jinzd-ai-cli 0.4.101 → 0.4.103

package/dist/{batch-DL3IJ7JQ.js → batch-LNTG2IRQ.js}

@@ -1,9 +1,9 @@
 #!/usr/bin/env node
 import {
   ConfigManager
-} from "./chunk-GXR2VZDI.js";
+} from "./chunk-SN56X6RE.js";
 import "./chunk-2ZD3YTVM.js";
-import "./chunk-3MEH7D36.js";
+import "./chunk-JHPSWYO3.js";
 
 // src/cli/batch.ts
 import Anthropic from "@anthropic-ai/sdk";

package/dist/{chunk-5EQP4IAM.js → chunk-DGXUO7D4.js}

@@ -6,7 +6,7 @@ import { platform } from "os";
 import chalk from "chalk";
 
 // src/core/constants.ts
-var VERSION = "0.4.101";
+var VERSION = "0.4.103";
 var APP_NAME = "ai-cli";
 var CONFIG_DIR_NAME = ".aicli";
 var CONFIG_FILE_NAME = "config.json";

package/dist/{chunk-3MEH7D36.js → chunk-JHPSWYO3.js}

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
 
 // src/core/constants.ts
-var VERSION = "0.4.101";
+var VERSION = "0.4.103";
 var APP_NAME = "ai-cli";
 var CONFIG_DIR_NAME = ".aicli";
 var CONFIG_FILE_NAME = "config.json";

package/dist/{chunk-TFCBOPQV.js → chunk-OVYOYUP7.js}

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
 import {
   TEST_TIMEOUT
-} from "./chunk-3MEH7D36.js";
+} from "./chunk-JHPSWYO3.js";
 
 // src/tools/builtin/run-tests.ts
 import { execSync } from "child_process";

package/dist/{chunk-MBPA2FMK.js → chunk-RZWWODW7.js}

@@ -23,7 +23,7 @@ import {
 } from "./chunk-6VRJGH25.js";
 import {
   runTestsTool
-} from "./chunk-TFCBOPQV.js";
+} from "./chunk-OVYOYUP7.js";
 import {
   CONFIG_DIR_NAME,
   DEFAULT_MAX_TOOL_OUTPUT_CHARS_CAP,
@@ -31,7 +31,7 @@ import {
   SUBAGENT_ALLOWED_TOOLS,
   SUBAGENT_DEFAULT_MAX_ROUNDS,
   SUBAGENT_MAX_ROUNDS_LIMIT
-} from "./chunk-3MEH7D36.js";
+} from "./chunk-JHPSWYO3.js";
 
 // src/tools/types.ts
 function isFileWriteTool(name) {
@@ -1827,8 +1827,13 @@ var ToolExecutor = class {
 var writeFileTool = {
   definition: {
     name: "write_file",
-    description: `Write content to a file. Creates the file if it doesn't exist, overwrites (append=false) or appends (append=true). Automatically creates parent directories.
-Important: For long content (over 500 lines or 3000 chars), you MUST split into multiple calls using append=true, each no more than 300 lines, to avoid truncation. First call uses append=false (overwrite), subsequent calls use append=true.`,
+    description: `Write content to a file. Creates the file if it doesn't exist; overwrites by default or appends with append=true. Automatically creates parent directories.
+
+[CRITICAL: pick the right tool by content size]
+- Short content (under ~300 lines / ~3 KB): use write_file in ONE call. Do NOT chunk.
+- Long content (300+ lines, or 3 KB+, or any document you'd describe as "complete file" / "full directory" / "long article" / "exam paper"): you MUST use \`save_last_response\` instead. It generates and saves the file in a single streaming pass, avoiding the tool_call argument truncation that happens around 2 KB.
+
+Do NOT split a long document into many write_file(append=true) calls. That pattern is fragile, slow, and triggers many approval prompts. If you find yourself about to call write_file twice for the same path in one turn \u2014 STOP and use save_last_response.`,
     parameters: {
       path: {
         type: "string",
@@ -3207,16 +3212,22 @@ var lastResponseStore = { content: "" };
 var saveLastResponseTool = {
   definition: {
     name: "save_last_response",
-    description: `Tool for generating and saving large documents (exams, reports, long articles) to a file.
+    description: `Generate AND save a long document to a file in one streaming pass. Use this whenever the user asks you to PRODUCE and SAVE a complete document.
 
