jinzd-ai-cli 0.4.268 → 0.4.269

package/dist/{server-NJ3ZNPQW.js → server-WTJSM2MT.js}

@@ -22,15 +22,19 @@ import {
   buildInitPrompt,
   buildSessionCostReport,
   costSourceNote,
+  decideModel,
   estimateConversationTokens,
   evaluateContextPressure,
+  filterMemoryEntries,
   loadDevState,
+  parseMemoryShowFilters,
   persistToolRound,
   readGitDiff,
   scanDirTree,
   scanProject,
-  setupProxy
-} from "./chunk-VNCBY4KY.js";
+  setupProxy,
+  stripRoutingTags
+} from "./chunk-S3F5LOV3.js";
 import {
   CostTracker,
   computeCost,
@@ -75,7 +79,7 @@ import {
   spawnAgentContext,
   truncateOutput,
   undoStack
-} from "./chunk-74BK22BM.js";
+} from "./chunk-DJNGUPM3.js";
 import {
   addMemoryEntry,
   clearMemoryEntries,
@@ -85,11 +89,12 @@ import {
   formatMemoryForPrompt,
   isMemoryExpired,
   listMemoryEntries,
+  listOrphanedProjectMemories,
   memoryStorePath,
   parseWritableScope,
   updateMemoryApproval,
   updateMemoryEntry
-} from "./chunk-WNG3YVZX.js";
+} from "./chunk-QPPUYCRF.js";
 import "./chunk-GLSR2NR3.js";
 import "./chunk-UUSRWSSX.js";
 import "./chunk-CKH4KQ4E.js";
@@ -103,13 +108,13 @@ import {
   buildDoctorReport,
   formatDoctorReport,
   loadContextFiles
-} from "./chunk-YLSM7RF6.js";
+} from "./chunk-6PROC2JO.js";
 import {
   ProviderRegistry
 } from "./chunk-AUEX6RNY.js";
 import {
   ConfigManager
-} from "./chunk-VGVKJTHF.js";
+} from "./chunk-5IUFUCIT.js";
 import {
   describePlugin,
   getActivePluginAssets,
@@ -124,7 +129,7 @@ import {
   runTool,
   setPluginEnabled,
   trustPlugin
-} from "./chunk-2FQGMR34.js";
+} from "./chunk-5PRWBJCI.js";
 import {
   CONTENT_ONLY_STREAM_REMINDER,
   TEE_FINAL_USER_NUDGE,
@@ -149,7 +154,7 @@ import {
   getGitRoot
 } from "./chunk-HOSJZMQS.js";
 import "./chunk-4BKXL7SM.js";
-import "./chunk-WWLXW4SY.js";
+import "./chunk-V6KQ67VF.js";
 import {
   AGENTIC_BEHAVIOR_GUIDELINE,
   AUTHOR,
@@ -165,7 +170,7 @@ import {
   SKILLS_DIR_NAME,
   VERSION,
   buildUserIdentityPrompt
-} from "./chunk-UT2WAJJG.js";
+} from "./chunk-3TOMHXDW.js";
 import "./chunk-6RPGQ5CG.js";
 import "./chunk-KHYD3WXE.js";
 import "./chunk-NYNSYLFY.js";
@@ -845,7 +850,7 @@ Cache:  write=${cacheCreate} read=${cacheRead}` : "";
 Cost:   ${formatCost(cost)}` : "";
   let memoryLine = "";
   try {
-    const { countPendingMemories } = await import("./persistent-memory-XVSSR277.js");
+    const { countPendingMemories } = await import("./persistent-memory-H4DNDIOQ.js");
     const pending = countPendingMemories(ctx.config.getConfigDir());
     if (pending > 0) memoryLine = `
 Memory: \u26A0 ${pending} pending approval \u2014 see the Memory panel or /memory`;
@@ -1658,7 +1663,7 @@ 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-33X3D7PV.js")).executeTests;
+    const runTests = ctx.runTests ?? (await import("./run-tests-FZSYHWYX.js")).executeTests;
     const report = await runTests(testArgs);
     ctx.send({ type: "info", message: report });
   } catch (err) {
@@ -1916,7 +1921,7 @@ async function handleMemory(args, ctx) {
     ctx.handleMemoryManage(sub, args[1], sub === "expire" ? args[2] : void 0);
   } else if (sub === "export") {
     if (args[1] === "md") {
-      const { exportMemoryEntries: exportMemoryEntries2 } = await import("./persistent-memory-XVSSR277.js");
+      const { exportMemoryEntries: exportMemoryEntries2 } = await import("./persistent-memory-H4DNDIOQ.js");
       ctx.send({
         type: "export_data",
         format: "md",
@@ -1936,7 +1941,8 @@ async function handleMemory(args, ctx) {
       });
     }
   } else {
-    ctx.memoryShow();
+    const isShow = sub === "show" || sub === "view" || sub === "list";
+    ctx.memoryShow(isShow ? args.slice(1) : args);
   }
 }
 async function handleAddDir(args, ctx) {
@@ -2386,17 +2392,18 @@ var SessionHandler = class {
   sessionAccruedCostUsd = 0;
   sessionCostIncomplete = false;
   /** Accumulate a TokenUsage (with optional cache fields) into sessionTokenUsage. */
-  addWebSessionUsage(u) {
+  addWebSessionUsage(u, modelOverride) {
+    const usageModel = modelOverride ?? this.currentModel;
     this.sessionTokenUsage.inputTokens += u.inputTokens;
     this.sessionTokenUsage.outputTokens += u.outputTokens;
     this.sessionTokenUsage.cacheCreationTokens += u.cacheCreationTokens ?? 0;
     this.sessionTokenUsage.cacheReadTokens += u.cacheReadTokens ?? 0;
-    const priced = this.costTracker.addCost(this.currentProvider, this.currentModel, u);
+    const priced = this.costTracker.addCost(this.currentProvider, usageModel, u);
     if (priced === null) this.sessionCostIncomplete = true;
     else this.sessionAccruedCostUsd += priced;
     const totalTokens = u.inputTokens + u.outputTokens + (u.cacheCreationTokens ?? 0) + (u.cacheReadTokens ?? 0);
-    if (totalTokens > 0 && !getPricing(this.currentProvider, this.currentModel)) {
-      const key = `${this.currentProvider}/${this.currentModel}`;
+    if (totalTokens > 0 && !getPricing(this.currentProvider, usageModel)) {
+      const key = `${this.currentProvider}/${usageModel}`;
       if (!this.unknownPricingWarnings.has(key)) {
