jinzd-ai-cli 0.4.57 → 0.4.59

package/dist/{chunk-G6K64M6X.js → chunk-7RX7675B.js}

@@ -10,7 +10,7 @@ import {
   SUBAGENT_DEFAULT_MAX_ROUNDS,
   SUBAGENT_MAX_ROUNDS_LIMIT,
   runTestsTool
-} from "./chunk-C2MNNHJ6.js";
+} from "./chunk-YJCJBUOG.js";
 
 // src/tools/builtin/bash.ts
 import { execSync } from "child_process";
@@ -221,7 +221,7 @@ Important rules:
       },
       timeout: {
         type: "number",
-        description: "Timeout in milliseconds, defaults to 30000",
+        description: "Timeout in milliseconds, defaults to 120000 (2 min), max 300000 (5 min). For recursive filesystem operations (e.g. Get-ChildItem -Recurse on large trees, find on deep dirs), pass a larger value explicitly (e.g. 240000) \u2014 the default may be too short.",
         required: false
       }
     },
@@ -229,8 +229,9 @@ Important rules:
   },
   async execute(args) {
     const command = String(args["command"] ?? "");
+    const DEFAULT_TIMEOUT = 12e4;
     const MAX_TIMEOUT = 3e5;
-    const timeout = Math.min(Math.max(Number(args["timeout"] ?? 3e4), 1e3), MAX_TIMEOUT);
+    const timeout = Math.min(Math.max(Number(args["timeout"] ?? DEFAULT_TIMEOUT), 1e3), MAX_TIMEOUT);
     const cwdArg = args["cwd"] ? String(args["cwd"]) : void 0;
     if (!command.trim()) {
       throw new ToolError("bash", "command is required");
@@ -287,18 +288,38 @@ Important rules:
       return result || "(command completed with no output)";
     } catch (err) {
       pushBashUndoEntries(beforeSnapshot, parsedTargetsBefore, effectiveCwd);
-      if (err && typeof err === "object" && "status" in err) {
+      if (err && typeof err === "object") {
         const execErr = err;
-        const stderr = IS_WINDOWS && Buffer.isBuffer(execErr.stderr) ? execErr.stderr.toString("utf-8").trim() : execErr.stderr?.toString().trim() ?? "";
-        const stdout = IS_WINDOWS && Buffer.isBuffer(execErr.stdout) ? execErr.stdout.toString("utf-8").trim() : execErr.stdout?.toString().trim() ?? "";
-        const combined = [stdout, stderr].filter(Boolean).join("\n");
-        throw new ToolError(
-          "bash",
-          `Exit code ${execErr.status}:
+        const isTimeout = execErr.code === "ETIMEDOUT" || execErr.status == null && execErr.signal === "SIGTERM" || /ETIMEDOUT/i.test(execErr.message ?? "");
+        if (isTimeout) {
+          const seconds = Math.round(timeout / 1e3);
+          throw new ToolError(
+            "bash",
+            `Command timed out after ${seconds}s.
+
+The previous command ran for longer than the timeout limit and was killed. This usually means it is scanning a large filesystem tree (recursive Get-ChildItem / find), compressing a big archive, or waiting on a network request.
+
+How to recover (pick ONE \u2014 do NOT retry the same command):
+  1. Pass an explicit longer timeout, e.g. timeout: ${Math.min(timeout * 2, MAX_TIMEOUT)}
+  2. Use a non-recursive / narrower alternative (e.g. 'Get-ChildItem -File' without -Recurse, or limit depth with -Depth 1)
+  3. Split the work into smaller batches (process subdirectories one at a time)
+  4. For size/count queries, use '(Get-ChildItem X).Count' per directory instead of one giant pipeline
+
+[Do NOT retry the identical command \u2014 it will time out again.]`
+          );
+        }
+        if ("status" in execErr && execErr.status !== void 0) {
+          const stderr = IS_WINDOWS && Buffer.isBuffer(execErr.stderr) ? execErr.stderr.toString("utf-8").trim() : execErr.stderr?.toString().trim() ?? "";
+          const stdout = IS_WINDOWS && Buffer.isBuffer(execErr.stdout) ? execErr.stdout.toString("utf-8").trim() : execErr.stdout?.toString().trim() ?? "";
+          const combined = [stdout, stderr].filter(Boolean).join("\n");
+          throw new ToolError(
+            "bash",
+            `Exit code ${execErr.status}:
 ${combined || (execErr.message ?? "Unknown error")}
 
