +552,7 @@ var ClaudeProvider = class extends BaseProvider {
           }
         } else if (event.type === "content_block_stop") {
           if (currentBlockType === "thinking") {
-            yield { type: "thinking_end" };
+            if (thinkingShown) yield { type: "thinking_end" };
           } else if (currentBlockType === "tool_use") {
             yield { type: "tool_call_end", index: currentToolIndex };
             currentToolIndex++;
@@ -662,10 +715,12 @@ var ClaudeProvider = class extends BaseProvider {
     try {
       const response = await this.client.models.list();
       const data = response.data ?? [];
+      const known = new Map(this.info.models.map((m) => [m.id, m.contextWindow]));
       return data.map((model) => ({
         id: model.id,
         displayName: model.id,
-        contextWindow: 2e5,
+        // Models API 返回 max_input_tokens；SDK 类型未必带，取不到时用清单值或保守的 200K
+        contextWindow: model.max_input_tokens ?? known.get(model.id) ?? 2e5,
         supportsStreaming: true,
         source: "live",
         available: true,
@@ -2052,13 +2107,15 @@ var OpenAIProvider = createStaticProviderClass(OPENAI_CONFIG);
 var OPENROUTER_CONFIG = {
   id: "openrouter",
   displayName: "OpenRouter",
-  defaultModel: "anthropic/claude-sonnet-4",
+  defaultModel: "anthropic/claude-sonnet-5.5",
   apiKeyEnvVar: "AICLI_API_KEY_OPENROUTER",
   baseUrl: "https://openrouter.ai/api/v1",
   models: [
-    { id: "anthropic/claude-opus-4", displayName: "Claude Opus 4", contextWindow: 2e5, supportsStreaming: true },
-    { id: "anthropic/claude-sonnet-4", displayName: "Claude Sonnet 4", contextWindow: 2e5, supportsStreaming: true },
-    { id: "anthropic/claude-haiku-4", displayName: "Claude Haiku 4", contextWindow: 2e5, supportsStreaming: true },
+    // 2026-10-09 对 /api/v1/models 实测：OpenRouter 的 Claude id 用点号（claude-sonnet-5.5），
+    // 此前清单里的 `anthropic/claude-opus-4` / `anthropic/claude-haiku-4` 在列表里不存在
+    { id: "anthropic/claude-opus-5.5", displayName: "Claude Opus 5.5", contextWindow: 1e6, supportsStreaming: true },
+    { id: "anthropic/claude-sonnet-5.5", displayName: "Claude Sonnet 5.5", contextWindow: 1e6, supportsStreaming: true },
+    { id: "anthropic/claude-haiku-5.5", displayName: "Claude Haiku 5.5", contextWindow: 1e6, supportsStreaming: true },
     { id: "openai/gpt-5.4", displayName: "GPT-5.4", contextWindow: 105e4, supportsStreaming: true },
     { id: "google/gemini-2.5-pro", displayName: "Gemini 2.5 Pro", contextWindow: 1048576, supportsStreaming: true },
     { id: "google/gemini-2.5-flash", displayName: "Gemini 2.5 Flash", contextWindow: 1048576, supportsStreaming: true },

package/dist/{chunk-MX3SLCEP.js → chunk-5EDSZ3JF.js}

@@ -2,7 +2,7 @@
 import {
   evaluateCiGate,
   normalizeCiThresholds
-} from "./chunk-CSXGOHLG.js";
+} from "./chunk-OIYQX2HM.js";
 import {
   buildReviewPrompt,
   buildSecurityReviewPrompt,
@@ -13,13 +13,13 @@ import {
 } from "./chunk-6YZEZFX5.js";
 import {
   ProviderRegistry
-} from "./chunk-PIX4ZPFB.js";
+} from "./chunk-5CE64V7X.js";
 import {
   ConfigManager
-} from "./chunk-2C7NMQBY.js";
+} from "./chunk-FBT5VPT4.js";
 import {
   VERSION
-} from "./chunk-IAI2FAX4.js";
+} from "./chunk-LCI2GVAU.js";
 
 // src/cli/ci.ts
 import { execFileSync } from "child_process";

package/dist/{chunk-JYDLKJEQ.js → chunk-7OY3WOJD.js}

@@ -4,7 +4,7 @@ import {
 } from "./chunk-QMFMST2T.js";
 import {
   TEST_TIMEOUT
-} from "./chunk-RFJV47EQ.js";
+} from "./chunk-MGFFZFOA.js";
 
 // src/tools/builtin/run-tests.ts
 import { spawn as spawn2 } from "child_process";

package/dist/{chunk-YCGWWBHF.js → chunk-BGFFD3IL.js}

@@ -13,7 +13,7 @@ import {
   MEMORY_FILE_NAME,
   MEMORY_MAX_CHARS,
   MEMORY_STORE_FILE_NAME
-} from "./chunk-IAI2FAX4.js";
+} from "./chunk-LCI2GVAU.js";
 