         this.unknownPricingWarnings.add(key);
         this.send({ type: "info", message: `Cost notice: pricing is unknown for ${key}. Tokens are tracked, but cost is excluded from totals. Run /cost or aicli usage to review.` });
@@ -2844,6 +2851,16 @@ var SessionHandler = class {
       } else {
         msgContent = content;
       }
+      const routingDecision = this.computeRoutingDecision(typeof msgContent === "string" ? msgContent : content);
+      if (typeof msgContent === "string") {
+        msgContent = stripRoutingTags(msgContent);
+      } else if (msgContent.length > 0 && msgContent[msgContent.length - 1]?.type === "text") {
+        const last = msgContent[msgContent.length - 1];
+        msgContent = [...msgContent.slice(0, -1), { type: "text", text: stripRoutingTags(last.text) }];
+      }
+      if (routingDecision.overridden) {
+        this.send({ type: "info", message: `\u2192 Routed to ${routingDecision.model} (${routingDecision.reason})` });
+      }
       const userMessage = { role: "user", content: msgContent, timestamp: /* @__PURE__ */ new Date() };
       session.addMessage(userMessage);
       const provider = this.providers.get(this.currentProvider);
@@ -2855,11 +2872,13 @@ var SessionHandler = class {
       const hasToolSupport = typeof provider.chatWithTools === "function";
       const { toolDefs, mcpBudgetNote } = hasToolSupport ? this.getFilteredToolDefs() : { toolDefs: [], mcpBudgetNote: null };
       if (hasToolSupport && toolDefs.length > 0) {
-        await this.handleChatWithTools(provider, session.messages, toolDefs, mcpBudgetNote);
+        await this.handleChatWithTools(provider, session.messages, toolDefs, mcpBudgetNote, routingDecision.model);
       } else {
-        await this.handleChatSimple(provider, session.messages);
+        await this.handleChatSimple(provider, session.messages, routingDecision.model);
       }
       await this.checkContextPressure();
+      const budgetWarning = this.costTracker.checkBudget(this.config.get("monthlyBudget"));
+      if (budgetWarning) this.send({ type: "info", message: `\u26A0 ${budgetWarning}` });
     } catch (err) {
       const parts = [];
       let current = err;
@@ -2878,8 +2897,9 @@ var SessionHandler = class {
       this.sendSessionList();
     }
   }
-  async handleChatSimple(provider, messages) {
+  async handleChatSimple(provider, messages, modelOverride) {
     const session = this.sessions.current;
+    const effectiveModel = modelOverride ?? this.currentModel;
     const modelParams = this.getModelParams();
     const ac = new AbortController();
     this.abortController = ac;
@@ -2887,7 +2907,7 @@ var SessionHandler = class {
       const { stable: simplStable, volatile: simplVolatile } = this.buildSystemPrompt();
       const stream = provider.chatStream({
         messages,
-        model: this.currentModel,
+        model: effectiveModel,
         systemPrompt: simplStable,
         systemPromptVolatile: simplVolatile,
         stream: true,
@@ -2906,7 +2926,7 @@ var SessionHandler = class {
         if (chunk.done) {
           this.send({ type: "response_done", content: fullContent, usage: chunk.usage });
           if (chunk.usage) {
-            this.addWebSessionUsage(chunk.usage);
+            this.addWebSessionUsage(chunk.usage, effectiveModel);
             session.addTokenUsage(chunk.usage);
           }
           if (chunk.reasoningContent) simpleReasoning = chunk.reasoningContent;
@@ -2931,8 +2951,9 @@ var SessionHandler = class {
       this.abortController = null;
     }
   }
-  async handleChatWithTools(provider, messages, toolDefs, mcpBudgetNote) {
+  async handleChatWithTools(provider, messages, toolDefs, mcpBudgetNote, modelOverride) {
     const session = this.sessions.current;
+    const effectiveModel = modelOverride ?? this.currentModel;
     const apiMessages = [...messages];
     const maxToolRounds = this.config.get("maxToolRounds") ?? DEFAULT_MAX_TOOL_ROUNDS;
     const autoPauseIntervalRaw = this.config.get("autoPauseInterval");
@@ -2962,7 +2983,7 @@ ${mcpBudgetNote}` : "");
         callModel: async (_round, extraMessages) => {
           const chatRequest = {
             messages: apiMessages,
-            model: this.currentModel,
+            model: effectiveModel,
             systemPrompt,
             systemPromptVolatile,
             stream: false,
@@ -3007,7 +3028,7 @@ Details: ${errMsg.split("\n")[0]}
           const summaryResult = await provider.chatWithTools(
             {
               messages: apiMessages,
-              model: this.currentModel,
+              model: effectiveModel,
               systemPrompt,
               systemPromptVolatile,
               stream: false,
@@ -3023,7 +3044,7 @@ Details: ${errMsg.split("\n")[0]}
         executeTools: async (toolCalls) => {
           googleSearchContext.configManager = this.config;
           spawnAgentContext.provider = provider;
-          spawnAgentContext.model = this.currentModel;
+          spawnAgentContext.model = effectiveModel;
           spawnAgentContext.systemPrompt = systemPromptVolatile ? `${systemPrompt}
 
