jinzd-ai-cli 0.4.325 → 0.4.327

package/dist/electron-server.js

@@ -55,7 +55,7 @@ import {
   touchMemoryReferences,
   updateMemoryApproval,
   updateMemoryEntry
-} from "./chunk-PXJ3XVR5.js";
+} from "./chunk-CLGEU35T.js";
 import {
   DEFAULT_PATTERNS,
   redactJson,
@@ -81,7 +81,7 @@ import {
   recoverGbkMojibake,
   runTestsTool,
   shellInvocation
-} from "./chunk-TJFHBYQ3.js";
+} from "./chunk-C2625RDR.js";
 import {
   getCurrentSessionKey,
   getExecutionConfigDir,
@@ -123,7 +123,7 @@ import {
   SUBAGENT_MAX_ROUNDS_LIMIT,
   VERSION,
   buildUserIdentityPrompt
-} from "./chunk-GCYH5WOZ.js";
+} from "./chunk-I7J7E7OB.js";
 
 // src/web/access.ts
 import { randomBytes, timingSafeEqual } from "crypto";
@@ -693,7 +693,18 @@ ${err}`
     const DEEPSEEK_TARGET = "deepseek-flash";
     const LEGACY_KIMI = /* @__PURE__ */ new Set(["kimi-k2-0711-preview", "kimi-k2-turbo-preview"]);
     const KIMI_TARGET = "kimi-k2.6";
+    const LEGACY_CLAUDE = {
+      "claude-sonnet-4-5-20250929": "claude-sonnet-5-5",
+      "claude-sonnet-4-5": "claude-sonnet-5-5",
+      "claude-sonnet-4-20250514": "claude-sonnet-5-5",
+      "claude-3-7-sonnet-20250219": "claude-sonnet-5-5",
+      "claude-opus-4-20250514": "claude-opus-5-5",
+      "claude-opus-4-1-20250805": "claude-opus-5-5",
+      "claude-3-5-haiku-20241022": "claude-haiku-5-5",
+      "claude-3-haiku-20240307": "claude-haiku-5-5"
+    };
     let changed = false;
+    let claudeTarget;
     const dm = this.config.defaultModels;
     if (dm) {
       if (LEGACY_DEEPSEEK.has(dm.deepseek)) {
@@ -704,6 +715,17 @@ ${err}`
         dm.kimi = KIMI_TARGET;
         changed = true;
       }
+      const orTo = dm.openrouter ? { "anthropic/claude-opus-4": "anthropic/claude-opus-5.5", "anthropic/claude-haiku-4": "anthropic/claude-haiku-5.5" }[dm.openrouter] : void 0;
+      if (orTo) {
+        dm.openrouter = orTo;
+        changed = true;
+      }
+      const claudeTo = dm.claude ? LEGACY_CLAUDE[dm.claude] : void 0;
+      if (claudeTo) {
+        dm.claude = claudeTo;
+        claudeTarget = claudeTo;
+        changed = true;
+      }
     }
     const mp = this.config.modelParams;
     if (mp) {
@@ -725,10 +747,20 @@ ${err}`
           changed = true;
         }
       }
+      for (const [legacy, target] of Object.entries(LEGACY_CLAUDE)) {
+        if (legacy in mp) {
+          if (!(target in mp)) {
+            mp[target] = mp[legacy];
+          }
+          delete mp[legacy];
+          claudeTarget ??= target;
+          changed = true;
+        }
+      }
     }
     if (changed) {
       process.stderr.write(
-        `[ai-cli] Migrated retired model ids in config.json (deepseek \u2192 ${DEEPSEEK_TARGET}, kimi \u2192 ${KIMI_TARGET})
+        `[ai-cli] Migrated retired model ids in config.json (deepseek \u2192 ${DEEPSEEK_TARGET}, kimi \u2192 ${KIMI_TARGET}` + (claudeTarget ? `, claude \u2192 ${claudeTarget}` : "") + `)
 `
       );
     }
@@ -918,12 +950,23 @@ 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)
@@ -1058,8 +1101,15 @@ 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;
@@ -1366,6 +1416,10 @@ 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）。
@@ -1511,6 +1565,75 @@ 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 {
@@ -1520,30 +1643,28 @@ 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,
@@ -1651,24 +1772,6 @@ 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，
@@ -1679,7 +1782,8 @@ 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);
       }