 // src/memory/persistent-memory.ts
 import { copyFileSync, existsSync, mkdirSync, readFileSync, statSync } from "fs";

package/dist/{chunk-NFPJ6KUU.js → chunk-BRZZ3QAI.js}

@@ -45,12 +45,23 @@ function getDeepSeekV4Pricing(model, at) {
 }
 var PRICING_TABLE = {
   // ── Anthropic Claude ──────────────────────────────────────────
-  "claude-opus-4-8": { input: 15, output: 75, cacheWrite: 18.75, cacheRead: 1.5 },
-  "claude-opus-4-7": { input: 15, output: 75, cacheWrite: 18.75, cacheRead: 1.5 },
-  "claude-opus-4-6": { input: 15, output: 75, cacheWrite: 18.75, cacheRead: 1.5 },
-  "claude-opus-4-5": { input: 15, output: 75, cacheWrite: 18.75, cacheRead: 1.5 },
+  // 2026-10-09 按官方 Pricing 页核实（USD/MTok）。cacheWrite 取 5 分钟缓存写入价（aicli 只用
+  // ephemeral 默认 5m）。缓存读取倍率不统一：Fable 5.1 0.025x、Opus 5.5 / Sonnet 5.5 0.05x、其余 0.1x。
+  // 此前 Opus 4.5–4.8 记成 $15/$75（Opus 4.1 的价），费用显示高了 3 倍。
+  "claude-fable-5-1": { input: 10, output: 50, cacheWrite: 12.5, cacheRead: 0.25 },
+  "claude-fable-5": { input: 10, output: 50, cacheWrite: 12.5, cacheRead: 1 },
+  "claude-opus-5-5": { input: 4, output: 20, cacheWrite: 5, cacheRead: 0.2 },
+  "claude-opus-5": { input: 5, output: 25, cacheWrite: 6.25, cacheRead: 0.5 },
+  "claude-opus-4-8": { input: 5, output: 25, cacheWrite: 6.25, cacheRead: 0.5 },
+  "claude-opus-4-7": { input: 5, output: 25, cacheWrite: 6.25, cacheRead: 0.5 },
+  "claude-opus-4-6": { input: 5, output: 25, cacheWrite: 6.25, cacheRead: 0.5 },
+  "claude-opus-4-5": { input: 5, output: 25, cacheWrite: 6.25, cacheRead: 0.5 },
+  "claude-sonnet-5-5": { input: 2, output: 10, cacheWrite: 2.5, cacheRead: 0.1 },
+  "claude-sonnet-5": { input: 2, output: 10, cacheWrite: 2.5, cacheRead: 0.2 },
   "claude-sonnet-4-6": { input: 3, output: 15, cacheWrite: 3.75, cacheRead: 0.3 },
   "claude-sonnet-4-5-20250929": { input: 3, output: 15, cacheWrite: 3.75, cacheRead: 0.3 },
+  // Haiku 5.5 按提示长度分档：此处为 ≤100K 档，超过的那档见 computeCost 的 CLAUDE_HAIKU_55_LONG
+  "claude-haiku-5-5": { input: 0.1, output: 0.5, cacheWrite: 0.125, cacheRead: 0.01 },
   "claude-haiku-4-5-20251001": { input: 1, output: 5, cacheWrite: 1.25, cacheRead: 0.1 },
   "claude-haiku-4-5": { input: 1, output: 5, cacheWrite: 1.25, cacheRead: 0.1 },
   // Legacy Claude 3.x families (prefix fallback handles minor date suffixes)
@@ -185,8 +196,15 @@ function getPricing(provider, model, at = /* @__PURE__ */ new Date()) {
   }
   return null;
 }
+var CLAUDE_HAIKU_55_LONG = { input: 0.5, output: 2.5, cacheWrite: 0.625, cacheRead: 0.05 };
+var CLAUDE_HAIKU_55_TIER_TOKENS = 1e5;
+function longPromptTier(model, usage) {
+  if (!model.toLowerCase().startsWith("claude-haiku-5-5")) return null;
+  const prompt = usage.inputTokens + (usage.cacheCreationTokens ?? 0) + (usage.cacheReadTokens ?? 0);
+  return prompt > CLAUDE_HAIKU_55_TIER_TOKENS ? CLAUDE_HAIKU_55_LONG : null;
+}
 function computeCost(provider, model, usage, at = /* @__PURE__ */ new Date()) {
-  const p = getPricing(provider, model, at);
+  const p = longPromptTier(model, usage) ?? getPricing(provider, model, at);
   if (!p) return null;
   const input = usage.inputTokens * p.input;
   const output = usage.outputTokens * p.output;