 ---
@@ -3397,6 +3418,24 @@ ${summaryContent}`,
    * 到达上下文压力阈值时自动压缩（与 REPL 同一判据，见 core/context-pressure.ts）。
    * 上下文窗口未知时不动——不基于猜测改用户的对话历史。
    */
+  /**
+   * 智能模型路由决策（审计附录 D-5）。候选限定在当前 provider 注册的模型内——
+   * 跨 provider 换模型会破坏 API key 与对话连续性。
+   */
+  computeRoutingDecision(userInput) {
+    let availableModels = [];
+    try {
+      availableModels = this.providers.get(this.currentProvider).info.models.map((m) => m.id);
+    } catch {
+      availableModels = [];
+    }
+    return decideModel({
+      userInput,
+      currentModel: this.currentModel,
+      routingConfig: this.config.get("routing"),
+      availableModels
+    });
+  }
   async checkContextPressure() {
     const contextWindow = this.getContextWindowSize();
     const session = this.sessions.current;
@@ -3559,18 +3598,27 @@ ${summaryContent}`,
       });
     }
   }
-  memoryShow() {
+  memoryShow(args = []) {
     const configDir = this.config.getConfigDir();
-    const entries = listMemoryEntries(configDir, { includeExpired: true, includeRejected: true });
-    const content = entries.length === 0 ? "(empty \u2014 no persistent memory entries yet)" : entries.map((entry) => {
+    const filters = parseMemoryShowFilters(args);
+    if (filters.invalidStale) {
+      this.send({ type: "error", message: "Usage: /memory show --stale <days> (positive number)" });
+      return;
+    }
+    const source = filters.orphaned ? listOrphanedProjectMemories(configDir) : listMemoryEntries(configDir, { includeExpired: true, includeRejected: true });
+    const { entries, applied } = filterMemoryEntries(source, filters);
+    const filterNote = applied.length > 0 ? ` (filtered: ${applied.join(", ")})` : "";
+    const content = entries.length === 0 ? applied.length > 0 ? `(no entries match [${applied.join(", ")}])` : "(empty \u2014 no persistent memory entries yet)" : `Governed persistent memory${filterNote}
+
+` + entries.map((entry) => {
       const status = isMemoryExpired(entry) ? "expired" : entry.approved ? "approved" : "pending";
-      const source = [
+      const source2 = [
         entry.source ?? "unknown",
         entry.sourceSession ? `session ${entry.sourceSession.slice(0, 8)}` : "",
         entry.sourceProject ? `project ${entry.sourceProject}` : ""
       ].filter(Boolean).join(" \xB7 ");
       return `## ${entry.id.slice(0, 8)} \xB7 ${status} \xB7 scope:${entry.scope} \xB7 sensitivity:${entry.sensitivity}
-Source: ${source}
+Source: ${source2}
 ${entry.content}`;
     }).join("\n\n");
     this.send({