 [Command failed. Report this error to the user. Do not retry with variant commands.]`
-        );
+          );
+        }
       }
       throw err;
     }

package/dist/{chunk-H7MNK3YO.js → chunk-F5WLEWN2.js}

@@ -6,7 +6,7 @@ import { platform } from "os";
 import chalk from "chalk";
 
 // src/core/constants.ts
-var VERSION = "0.4.57";
+var VERSION = "0.4.59";
 var APP_NAME = "ai-cli";
 var CONFIG_DIR_NAME = ".aicli";
 var CONFIG_FILE_NAME = "config.json";

package/dist/{chunk-G5REL4FK.js → chunk-NXSYL5OP.js}

@@ -7,7 +7,7 @@ import {
   ProviderNotFoundError,
   RateLimitError,
   schemaToJsonSchema
-} from "./chunk-G6K64M6X.js";
+} from "./chunk-7RX7675B.js";
 import {
   APP_NAME,
   CONFIG_DIR_NAME,
@@ -20,7 +20,7 @@ import {
   MCP_TOOL_PREFIX,
   PLUGINS_DIR_NAME,
   VERSION
-} from "./chunk-C2MNNHJ6.js";
+} from "./chunk-YJCJBUOG.js";
 
 // src/config/config-manager.ts
 import { readFileSync, writeFileSync, existsSync, mkdirSync } from "fs";

package/dist/{chunk-C2MNNHJ6.js → chunk-YJCJBUOG.js}

@@ -8,7 +8,7 @@ import { platform } from "os";
 import chalk from "chalk";
 
 // src/core/constants.ts
-var VERSION = "0.4.57";
+var VERSION = "0.4.59";
 var APP_NAME = "ai-cli";
 var CONFIG_DIR_NAME = ".aicli";
 var CONFIG_FILE_NAME = "config.json";

package/dist/{hub-P3BR4JB5.js → hub-3BY5W4VE.js}

@@ -385,7 +385,7 @@ ${content}`);
   }
 }
 async function runTaskMode(config, providers, configManager, topic) {
-  const { TaskOrchestrator } = await import("./task-orchestrator-FRF6LTWK.js");
+  const { TaskOrchestrator } = await import("./task-orchestrator-I5YCZ72U.js");
   const orchestrator = new TaskOrchestrator(config, providers, configManager);
   let interrupted = false;
   const onSigint = () => {

package/dist/index.js

@@ -27,7 +27,7 @@ import {
   saveDevState,
   sessionHasMeaningfulContent,
   setupProxy
-} from "./chunk-G5REL4FK.js";
+} from "./chunk-NXSYL5OP.js";
 import {
   ToolExecutor,
   ToolRegistry,
@@ -41,7 +41,7 @@ import {
   spawnAgentContext,
   theme,
   undoStack
-} from "./chunk-G6K64M6X.js";
+} from "./chunk-7RX7675B.js";
 import {
   fileCheckpoints
 } from "./chunk-4BKXL7SM.js";
@@ -66,7 +66,7 @@ import {
   SKILLS_DIR_NAME,
   VERSION,
   buildUserIdentityPrompt
-} from "./chunk-C2MNNHJ6.js";
+} from "./chunk-YJCJBUOG.js";
 