-[Usage] When the user requests generating and saving a large document, call this tool with the target path:
-  - The system will automatically initiate a streaming generation request, outputting content to both terminal and disk (tee mode)
-  - No need to output content first then call \u2014 the system handles "generate + save" in one step
-  - Only the file path is passed as a parameter, content is NOT passed via arguments, avoiding API argument truncation
+[When to use \u2014 pick this over write_file]
+Any of these triggers means use save_last_response, NOT write_file:
+  - The expected output is 300+ lines or 3 KB+ (exam papers, table-of-contents, full chapter, long report, structured outline, complete teaching plan, full code module)
+  - The user said "save as ...", "\u4FDD\u5B58\u4E3A ...", "write the full ...", "complete the entire ...", "list the whole directory"
+  - You'd otherwise need multiple write_file(append=true) calls to fit the content
 
-[Warning] Do NOT use write_file to save large content \u2014 write_file's content parameter gets truncated by the API for large documents, resulting in incomplete files.
+[How it works]
+  - You only pass the target file path as an argument
+  - The system runs a fresh streaming generation; content streams to the user AND to disk simultaneously (tee mode)
+  - No content is passed via tool arguments \u2192 not subject to the ~2 KB tool_call argument truncation that breaks write_file for large docs
 
-[Use cases] Exam papers (600-700 lines, 15-25KB), technical reports, long articles \u2014 any content over 2KB.`,
+[Anti-pattern \u2014 DO NOT DO]
+  - Do NOT call write_file with chunked append=true to build up a long file. Use save_last_response for the whole file in one pass.
+  - Do NOT first write a partial file with write_file, then "extend" it with edit_file inserts. Use save_last_response.`,
     parameters: {
       path: {
         type: "string",

package/dist/{chunk-GXR2VZDI.js → chunk-SN56X6RE.js}

@@ -8,7 +8,7 @@ import {
   CONFIG_FILE_NAME,
   HISTORY_DIR_NAME,
   PLUGINS_DIR_NAME
-} from "./chunk-3MEH7D36.js";
+} from "./chunk-JHPSWYO3.js";
 
 // src/config/config-manager.ts
 import { readFileSync, writeFileSync, existsSync, mkdirSync } from "fs";

package/dist/{chunk-VWZW5JWK.js → chunk-VOWVIR2U.js}

@@ -2,7 +2,7 @@
 import {
   schemaToJsonSchema,
   truncateForPersist
-} from "./chunk-MBPA2FMK.js";
+} from "./chunk-RZWWODW7.js";
 import {
   AuthError,
   ProviderError,
@@ -21,7 +21,7 @@ import {
   MCP_PROTOCOL_VERSION,
   MCP_TOOL_PREFIX,
   VERSION
-} from "./chunk-3MEH7D36.js";
+} from "./chunk-JHPSWYO3.js";
 
 // src/providers/claude.ts
 import Anthropic from "@anthropic-ai/sdk";

package/dist/electron-server.js

@@ -36,7 +36,7 @@ import {
   VERSION,
   buildUserIdentityPrompt,
   runTestsTool
-} from "./chunk-5EQP4IAM.js";
+} from "./chunk-DGXUO7D4.js";
 import {
   hasSemanticIndex,
   semanticSearch
@@ -56,7 +56,7 @@ import "./chunk-JV5N65KN.js";
 import express from "express";
 import { createServer } from "http";
 import { WebSocketServer } from "ws";
-import { join as join15, dirname as dirname4, resolve as resolve5, relative as relative3, sep as sep2 } from "path";
+import { join as join15, dirname as dirname5, resolve as resolve5, relative as relative3, sep as sep2 } from "path";
 import { existsSync as existsSync22, readFileSync as readFileSync15, readdirSync as readdirSync11, statSync as statSync9, realpathSync } from "fs";
 import { networkInterfaces } from "os";
 
@@ -5179,8 +5179,13 @@ var ToolExecutor = class {
 var writeFileTool = {
   definition: {
     name: "write_file",
-    description: `Write content to a file. Creates the file if it doesn't exist, overwrites (append=false) or appends (append=true). Automatically creates parent directories.
-Important: For long content (over 500 lines or 3000 chars), you MUST split into multiple calls using append=true, each no more than 300 lines, to avoid truncation. First call uses append=false (overwrite), subsequent calls use append=true.`,
+    description: `Write content to a file. Creates the file if it doesn't exist; overwrites by default or appends with append=true. Automatically creates parent directories.
+
+[CRITICAL: pick the right tool by content size]
+- Short content (under ~300 lines / ~3 KB): use write_file in ONE call. Do NOT chunk.
+- Long content (300+ lines, or 3 KB+, or any document you'd describe as "complete file" / "full directory" / "long article" / "exam paper"): you MUST use \`save_last_response\` instead. It generates and saves the file in a single streaming pass, avoiding the tool_call argument truncation that happens around 2 KB.
+
+Do NOT split a long document into many write_file(append=true) calls. That pattern is fragile, slow, and triggers many approval prompts. If you find yourself about to call write_file twice for the same path in one turn \u2014 STOP and use save_last_response.`,
     parameters: {
       path: {
         type: "string",
@@ -6559,16 +6564,22 @@ var lastResponseStore = { content: "" };
 var saveLastResponseTool = {
   definition: {
     name: "save_last_response",
-    description: `Tool for generating and saving large documents (exams, reports, long articles) to a file.
-
-[Usage] When the user requests generating and saving a large document, call this tool with the target path:
-  - The system will automatically initiate a streaming generation request, outputting content to both terminal and disk (tee mode)
-  - No need to output content first then call \u2014 the system handles "generate + save" in one step
-  - Only the file path is passed as a parameter, content is NOT passed via arguments, avoiding API argument truncation
-
-[Warning] Do NOT use write_file to save large content \u2014 write_file's content parameter gets truncated by the API for large documents, resulting in incomplete files.
-
-[Use cases] Exam papers (600-700 lines, 15-25KB), technical reports, long articles \u2014 any content over 2KB.`,
+    description: `Generate AND save a long document to a file in one streaming pass. Use this whenever the user asks you to PRODUCE and SAVE a complete document.
+
+[When to use \u2014 pick this over write_file]
+Any of these triggers means use save_last_response, NOT write_file:
+  - The expected output is 300+ lines or 3 KB+ (exam papers, table-of-contents, full chapter, long report, structured outline, complete teaching plan, full code module)
+  - The user said "save as ...", "\u4FDD\u5B58\u4E3A ...", "write the full ...", "complete the entire ...", "list the whole directory"
+  - You'd otherwise need multiple write_file(append=true) calls to fit the content
+
+[How it works]
+  - You only pass the target file path as an argument
+  - The system runs a fresh streaming generation; content streams to the user AND to disk simultaneously (tee mode)
+  - No content is passed via tool arguments \u2192 not subject to the ~2 KB tool_call argument truncation that breaks write_file for large docs
+
+[Anti-pattern \u2014 DO NOT DO]
+  - Do NOT call write_file with chunked append=true to build up a long file. Use save_last_response for the whole file in one pass.
+  - Do NOT first write a partial file with write_file, then "extend" it with edit_file inserts. Use save_last_response.`,
     parameters: {
       path: {
         type: "string",
@@ -9615,8 +9626,8 @@ function autoTrimSessionIfNeeded(session, sizeLimit = SESSION_SIZE_LIMIT) {
 }
 