@@ -1694,14 +1798,11 @@ 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);
@@ -1717,33 +1818,31 @@ 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 };
         }
@@ -1773,15 +1872,12 @@ 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);
@@ -1827,7 +1923,6 @@ 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 {
@@ -1836,12 +1931,11 @@ 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) {
@@ -1855,7 +1949,7 @@ 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: "" };
@@ -1877,7 +1971,13 @@ 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;
@@ -1892,7 +1992,7 @@ 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++;
@@ -2055,10 +2155,12 @@ 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,
@@ -4709,13 +4811,15 @@ 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 },
@@ -16517,6 +16621,7 @@ function applyThinkCommand(sub, state2) {
     details: effort && !state2.supportsEffort ? [`Note: ${state2.provider} has no reasoning-effort levels \u2014 the level is ignored.`] : [],
     changed: true
   });
+  const alwaysThinksNote = state2.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
@@ -16524,11 +16629,12 @@ function applyThinkCommand(sub, state2) {
     effective: false,
     origin: "runtime override",
     summary: "Extended Thinking: OFF",
-    details: [],
+    details: alwaysThinksNote,
     changed: true
   });
   if (key === "status") {
     const details = current.effective ? [effortLine(state2, null)] : [
+      ...alwaysThinksNote,
       "Enable with /think on | low | high | max, or set",
       "modelParams.<model>.thinking in ~/.aicli/config.json"
     ];
@@ -16675,7 +16781,7 @@ 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 = [];
@@ -16835,7 +16941,8 @@ async function handleThink(args, ctx) {
     configuredThinking: modelParams?.thinking,
     configuredEffort: modelParams?.thinkingEffort,
     supportsEffort: ctx.providers.has(ctx.currentProvider) ? ctx.providers.get(ctx.currentProvider).supportsReasoningEffort : false,
-    provider: ctx.currentProvider
+    provider: ctx.currentProvider,
+    alwaysThinks: ctx.providers.has(ctx.currentProvider) && ctx.providers.get(ctx.currentProvider).canDisableThinking?.(ctx.currentModel) === false
   });
   if (outcome.changed) {
     ctx.runtimeThinking = outcome.runtimeThinking;
@@ -17727,7 +17834,7 @@ Cache:  write=${cacheCreate} read=${cacheRead}` : "";
 Cost:   ${formatCost(cost)}` : "";
   let memoryLine = "";
   try {
-    const { countPendingMemories: countPendingMemories2 } = await import("./persistent-memory-OAFPFGKB.js");
+    const { countPendingMemories: countPendingMemories2 } = await import("./persistent-memory-Q34UVU4P.js");
     const pending = countPendingMemories2(ctx.config.getConfigDir());
     if (pending > 0) memoryLine = `
 Memory: \u26A0 ${pending} pending approval \u2014 see the Memory panel or /memory`;
@@ -18715,7 +18822,7 @@ async function handleTest(args, ctx) {
       const isCommand = argStr.includes(" ") || /^(mvn|gradle|npm|pytest|cargo|go)\b/.test(argStr);
       testArgs = isCommand ? { command: argStr } : { filter: argStr };
     }
-    const runTests = ctx.runTests ?? (await import("./run-tests-7AOMBU2D.js")).executeTests;
+    const runTests = ctx.runTests ?? (await import("./run-tests-6EB6ABYH.js")).executeTests;
     const report = await runTests(testArgs);
     ctx.send({ type: "info", message: report });
   } catch (err) {
@@ -19106,7 +19213,7 @@ async function handleMemory(args, ctx) {
   }
   if (intent.kind === "status") {