 // src/index.ts
 import { program } from "commander";
@@ -2161,7 +2161,7 @@ ${hint}` : "")
       usage: "/test [command|filter]",
       async execute(args, ctx) {
         try {
-          const { executeTests } = await import("./run-tests-NVCAP42D.js");
+          const { executeTests } = await import("./run-tests-IW6GHAVV.js");
           const argStr = args.join(" ").trim();
           let testArgs = {};
           if (argStr) {
@@ -4322,7 +4322,29 @@ Session '${this.resumeSessionId}' not found.
         sendNotification("ai-cli", `Task completed in ${Math.round(elapsed / 1e3)}s`);
       }
     } catch (err) {
-      this.renderer.renderError(err);
+      const errMsg = err instanceof Error ? err.message : String(err);
+      const isCtxLengthError = /maximum context length|context_length_exceeded|context window|too many tokens|reduce the length of the messages/i.test(errMsg);
+      if (isCtxLengthError) {
+        process.stderr.write(
+          theme.error(`
+\u26A0 Context length exceeded \u2014 the conversation is too long for this model.
+`)
+        );
+        process.stderr.write(theme.dim(`  Details: ${errMsg.split("\n")[0]}
+`));
+        process.stderr.write(
+          theme.dim(
+            `  Recovery options:
+    1. Run /compact to summarize old messages and free context
+    2. Run /clear to start a fresh session (keeps this terminal)
+    3. Run /model to switch to a model with a larger context window
+
+`
+          )
+        );
+      } else {
+        this.renderer.renderError(err);
+      }
     }
   }
   /**
@@ -4397,6 +4419,23 @@ Session '${this.resumeSessionId}' not found.
     }
     return total;
   }
+  /**
+   * 估算 agentic 循环当前请求的 token 数（包含 session messages + extraMessages + system prompt）。
+   * extraMessages 结构复杂（含 tool_calls、tool 结果等），这里用 JSON.stringify 后除以字符/token 比。
+   * 用于 handleChatWithTools 循环内每轮发 API 前做上下文压力检查。
+   */
+  estimateRequestTokens(systemPrompt, extraMessages) {
+    let total = this.estimateConversationTokens();
+    if (extraMessages.length > 0) {
+      try {
+        const serialized = JSON.stringify(extraMessages);
+        total += this.estimateTokens(serialized);
+      } catch {
+      }
+    }
+    void systemPrompt;
+    return total;
+  }
   /**
    * 获取当前模型的 context window 大小。
    */
@@ -4865,6 +4904,7 @@ You have a maximum of ${maxToolRounds} tool call rounds for this task. Plan effi
     let lastToolCallSignature = "";
     let repeatedToolCallCount = 0;
     let emptyResponseRetries = 0;
+    let warnedCtx80 = false;
     const roundToolHistory = [];
     this.setupInterjectionListener();
     try {
@@ -4923,6 +4963,52 @@ You have a maximum of ${maxToolRounds} tool call rounds for this task. Plan effi
 `));
           extraMessages.push({ role: "user", content: msg });
         }
+        const ctxWindow = this.getContextWindowSize();
+        if (ctxWindow > 0) {
+          const reqTokens = this.estimateRequestTokens(systemPrompt, extraMessages);
+          const reqRatio = reqTokens / ctxWindow;
+          if (reqRatio >= 0.95) {
+            spinner.stop();
+            process.stderr.write(
+              theme.error(
+                `
+\u26A0 Context at ${Math.round(reqRatio * 100)}% of ${fmtTokens(ctxWindow)} \u2014 aborting agentic loop before API rejection.
+`
+              )
+            );
+            process.stderr.write(
+              theme.dim(
+                `  Too much tool output accumulated this turn. Your work so far is preserved.
+  Recovery: run /compact to shrink history, then ask the AI to continue.
+
+`
+              )
+            );
+            if (roundUsage.inputTokens > 0 || roundUsage.outputTokens > 0) {
+              this.addSessionUsage(roundUsage);
+              session.addTokenUsage(roundUsage);
+              if (this.shouldShowTokens()) {
+                this.renderer.renderUsage(roundUsage, this.sessionTokenUsage);
+              }
+            }
+            return;
+          } else if (reqRatio >= 0.8 && !warnedCtx80) {
+            warnedCtx80 = true;
+            spinner.stop();
+            process.stdout.write(
+              theme.warning(
+                `
+\u26A0 Context at ${Math.round(reqRatio * 100)}% of ${fmtTokens(ctxWindow)} \u2014 asking AI to wrap up.
+`
+              )
+            );
+            extraMessages.push({
+              role: "user",
+              content: `\u26A0\uFE0F Context pressure: ~${Math.round(reqRatio * 100)}% of the ${fmtTokens(ctxWindow)} context window is used. Avoid reading more files or running broad scans. Finish the current critical step, then produce a final summary. Every unnecessary tool call now risks breaking the conversation.`
+            });
+            spinner.start(`Thinking... (round ${round + 1}/${maxToolRounds})`);
+          }
+        }
         let result;
         let alreadyRendered = false;
         const chatRequest = {
@@ -5609,7 +5695,7 @@ 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-46J5MXHG.js");
+  const { startWebServer } = await import("./server-J7PNU32E.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) => {
@@ -5842,7 +5928,7 @@ program.command("hub [topic]").description("Start multi-agent hub (discuss / bra
     }),
     config.get("customProviders")
   );
-  const { startHub } = await import("./hub-P3BR4JB5.js");
+  const { startHub } = await import("./hub-3BY5W4VE.js");
   await startHub(
     {
       topic: topic ?? "",