 // src/web/session-handler.ts
-import { existsSync as existsSync20, readFileSync as readFileSync13, appendFileSync as appendFileSync3, writeFileSync as writeFileSync8, mkdirSync as mkdirSync9, readdirSync as readdirSync9, statSync as statSync8 } from "fs";
-import { join as join13, resolve as resolve4 } from "path";
+import { existsSync as existsSync20, readFileSync as readFileSync13, appendFileSync as appendFileSync3, writeFileSync as writeFileSync8, mkdirSync as mkdirSync9, readdirSync as readdirSync9, statSync as statSync8, createWriteStream } from "fs";
+import { join as join13, resolve as resolve4, dirname as dirname4 } from "path";
 import { execSync as execSync3 } from "child_process";
 
 // src/tools/git-context.ts
@@ -10306,6 +10317,45 @@ ${systemPromptVolatile}` : systemPrompt;
           spawnAgentContext.modelParams = modelParams;
           spawnAgentContext.configManager = this.config;
           ToolExecutor.currentMessageIndex = this.sessions.current?.messages.length ?? 0;
+          const saveLastResponseCall = result.toolCalls.find((tc) => tc.name === "save_last_response");
+          const saveLastResponsePath = saveLastResponseCall ? String(saveLastResponseCall.arguments["path"] ?? "") : "";
+          if (saveLastResponseCall && saveLastResponsePath) {
+            const teeResult = await this.runSaveLastResponseTee(
+              provider,
+              saveLastResponseCall,
+              saveLastResponsePath,
+              apiMessages,
+              extraMessages,
+              systemPrompt,
+              systemPromptVolatile,
+              modelParams,
+              ac,
+              roundUsage
+            );
+            const teeToolResults = result.toolCalls.map((tc) => {
+              if (tc.id === saveLastResponseCall.id) {
+                return {
+                  callId: tc.id,
+                  content: teeResult.summary,
+                  isError: teeResult.isError
+                };
+              }
+              return {
+                callId: tc.id,
+                content: "[skipped: file already saved by tee streaming]",
+                isError: false
+              };
+            });
+            const reasoningContent2 = result.reasoningContent;
+            const newMsgs2 = provider.buildToolResultMessages(result.toolCalls, teeToolResults, reasoningContent2);
+            extraMessages.push(...newMsgs2);
+            persistToolRound(session, result.toolCalls, teeToolResults, {
+              assistantContent: teeResult.content,
+              reasoningContent: reasoningContent2
+            });
+            consecutiveFreeRounds = 0;
+            continue;
+          }
           const toolResults = await this.toolExecutor.executeAll(result.toolCalls);
           const reasoningContent = result.reasoningContent;
           const newMsgs = provider.buildToolResultMessages(result.toolCalls, toolResults, reasoningContent);
@@ -10417,6 +10467,98 @@ ${summaryResult.content}`,
       this.abortController = null;
     }
   }
+  /**
+   * Tee-streaming for save_last_response in Web mode (v0.4.102+).
+   *
+   * 复刻 REPL 的 [repl.ts:2576] 路径——AI 调用 save_last_response(path) 时，
+   * 不走默认 executor，而是发起一次新的 chatStream，把生成的内容同步推送到
+   * WebSocket（`text_delta` 事件，Web UI 实时渲染）+ 写入磁盘文件。这样 AI
+   * 就能一次性输出 >2KB 的大文档（如教材目录、模考试卷），不会被 tool_call
+   * arguments 截断。
+   *
+   * 已知约束：
+   * - 直接跳过 confirm 流程（与 REPL tee 路径一致）。tee 文本会实时显示，用户
+   *   看到不对可中断。
+   * - 同一轮其余 tool_calls 会被标记为 skipped——AI 通常只调一个工具，多并发
+   *   场景极少。
+   */
+  async runSaveLastResponseTee(provider, call, saveToFile, apiMessages, extraMessages, systemPrompt, systemPromptVolatile, modelParams, ac, roundUsage) {
+    this.send({
+      type: "tool_call_start",
+      callId: call.id,
+      toolName: call.name,
+      args: call.arguments,
+      dangerLevel: "write",
+      round: 0,
+      totalRounds: 0,
+      startTime: Date.now()
+    });
+    let fileStream = null;
+    let fullContent = "";
+    let teeUsage;
+    let isError = false;
+    let summary;
+    try {
+      mkdirSync9(dirname4(saveToFile), { recursive: true });
+      fileStream = createWriteStream(saveToFile, { encoding: "utf-8" });
+      const chatRequest = {
+        messages: apiMessages,
+        model: this.currentModel,
+        systemPrompt,
+        systemPromptVolatile,
+        stream: true,
+        temperature: modelParams.temperature,