package/dist/{server-3LERX42X.js → server-ZSTVYSYC.js}

@@ -1,8 +1,8 @@
 #!/usr/bin/env node
 import {
   ToolRegistry
-} from "./chunk-74BK22BM.js";
-import "./chunk-WNG3YVZX.js";
+} from "./chunk-DJNGUPM3.js";
+import "./chunk-QPPUYCRF.js";
 import "./chunk-GLSR2NR3.js";
 import "./chunk-UUSRWSSX.js";
 import "./chunk-CKH4KQ4E.js";
@@ -10,7 +10,7 @@ import "./chunk-BXP6YZ2P.js";
 import "./chunk-5C76XDDF.js";
 import {
   runTool
-} from "./chunk-2FQGMR34.js";
+} from "./chunk-5PRWBJCI.js";
 import {
   getDangerLevel,
   schemaToJsonSchema
@@ -18,10 +18,10 @@ import {
 import "./chunk-TU3L3PW7.js";
 import "./chunk-HOSJZMQS.js";
 import "./chunk-4BKXL7SM.js";
-import "./chunk-WWLXW4SY.js";
+import "./chunk-V6KQ67VF.js";
 import {
   VERSION
-} from "./chunk-UT2WAJJG.js";
+} from "./chunk-3TOMHXDW.js";
 import "./chunk-6RPGQ5CG.js";
 import "./chunk-KHYD3WXE.js";
 import "./chunk-NYNSYLFY.js";

package/dist/{task-orchestrator-JN2WFZSF.js → task-orchestrator-PNLI3V36.js}

@@ -3,8 +3,8 @@ import {
   ToolRegistry,
   googleSearchContext,
   truncateOutput
-} from "./chunk-74BK22BM.js";
-import "./chunk-WNG3YVZX.js";
+} from "./chunk-DJNGUPM3.js";
+import "./chunk-QPPUYCRF.js";
 import "./chunk-GLSR2NR3.js";
 import "./chunk-UUSRWSSX.js";
 import "./chunk-CKH4KQ4E.js";
@@ -12,7 +12,7 @@ import "./chunk-BXP6YZ2P.js";
 import "./chunk-5C76XDDF.js";
 import {
   runTool
-} from "./chunk-2FQGMR34.js";
+} from "./chunk-5PRWBJCI.js";
 import {
   getDangerLevel,
   runLeanAgentLoop
@@ -20,10 +20,10 @@ import {
 import "./chunk-TU3L3PW7.js";
 import "./chunk-HOSJZMQS.js";
 import "./chunk-4BKXL7SM.js";
-import "./chunk-WWLXW4SY.js";
+import "./chunk-V6KQ67VF.js";
 import {
   SUBAGENT_ALLOWED_TOOLS
-} from "./chunk-UT2WAJJG.js";
+} from "./chunk-3TOMHXDW.js";
 import "./chunk-6RPGQ5CG.js";
 import "./chunk-KHYD3WXE.js";
 import "./chunk-NYNSYLFY.js";

package/dist/{usage-IFTNXTWC.js → usage-GVFUVYUG.js}

@@ -6,9 +6,9 @@ import {
 } from "./chunk-R7GERZDA.js";
 import {
   ConfigManager
-} from "./chunk-VGVKJTHF.js";
+} from "./chunk-5IUFUCIT.js";
 import "./chunk-TU3L3PW7.js";
-import "./chunk-UT2WAJJG.js";
+import "./chunk-3TOMHXDW.js";
 import "./chunk-IW3Q7AE5.js";
 
 // src/cli/usage.ts

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "jinzd-ai-cli",
-  "version": "0.4.268",
+  "version": "0.4.269",
   "description": "Cross-platform REPL-style AI CLI with multi-provider support",
   "type": "module",
   "main": "./dist/index.js",