package/dist/{run-tests-NVCAP42D.js → run-tests-IW6GHAVV.js}

@@ -2,7 +2,7 @@
 import {
   executeTests,
   runTestsTool
-} from "./chunk-C2MNNHJ6.js";
+} from "./chunk-YJCJBUOG.js";
 export {
   executeTests,
   runTestsTool

package/dist/{run-tests-XAWG6R73.js → run-tests-OT3MSOUF.js}

@@ -1,7 +1,7 @@
 import {
   executeTests,
   runTestsTool
-} from "./chunk-H7MNK3YO.js";
+} from "./chunk-F5WLEWN2.js";
 export {
   executeTests,
   runTestsTool

package/dist/{server-46J5MXHG.js → server-J7PNU32E.js}

@@ -17,7 +17,7 @@ import {
   hadPreviousWriteToolCalls,
   loadDevState,
   setupProxy
-} from "./chunk-G5REL4FK.js";
+} from "./chunk-NXSYL5OP.js";
 import {
   AuthManager
 } from "./chunk-BYNY5JPB.js";
@@ -36,7 +36,7 @@ import {
   spawnAgentContext,
   truncateOutput,
   undoStack
-} from "./chunk-G6K64M6X.js";
+} from "./chunk-7RX7675B.js";
 import "./chunk-4BKXL7SM.js";
 import {
   AGENTIC_BEHAVIOR_GUIDELINE,
@@ -56,7 +56,7 @@ import {
   SKILLS_DIR_NAME,
   VERSION,
   buildUserIdentityPrompt
-} from "./chunk-C2MNNHJ6.js";
+} from "./chunk-YJCJBUOG.js";
 