+        maxTokens: modelParams.maxTokens,
+        timeout: modelParams.timeout,
+        thinking: modelParams.thinking,
+        thinkingBudget: modelParams.thinkingBudget,
+        signal: ac.signal,
+        ...extraMessages.length > 0 ? { _extraMessages: extraMessages } : {}
+      };
+      const stream = provider.chatStream(chatRequest);
+      for await (const chunk of stream) {
+        if (ac.signal.aborted) break;
+        if (chunk.usage) teeUsage = chunk.usage;
+        if (chunk.delta) {
+          fullContent += chunk.delta;
+          this.send({ type: "text_delta", delta: chunk.delta });
+          fileStream.write(chunk.delta);
+        }
+        if (chunk.done) break;
+      }
+      await new Promise((resolve6, reject) => {
+        fileStream.end((err) => err ? reject(err) : resolve6());
+      });
+      const lines = fullContent.split("\n").length;
+      const bytes = Buffer.byteLength(fullContent, "utf-8");
+      summary = `File saved: ${saveToFile} (${lines} lines, ${bytes} bytes)`;
+      undoStack.push(saveToFile, `save_last_response: ${saveToFile}`);
+      if (teeUsage) {
+        roundUsage.inputTokens += teeUsage.inputTokens;
+        roundUsage.outputTokens += teeUsage.outputTokens;
+        roundUsage.cacheCreationTokens += teeUsage.cacheCreationTokens ?? 0;
+        roundUsage.cacheReadTokens += teeUsage.cacheReadTokens ?? 0;
+      }
+    } catch (err) {
+      if (fileStream) {
+        try {
+          await new Promise((resolve6) => fileStream.end(() => resolve6()));
+        } catch {
+        }
+      }
+      isError = true;
+      const msg = err instanceof Error ? err.message : String(err);
+      summary = `[save_last_response failed] ${msg}`;
+    }
+    this.send({
+      type: "tool_call_result",
+      callId: call.id,
+      toolName: call.name,
+      content: summary,
+      isError
+    });
+    return { content: fullContent, summary, isError };
+  }
   /** Consume streaming tool call events and forward to client */
   async consumeToolStream(streamGen, ac) {
     let textContent = "";
@@ -11485,7 +11627,7 @@ ${undoResults.map((r) => `  \u2022 ${r}`).join("\n")}` });
       case "test": {
         this.send({ type: "info", message: "\u{1F9EA} Running tests..." });
         try {
-          const { executeTests } = await import("./run-tests-7LVMNX4S.js");
+          const { executeTests } = await import("./run-tests-SN74WT4Z.js");
           const argStr = args.join(" ").trim();
           let testArgs = {};
           if (argStr) {
@@ -12174,17 +12316,16 @@ Add .md files to create commands.` });
     };
   }
   getFilteredToolDefs() {
-    const excludeWeb = (t) => t.name !== "save_last_response";
     if (this.planMode) {
       return {
-        toolDefs: this.toolRegistry.getDefinitions().filter((t) => PLAN_MODE_READONLY_TOOLS.has(t.name) && excludeWeb(t)),
+        toolDefs: this.toolRegistry.getDefinitions().filter((t) => PLAN_MODE_READONLY_TOOLS.has(t.name)),
         mcpBudgetNote: null
       };
     }
     const skillFilter = this.skillManager?.getActiveToolFilter();
     if (skillFilter) {
       return {
-        toolDefs: this.toolRegistry.getDefinitions().filter((t) => skillFilter.has(t.name) && excludeWeb(t)),
+        toolDefs: this.toolRegistry.getDefinitions().filter((t) => skillFilter.has(t.name)),
         mcpBudgetNote: null
       };
     }
@@ -12192,9 +12333,9 @@ Add .md files to create commands.` });
     if (contextWindow > 0) {
       const toolBudget = Math.floor(contextWindow * 0.2);
       const { definitions, systemNote } = this.toolRegistry.getDefinitionsWithBudget(toolBudget, this.usedMcpToolNames);
-      return { toolDefs: definitions.filter(excludeWeb), mcpBudgetNote: systemNote };
+      return { toolDefs: definitions, mcpBudgetNote: systemNote };
     }
-    return { toolDefs: this.toolRegistry.getDefinitions().filter(excludeWeb), mcpBudgetNote: null };
+    return { toolDefs: this.toolRegistry.getDefinitions(), mcpBudgetNote: null };
   }
   /**
    * Find first matching context file in a directory.
@@ -12686,7 +12827,7 @@ function getModuleDir() {
     if (typeof import.meta?.url === "string") {
       const url = new URL(import.meta.url);
       const filePath = url.pathname.replace(/^\/([A-Z]:)/i, "$1");
-      return dirname4(filePath);
+      return dirname5(filePath);
     }
   } catch {
   }