     const { getChatIndexStatus: getChatIndexStatus2 } = await import("./chat-index-RLVLQRRC.js");
-    const { listMemoryEntries: listMemoryEntries2, isMemoryExpired: isMemoryExpired2, formatMemoryForPrompt: formatMemoryForPrompt2 } = await import("./persistent-memory-OAFPFGKB.js");
+    const { listMemoryEntries: listMemoryEntries2, isMemoryExpired: isMemoryExpired2, formatMemoryForPrompt: formatMemoryForPrompt2 } = await import("./persistent-memory-Q34UVU4P.js");
     const memories = listMemoryEntries2(configDir, { includeExpired: true, includeRejected: true });
     const injected = formatMemoryForPrompt2(configDir, process.cwd(), ctx.config.get("memory")?.maxChars);
     const view = buildMemoryStatusView({
@@ -19129,7 +19236,7 @@ async function handleMemory(args, ctx) {
     return;
   }
   if (intent.kind === "recall") {
-    const { searchPersistentMemories: searchPersistentMemories2, touchMemoryReferences: touchMemoryReferences2 } = await import("./persistent-memory-OAFPFGKB.js");
+    const { searchPersistentMemories: searchPersistentMemories2, touchMemoryReferences: touchMemoryReferences2 } = await import("./persistent-memory-Q34UVU4P.js");
     const { searchChatMemory: searchChatMemory2 } = await import("./chat-index-RLVLQRRC.js");
     const persistentHits = searchPersistentMemories2(configDir, intent.query, process.cwd(), 3);
     if (persistentHits.length > 0) {
@@ -19250,7 +19357,7 @@ async function handleMemory(args, ctx) {
     ctx.handleMemoryManage(intent.action, args[1], intent.action === "expire" ? args[2] : void 0);
   } else if (intent.kind === "export") {
     if (args[1] === "md") {
-      const { exportMemoryEntries: exportMemoryEntries2 } = await import("./persistent-memory-OAFPFGKB.js");
+      const { exportMemoryEntries: exportMemoryEntries2 } = await import("./persistent-memory-Q34UVU4P.js");
       ctx.send({
         type: "export_data",
         format: "md",
@@ -20040,7 +20147,8 @@ 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);
@@ -20255,7 +20363,7 @@ No tools match "${filter}".
           const { join: join26 } = await import("path");
           const { existsSync: existsSync32 } = await import("fs");
           const { getGitRoot: getGitRoot2 } = await import("./git-context-VNCA2GU7.js");
-          const { MCP_PROJECT_CONFIG_NAME: MCP_PROJECT_CONFIG_NAME2 } = await import("./constants-W2CH36YR.js");
+          const { MCP_PROJECT_CONFIG_NAME: MCP_PROJECT_CONFIG_NAME2 } = await import("./constants-QFXJCLTM.js");
           const { approveProject, hashMcpFile } = await import("./project-trust-6CJB5ZMX.js");
           const cwd = process.cwd();
           const projectRoot = getGitRoot2(cwd) ?? cwd;
@@ -21174,7 +21282,7 @@ function createReviewCommands() {
       usage: "/test [command|filter]",
       async execute(args, ctx) {
         try {
-          const { executeTests } = await import("./run-tests-7AOMBU2D.js");
+          const { executeTests } = await import("./run-tests-6EB6ABYH.js");
           const argStr = args.join(" ").trim();
           let testArgs = {};
           if (argStr) {

package/dist/{hub-RHMDL74E.js → hub-OE7M7DRD.js}

@@ -158,7 +158,7 @@ ${content}`);
   }
 }
 async function runTaskMode(config, providers, configManager, topic) {
-  const { TaskOrchestrator } = await import("./task-orchestrator-7H7UYCU7.js");
+  const { TaskOrchestrator } = await import("./task-orchestrator-MWWHWMBT.js");
   const orchestrator = new TaskOrchestrator(config, providers, configManager);
   let interrupted = false;
   const onSigint = () => {