 // src/web/server.ts
 import express from "express";
@@ -642,6 +642,49 @@ var SessionHandler = class _SessionHandler {
     } catch {
     }
   }
+  /** 获取当前模型的 context window 大小（0 表示未知）*/
+  getContextWindowSize() {
+    try {
+      const provider = this.providers.get(this.currentProvider);
+      const modelInfo = provider?.info.models.find((m) => m.id === this.currentModel);
+      return modelInfo?.contextWindow ?? 0;
+    } catch {
+      return 0;
+    }
+  }
+  /** 粗略估算文本 token 数（2.5 chars/token）*/
+  estTokens(text) {
+    return Math.ceil(text.length / 2.5);
+  }
+  /**
+   * 估算当前 agentic 请求总 token 数（session messages + extraMessages + system prompt）。
+   * 用于 handleChat 循环内每轮发 API 前的压力检查。
+   */
+  estimateRequestTokens(systemPrompt, extraMessages) {
+    let total = 0;
+    if (systemPrompt) total += this.estTokens(systemPrompt);
+    const session = this.sessions.current;
+    if (session) {
+      for (const msg of session.messages) {
+        if (typeof msg.content === "string") {
+          total += this.estTokens(msg.content);
+        } else if (Array.isArray(msg.content)) {
+          for (const part of msg.content) {
+            if (part.type === "text" && part.text) {
+              total += this.estTokens(part.text);
+            }
+          }
+        }
+      }
+    }
+    if (extraMessages.length > 0) {
+      try {
+        total += this.estTokens(JSON.stringify(extraMessages));
+      } catch {
+      }
+    }
+    return total;
+  }
   /** Save session only if it exists and has messages (never persist empty "Untitled" sessions). */
   saveIfNeeded() {
     if (this.sessions.current && this.sessions.current.messages.length > 0) {
@@ -801,6 +844,7 @@ You have a maximum of ${maxToolRounds} tool call rounds for this task. Plan effi
     let warnedLow = false;
     let warnedCritical = false;
     let emptyResponseRetries = 0;
+    let warnedCtx80 = false;
     const ac = new AbortController();
     this.abortController = ac;
     try {
@@ -836,6 +880,38 @@ You have a maximum of ${maxToolRounds} tool call rounds for this task. Plan effi
           this.send({ type: "info", message: `\u26A1 Interjection: "${msg}"` });
           extraMessages.push({ role: "user", content: msg });
         }
+        const ctxWindow = this.getContextWindowSize();
+        if (ctxWindow > 0) {
+          const reqTokens = this.estimateRequestTokens(systemPrompt, extraMessages);
+          const reqRatio = reqTokens / ctxWindow;
+          if (reqRatio >= 0.95) {
+            this.send({
+              type: "response_done",
+              content: `\u26A0 Context at ${Math.round(reqRatio * 100)}% of ${ctxWindow.toLocaleString()} tokens \u2014 aborting before API rejection.
+
+Too much tool output accumulated this turn. Your work so far is preserved.
+
+**Recovery**:
+1. Run \`/compact\` to shrink history, then ask the AI to continue
+2. Run \`/clear\` to start fresh
+3. Switch to a larger-context model`,
+              usage: roundUsage
+            });
+            this.addWebSessionUsage(roundUsage);
+            session.addTokenUsage(roundUsage);
+            return;
+          } else if (reqRatio >= 0.8 && !warnedCtx80) {
+            warnedCtx80 = true;
+            this.send({
+              type: "info",
+              message: `\u26A0 Context at ${Math.round(reqRatio * 100)}% \u2014 asking AI to wrap up`
+            });
+            extraMessages.push({
+              role: "user",
+              content: `\u26A0\uFE0F Context pressure: ~${Math.round(reqRatio * 100)}% of the ${ctxWindow.toLocaleString()}-token context window is used. Avoid reading more files or running broad scans. Finish the current critical step, then produce a final summary. Every unnecessary tool call now risks breaking the conversation.`
+            });
+          }
+        }
         const chatRequest = {
           messages: apiMessages,
           model: this.currentModel,
@@ -850,11 +926,34 @@ You have a maximum of ${maxToolRounds} tool call rounds for this task. Plan effi
           ...extraMessages.length > 0 ? { _extraMessages: extraMessages } : {}
         };
         let result;
-        if (supportsStreamingTools) {
-          const streamGen = provider.chatWithToolsStream(chatRequest, toolDefs);
-          result = await this.consumeToolStream(streamGen, ac);
-        } else {
-          result = await provider.chatWithTools(chatRequest, toolDefs);
+        try {
+          if (supportsStreamingTools) {
+            const streamGen = provider.chatWithToolsStream(chatRequest, toolDefs);
+            result = await this.consumeToolStream(streamGen, ac);
+          } else {
+            result = await provider.chatWithTools(chatRequest, toolDefs);
+          }
+        } catch (providerErr) {
+          const errMsg = providerErr instanceof Error ? providerErr.message : String(providerErr);
+          const isCtxLengthError = /maximum context length|context_length_exceeded|context window|too many tokens|reduce the length of the messages/i.test(errMsg);
+          if (isCtxLengthError) {
+            this.send({
+              type: "response_done",
+              content: `\u26A0 Context length exceeded \u2014 the conversation is too long for this model.
+
+Details: ${errMsg.split("\n")[0]}
+
+**Recovery options**:
+1. Run \`/compact\` to summarize old messages and free context
+2. Run \`/clear\` to start a fresh session
+3. Run \`/model\` to switch to a model with a larger context window`,
+              usage: roundUsage
+            });
+            this.addWebSessionUsage(roundUsage);
+            session.addTokenUsage(roundUsage);
+            return;
+          }
+          throw providerErr;
         }
         if (ac.signal.aborted) break;
         if (result.usage) {
@@ -1816,7 +1915,7 @@ ${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-NVCAP42D.js");
+          const { executeTests } = await import("./run-tests-IW6GHAVV.js");
           const argStr = args.join(" ").trim();
           let testArgs = {};
           if (argStr) {

package/dist/{task-orchestrator-FRF6LTWK.js → task-orchestrator-I5YCZ72U.js}

@@ -4,11 +4,11 @@ import {
   getDangerLevel,
   googleSearchContext,
   truncateOutput
-} from "./chunk-G6K64M6X.js";
+} from "./chunk-7RX7675B.js";
 import "./chunk-4BKXL7SM.js";
 import {
   SUBAGENT_ALLOWED_TOOLS
-} from "./chunk-C2MNNHJ6.js";
+} from "./chunk-YJCJBUOG.js";
 
 // src/hub/task-orchestrator.ts
 import { createInterface } from "readline";

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "jinzd-ai-cli",
-  "version": "0.4.57",
+  "version": "0.4.59",
   "description": "Cross-platform REPL-style AI CLI with multi-provider support",
   "type": "module",
   "main": "./dist/index.js",