package/dist/{hub-O5OX6KWP.js → hub-4YGZ4XHN.js}

@@ -385,7 +385,7 @@ ${content}`);
   }
 }
 async function runTaskMode(config, providers, configManager, topic) {
-  const { TaskOrchestrator } = await import("./task-orchestrator-ZATGIRNE.js");
+  const { TaskOrchestrator } = await import("./task-orchestrator-MUIH3XBY.js");
   const orchestrator = new TaskOrchestrator(config, providers, configManager);
   let interrupted = false;
   const onSigint = () => {

package/dist/index.js

@@ -28,10 +28,10 @@ import {
   saveDevState,
   sessionHasMeaningfulContent,
   setupProxy
-} from "./chunk-VWZW5JWK.js";
+} from "./chunk-VOWVIR2U.js";
 import {
   ConfigManager
-} from "./chunk-GXR2VZDI.js";
+} from "./chunk-SN56X6RE.js";
 import {
   ToolExecutor,
   ToolRegistry,
@@ -50,7 +50,7 @@ import {
   spawnAgentContext,
   theme,
   undoStack
-} from "./chunk-MBPA2FMK.js";
+} from "./chunk-RZWWODW7.js";
 import "./chunk-2ZD3YTVM.js";
 import {
   fileCheckpoints
@@ -68,7 +68,7 @@ import "./chunk-KJLJPUY2.js";
 import "./chunk-6VRJGH25.js";
 import "./chunk-2DXY7UGF.js";
 import "./chunk-KHYD3WXE.js";
-import "./chunk-TFCBOPQV.js";
+import "./chunk-OVYOYUP7.js";
 import {
   AGENTIC_BEHAVIOR_GUIDELINE,
   AUTHOR,
@@ -90,7 +90,7 @@ import {
   SKILLS_DIR_NAME,
   VERSION,
   buildUserIdentityPrompt
-} from "./chunk-3MEH7D36.js";
+} from "./chunk-JHPSWYO3.js";
 
 // src/index.ts
 import { program } from "commander";
@@ -2610,7 +2610,7 @@ ${hint}` : "")
       usage: "/test [command|filter]",
       async execute(args, ctx) {
         try {
-          const { executeTests } = await import("./run-tests-IG45VWEM.js");
+          const { executeTests } = await import("./run-tests-3YOJEN2Q.js");
           const argStr = args.join(" ").trim();
           let testArgs = {};
           if (argStr) {
@@ -6749,7 +6749,7 @@ program.command("web").description("Start Web UI server with browser-based chat
     console.error("Error: Invalid port number. Must be between 1 and 65535.");
     process.exit(1);
   }
-  const { startWebServer } = await import("./server-UBHVIJA6.js");
+  const { startWebServer } = await import("./server-TNPDHGQT.js");
   await startWebServer({ port, host: options.host });
 });
 program.command("user [action] [username]").description("Manage Web UI users (list | create <name> | delete <name> | reset-password <name> | migrate <name>)").action(async (action, username) => {
@@ -6872,7 +6872,7 @@ program.command("sessions").description("List recent conversation sessions").act
 });
 program.command("batch <action> [arg] [arg2]").description("Anthropic Message Batches: submit | list | status <id> | results <id> [out] | cancel <id>").option("--dry-run", "Parse and validate input without submitting (submit only)").action(async (action, arg, arg2, options) => {
   try {
-    const batch = await import("./batch-DL3IJ7JQ.js");
+    const batch = await import("./batch-LNTG2IRQ.js");
     switch (action) {
       case "submit":
         if (!arg) {
@@ -6915,7 +6915,7 @@ program.command("batch <action> [arg] [arg2]").description("Anthropic Message Ba
   }
 });
 program.command("mcp-serve").description("Start an MCP server over STDIO, exposing aicli's built-in tools to Claude Desktop / Cursor / other MCP clients").option("--allow-destructive", "Allow bash / run_interactive / task_create (always destructive in MCP mode)").option("--allow-outside-cwd", "Allow tool path arguments to escape the sandbox root \u2014 disabled by default").option("--tools <list>", "Comma-separated whitelist of tools to expose (default: all eligible tools)").option("--cwd <path>", "Working directory AND sandbox root (default: current directory)").action(async (options) => {
-  const { startMcpServer } = await import("./server-X5DQ5V6S.js");
+  const { startMcpServer } = await import("./server-BG4WR6RF.js");
   await startMcpServer({
     allowDestructive: !!options.allowDestructive,
     allowOutsideCwd: !!options.allowOutsideCwd,
@@ -7042,7 +7042,7 @@ program.command("hub [topic]").description("Start multi-agent hub (discuss / bra
     }),
     config.get("customProviders")
   );
-  const { startHub } = await import("./hub-O5OX6KWP.js");
+  const { startHub } = await import("./hub-4YGZ4XHN.js");
   await startHub(
     {
       topic: topic ?? "",