package/dist/{chunk-IOPPS6ST.js → chunk-DHJFMW3F.js}

@@ -2,7 +2,7 @@
 import {
   computeCost,
   formatCost
-} from "./chunk-NFPJ6KUU.js";
+} from "./chunk-BRZZ3QAI.js";
 import {
   clearRecentlyDeniedAutoActions,
   estimateTokens,
@@ -16,7 +16,7 @@ import {
   setPreferredAgentName,
   truncateForPersist,
   undoStack
-} from "./chunk-4YKQD5RW.js";
+} from "./chunk-NZGSCXA4.js";
 import {
   buildReviewPrompt,
   buildSecurityReviewPrompt
@@ -51,7 +51,7 @@ import {
   touchMemoryReferences,
   updateMemoryApproval,
   updateMemoryEntry
-} from "./chunk-YCGWWBHF.js";
+} from "./chunk-BGFFD3IL.js";
 import {
   theme
 } from "./chunk-5C76XDDF.js";
@@ -65,7 +65,7 @@ import {
 import {
   buildDoctorReport,
   formatDoctorReport
-} from "./chunk-TEJHNHNO.js";
+} from "./chunk-OW3PAC3K.js";
 import {
   describePlugin,
   getActivePluginAssets,
@@ -80,7 +80,7 @@ import {
   trustHook,
   trustPlugin,
   untrustHook
-} from "./chunk-IVVPYBGG.js";
+} from "./chunk-ERHFJIEP.js";
 import {
   isCleanDocumentBody,
   stripOuterCodeFence
@@ -109,7 +109,7 @@ import {
   MCP_TOOL_PREFIX,
   REPO_URL,
   VERSION
-} from "./chunk-IAI2FAX4.js";
+} from "./chunk-LCI2GVAU.js";
 
 // src/skills/manager.ts
 import { existsSync, readdirSync, mkdirSync, statSync } from "fs";
@@ -1348,6 +1348,7 @@ function applyThinkCommand(sub, state) {
     details: effort && !state.supportsEffort ? [`Note: ${state.provider} has no reasoning-effort levels \u2014 the level is ignored.`] : [],
     changed: true
   });
+  const alwaysThinksNote = state.alwaysThinks ? ["Note: this model always thinks and cannot turn it off \u2014 OFF only hides the thinking text; thinking tokens are still billed."] : [];
   const disable = () => ({
     runtimeThinking: false,
     // 刻意保留档位：重新开启时不该丢掉用户选过的档
@@ -1355,11 +1356,12 @@ function applyThinkCommand(sub, state) {
     effective: false,
     origin: "runtime override",
     summary: "Extended Thinking: OFF",
-    details: [],
+    details: alwaysThinksNote,
     changed: true
   });
   if (key === "status") {
     const details = current.effective ? [effortLine(state, null)] : [
+      ...alwaysThinksNote,
       "Enable with /think on | low | high | max, or set",
       "modelParams.<model>.thinking in ~/.aicli/config.json"
     ];
@@ -1506,7 +1508,7 @@ function parseRouteCommand(args) {
 }
 function routingEmptyRulesNote(routing) {
   if (routing && routing.rules.length > 0) return null;
-  return 'No rules configured yet. Add rules under `routing.rules` in ~/.aicli/config.json.\nExample: { match: { tag: "fast" }, model: "claude-haiku-4-5" }';
+  return 'No rules configured yet. Add rules under `routing.rules` in ~/.aicli/config.json.\nExample: { match: { tag: "fast" }, model: "claude-haiku-5-5" }';
 }
 function describeRoutingRule(rule, index) {
   const parts = [];
@@ -2168,7 +2170,8 @@ function createProviderCommands() {
           configuredThinking: ctx.getConfiguredThinking(),
           configuredEffort: ctx.getConfiguredThinkingEffort(),
           supportsEffort: ctx.getProviderSupportsReasoningEffort(),
-          provider: ctx.getCurrentProvider()
+          provider: ctx.getCurrentProvider(),
+          alwaysThinks: ctx.getModelCanDisableThinking ? !ctx.getModelCanDisableThinking() : false
         });
         if (outcome.changed) {
           ctx.setThinkingMode(outcome.runtimeThinking);
@@ -2442,7 +2445,7 @@ No tools match "${filter}".
           const { join: join6 } = await import("path");
           const { existsSync: existsSync6 } = await import("fs");
           const { getGitRoot } = await import("./git-context-EXOEHQSF.js");
-          const { MCP_PROJECT_CONFIG_NAME } = await import("./constants-7WO66DLZ.js");
+          const { MCP_PROJECT_CONFIG_NAME } = await import("./constants-XOD3IIW5.js");
           const { approveProject, hashMcpFile } = await import("./project-trust-KGG2BWKC.js");
           const cwd = process.cwd();
           const projectRoot = getGitRoot(cwd) ?? cwd;
@@ -3550,7 +3553,7 @@ function createReviewCommands() {
       usage: "/test [command|filter]",
       async execute(args, ctx) {
         try {
-          const { executeTests } = await import("./run-tests-WOPRQ64C.js");
+          const { executeTests } = await import("./run-tests-5HMWO676.js");
           const argStr = args.join(" ").trim();
           let testArgs = {};
           if (argStr) {

package/dist/{chunk-IVVPYBGG.js → chunk-ERHFJIEP.js}

@@ -8,7 +8,7 @@ import {
 import {
   CONFIG_DIR_NAME,
   PLUGINS_DIR_NAME
-} from "./chunk-IAI2FAX4.js";
+} from "./chunk-LCI2GVAU.js";
 
 // src/diagnostics/tool-stats.ts
 import { mkdirSync } from "fs";