package/dist/{run-tests-IG45VWEM.js → run-tests-3YOJEN2Q.js}

@@ -2,8 +2,8 @@
 import {
   executeTests,
   runTestsTool
-} from "./chunk-TFCBOPQV.js";
-import "./chunk-3MEH7D36.js";
+} from "./chunk-OVYOYUP7.js";
+import "./chunk-JHPSWYO3.js";
 export {
   executeTests,
   runTestsTool

package/dist/{run-tests-7LVMNX4S.js → run-tests-SN74WT4Z.js}

@@ -1,7 +1,7 @@
 import {
   executeTests,
   runTestsTool
-} from "./chunk-5EQP4IAM.js";
+} from "./chunk-DGXUO7D4.js";
 export {
   executeTests,
   runTestsTool

package/dist/{server-X5DQ5V6S.js → server-BG4WR6RF.js}

@@ -3,7 +3,7 @@ import {
   ToolRegistry,
   getDangerLevel,
   schemaToJsonSchema
-} from "./chunk-MBPA2FMK.js";
+} from "./chunk-RZWWODW7.js";
 import "./chunk-2ZD3YTVM.js";
 import "./chunk-4BKXL7SM.js";
 import "./chunk-ANYYM4CF.js";
@@ -12,10 +12,10 @@ import "./chunk-KJLJPUY2.js";
 import "./chunk-6VRJGH25.js";
 import "./chunk-2DXY7UGF.js";
 import "./chunk-KHYD3WXE.js";
-import "./chunk-TFCBOPQV.js";
+import "./chunk-OVYOYUP7.js";
 import {
   VERSION
-} from "./chunk-3MEH7D36.js";
+} from "./chunk-JHPSWYO3.js";
 
 // src/mcp/server.ts
 import { createInterface } from "readline";

package/dist/{server-UBHVIJA6.js → server-TNPDHGQT.js}

@@ -21,10 +21,10 @@ import {
   loadDevState,
   persistToolRound,
   setupProxy
-} from "./chunk-VWZW5JWK.js";
+} from "./chunk-VOWVIR2U.js";
 import {
   ConfigManager
-} from "./chunk-GXR2VZDI.js";
+} from "./chunk-SN56X6RE.js";
 import {
   ToolExecutor,
   ToolRegistry,
@@ -42,7 +42,7 @@ import {
   spawnAgentContext,
   truncateOutput,
   undoStack
-} from "./chunk-MBPA2FMK.js";
+} from "./chunk-RZWWODW7.js";
 import "./chunk-2ZD3YTVM.js";
 import "./chunk-4BKXL7SM.js";
 import "./chunk-ANYYM4CF.js";
@@ -51,7 +51,7 @@ import "./chunk-KJLJPUY2.js";
 import "./chunk-6VRJGH25.js";
 import "./chunk-2DXY7UGF.js";
 import "./chunk-KHYD3WXE.js";
-import "./chunk-TFCBOPQV.js";
+import "./chunk-OVYOYUP7.js";
 import {
   AGENTIC_BEHAVIOR_GUIDELINE,
   AUTHOR,
@@ -70,13 +70,13 @@ import {
   SKILLS_DIR_NAME,
   VERSION,
   buildUserIdentityPrompt
-} from "./chunk-3MEH7D36.js";
+} from "./chunk-JHPSWYO3.js";
 
 // src/web/server.ts
 import express from "express";
 import { createServer } from "http";
 import { WebSocketServer } from "ws";
-import { join as join3, dirname, resolve as resolve2, relative, sep } from "path";
+import { join as join3, dirname as dirname2, resolve as resolve2, relative, sep } from "path";
 import { existsSync as existsSync4, readFileSync as readFileSync4, readdirSync as readdirSync2, statSync as statSync2, realpathSync } from "fs";
 import { networkInterfaces } from "os";
 
@@ -470,8 +470,8 @@ function loadMemoryContent(configDir) {
 }
 
 // src/web/session-handler.ts
-import { existsSync as existsSync3, readFileSync as readFileSync3, appendFileSync, writeFileSync, mkdirSync, readdirSync, statSync } from "fs";
-import { join as join2, resolve } from "path";
+import { existsSync as existsSync3, readFileSync as readFileSync3, appendFileSync, writeFileSync, mkdirSync, readdirSync, statSync, createWriteStream } from "fs";
+import { join as join2, resolve, dirname } from "path";
 import { execSync } from "child_process";
 var FREE_ROUND_TOOLS = /* @__PURE__ */ new Set(["write_todos"]);
 var MAX_CONSECUTIVE_FREE_ROUNDS = 5;
@@ -1069,6 +1069,45 @@ ${systemPromptVolatile}` : systemPrompt;
           spawnAgentContext.modelParams = modelParams;
           spawnAgentContext.configManager = this.config;
           ToolExecutor.currentMessageIndex = this.sessions.current?.messages.length ?? 0;
+          const saveLastResponseCall = result.toolCalls.find((tc) => tc.name === "save_last_response");
+          const saveLastResponsePath = saveLastResponseCall ? String(saveLastResponseCall.arguments["path"] ?? "") : "";
+          if (saveLastResponseCall && saveLastResponsePath) {
+            const teeResult = await this.runSaveLastResponseTee(
+              provider,
+              saveLastResponseCall,
+              saveLastResponsePath,
+              apiMessages,
+              extraMessages,
+              systemPrompt,
+              systemPromptVolatile,
+              modelParams,
+              ac,
+              roundUsage
+            );
+            const teeToolResults = result.toolCalls.map((tc) => {
+              if (tc.id === saveLastResponseCall.id) {
+                return {
+                  callId: tc.id,
+                  content: teeResult.summary,
+                  isError: teeResult.isError
+                };
+              }
+              return {
+                callId: tc.id,
+                content: "[skipped: file already saved by tee streaming]",
+                isError: false
+              };
+            });
+            const reasoningContent2 = result.reasoningContent;
+            const newMsgs2 = provider.buildToolResultMessages(result.toolCalls, teeToolResults, reasoningContent2);
+            extraMessages.push(...newMsgs2);
+            persistToolRound(session, result.toolCalls, teeToolResults, {
+              assistantContent: teeResult.content,
+              reasoningContent: reasoningContent2
+            });
+            consecutiveFreeRounds = 0;
+            continue;
+          }
           const toolResults = await this.toolExecutor.executeAll(result.toolCalls);
           const reasoningContent = result.reasoningContent;
           const newMsgs = provider.buildToolResultMessages(result.toolCalls, toolResults, reasoningContent);
@@ -1180,6 +1219,98 @@ ${summaryResult.content}`,
       this.abortController = null;
     }
   }
+  /**
+   * Tee-streaming for save_last_response in Web mode (v0.4.102+).
+   *
+   * 复刻 REPL 的 [repl.ts:2576] 路径——AI 调用 save_last_response(path) 时，
+   * 不走默认 executor，而是发起一次新的 chatStream，把生成的内容同步推送到
+   * WebSocket（`text_delta` 事件，Web UI 实时渲染）+ 写入磁盘文件。这样 AI
+   * 就能一次性输出 >2KB 的大文档（如教材目录、模考试卷），不会被 tool_call
+   * arguments 截断。
+   *
+   * 已知约束：
+   * - 直接跳过 confirm 流程（与 REPL tee 路径一致）。tee 文本会实时显示，用户
+   *   看到不对可中断。
+   * - 同一轮其余 tool_calls 会被标记为 skipped——AI 通常只调一个工具，多并发
+   *   场景极少。
+   */
+  async runSaveLastResponseTee(provider, call, saveToFile, apiMessages, extraMessages, systemPrompt, systemPromptVolatile, modelParams, ac, roundUsage) {
+    this.send({
+      type: "tool_call_start",
+      callId: call.id,
+      toolName: call.name,
+      args: call.arguments,
+      dangerLevel: "write",
+      round: 0,
+      totalRounds: 0,
+      startTime: Date.now()
+    });
+    let fileStream = null;
+    let fullContent = "";
+    let teeUsage;
+    let isError = false;
+    let summary;
+    try {
+      mkdirSync(dirname(saveToFile), { recursive: true });
+      fileStream = createWriteStream(saveToFile, { encoding: "utf-8" });
+      const chatRequest = {
+        messages: apiMessages,
+        model: this.currentModel,
+        systemPrompt,
+        systemPromptVolatile,
+        stream: true,
+        temperature: modelParams.temperature,
+        maxTokens: modelParams.maxTokens,
+        timeout: modelParams.timeout,
+        thinking: modelParams.thinking,
+        thinkingBudget: modelParams.thinkingBudget,
+        signal: ac.signal,
+        ...extraMessages.length > 0 ? { _extraMessages: extraMessages } : {}
+      };
+      const stream = provider.chatStream(chatRequest);
+      for await (const chunk of stream) {
+        if (ac.signal.aborted) break;
+        if (chunk.usage) teeUsage = chunk.usage;
+        if (chunk.delta) {
+          fullContent += chunk.delta;
+          this.send({ type: "text_delta", delta: chunk.delta });
+          fileStream.write(chunk.delta);
+        }
+        if (chunk.done) break;
+      }
+      await new Promise((resolve3, reject) => {
+        fileStream.end((err) => err ? reject(err) : resolve3());
+      });
+      const lines = fullContent.split("\n").length;
+      const bytes = Buffer.byteLength(fullContent, "utf-8");
+      summary = `File saved: ${saveToFile} (${lines} lines, ${bytes} bytes)`;
+      undoStack.push(saveToFile, `save_last_response: ${saveToFile}`);
+      if (teeUsage) {
+        roundUsage.inputTokens += teeUsage.inputTokens;
+        roundUsage.outputTokens += teeUsage.outputTokens;
+        roundUsage.cacheCreationTokens += teeUsage.cacheCreationTokens ?? 0;
+        roundUsage.cacheReadTokens += teeUsage.cacheReadTokens ?? 0;
+      }
+    } catch (err) {
+      if (fileStream) {
+        try {
+          await new Promise((resolve3) => fileStream.end(() => resolve3()));
+        } catch {
+        }
+      }
+      isError = true;
+      const msg = err instanceof Error ? err.message : String(err);
+      summary = `[save_last_response failed] ${msg}`;
+    }
+    this.send({
+      type: "tool_call_result",
+      callId: call.id,
+      toolName: call.name,
+      content: summary,
+      isError
+    });
+    return { content: fullContent, summary, isError };
+  }
   /** Consume streaming tool call events and forward to client */
   async consumeToolStream(streamGen, ac) {
     let textContent = "";
@@ -2248,7 +2379,7 @@ ${undoResults.map((r) => `  \u2022 ${r}`).join("\n")}` });
       case "test": {
         this.send({ type: "info", message: "\u{1F9EA} Running tests..." });
         try {
-          const { executeTests } = await import("./run-tests-IG45VWEM.js");
+          const { executeTests } = await import("./run-tests-3YOJEN2Q.js");
           const argStr = args.join(" ").trim();
           let testArgs = {};
           if (argStr) {
@@ -2937,17 +3068,16 @@ Add .md files to create commands.` });
     };
   }
   getFilteredToolDefs() {
-    const excludeWeb = (t) => t.name !== "save_last_response";
     if (this.planMode) {
       return {
-        toolDefs: this.toolRegistry.getDefinitions().filter((t) => PLAN_MODE_READONLY_TOOLS.has(t.name) && excludeWeb(t)),
+        toolDefs: this.toolRegistry.getDefinitions().filter((t) => PLAN_MODE_READONLY_TOOLS.has(t.name)),
         mcpBudgetNote: null
       };
     }
     const skillFilter = this.skillManager?.getActiveToolFilter();
     if (skillFilter) {
       return {
-        toolDefs: this.toolRegistry.getDefinitions().filter((t) => skillFilter.has(t.name) && excludeWeb(t)),
+        toolDefs: this.toolRegistry.getDefinitions().filter((t) => skillFilter.has(t.name)),
         mcpBudgetNote: null
       };
     }
@@ -2955,9 +3085,9 @@ Add .md files to create commands.` });
     if (contextWindow > 0) {
       const toolBudget = Math.floor(contextWindow * 0.2);
       const { definitions, systemNote } = this.toolRegistry.getDefinitionsWithBudget(toolBudget, this.usedMcpToolNames);
-      return { toolDefs: definitions.filter(excludeWeb), mcpBudgetNote: systemNote };
+      return { toolDefs: definitions, mcpBudgetNote: systemNote };
     }
-    return { toolDefs: this.toolRegistry.getDefinitions().filter(excludeWeb), mcpBudgetNote: null };
+    return { toolDefs: this.toolRegistry.getDefinitions(), mcpBudgetNote: null };
   }
   /**
    * Find first matching context file in a directory.
@@ -3227,7 +3357,7 @@ function getModuleDir() {
     if (typeof import.meta?.url === "string") {
       const url = new URL(import.meta.url);
       const filePath = url.pathname.replace(/^\/([A-Z]:)/i, "$1");
-      return dirname(filePath);
+      return dirname2(filePath);
     }
   } catch {
   }

package/dist/{task-orchestrator-ZATGIRNE.js → task-orchestrator-MUIH3XBY.js}

@@ -4,7 +4,7 @@ import {
   getDangerLevel,
   googleSearchContext,
   truncateOutput
-} from "./chunk-MBPA2FMK.js";
+} from "./chunk-RZWWODW7.js";
 import "./chunk-2ZD3YTVM.js";
 import "./chunk-4BKXL7SM.js";
 import "./chunk-ANYYM4CF.js";
@@ -13,10 +13,10 @@ import "./chunk-KJLJPUY2.js";
 import "./chunk-6VRJGH25.js";
 import "./chunk-2DXY7UGF.js";
 import "./chunk-KHYD3WXE.js";
-import "./chunk-TFCBOPQV.js";
+import "./chunk-OVYOYUP7.js";
 import {
   SUBAGENT_ALLOWED_TOOLS
-} from "./chunk-3MEH7D36.js";
+} from "./chunk-JHPSWYO3.js";
 
 // src/hub/task-orchestrator.ts
 import { createInterface } from "readline";

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "jinzd-ai-cli",
-  "version": "0.4.101",
+  "version": "0.4.103",
   "description": "Cross-platform REPL-style AI CLI with multi-provider support",
   "type": "module",
   "main": "./dist/index.js",