nongyu-agent-sdk 1.0.0

package/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 TangLei
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.md

@@ -0,0 +1,55 @@
+# 农屿Agent SDK
+
+## 项目描述
+
+农屿Agent是一个希望内嵌在农屿系统中工作的Agent，是农屿系统内部专用的Agent，而非通用Agent。
+
+## 背景补充
+
+### 农屿系统是什么？
+
+- 农屿App（主站，面向四川农业大学学子的一站式智慧校园助手，是本Agent未来的主要宿主环境）
+- 农屿后台管理系统（在更远的将来，本Agent将拓展问数等能力，用于智能化后台管理系统）
+- 农屿官网（本Agent将暴露部分能力用于官网对外展示）
+
+## 能力边界设定
+
+### 核心能力
+
+1. 教务系统Agent
+   - 教务网首页通知询问
+     ```
+     用户问：27届毕业生毕业照采集的时间地点是什么？
+     Agent回答：雅安校区时间xxx，地点xxx；成都校区时间xxx，地点xxx
+     ```
+   - 教务网竞赛通知询问
+     ```
+     用户问： 第四届四川农业大学计算机设计大赛的报名方式是什么？
+     Agent回答：通过填写报名表.......
+     ```
+   - 学业进度查询
+     ```
+     用户问：我的学业进度是什么？还差多少才能修够学分？
+     Agent回答：您的学业进度为....还需要修读xxx学分
+     ```
+   - 个人成绩查询（不展示示例）
+   - 个人排名查询（不展示示例）
+   - 考试安排查询（不展示示例）
+   - 个人课表查询（不展示示例）
+   - 调停课查询（待扩展）
+   - 空教室查询（待扩展）
+   - 教师课表查询（待扩展）
+   - 按学号对某学生课表查询（待扩展）
+
+2. 二课系统Agent
+   - 二课活动查询/搜索（不展示示例）
+   - 个人二课数据查询（不展示示例）
+   - 二课活动报名、取消报名（待扩展）
+
+### 不能做什么？不应该做什么？不擅长做什么？
+
+1. 定位清晰
+   农屿Agent是一个专用Agent，它专用于农屿系统内部的各类问答、查询等操作，农屿Agent应该是全世界最懂川农的Agent，至少在我看来
+
+2. 局限
+   它不会写代码、写文章等等（理论上它会，毕竟大模型本身会，但我们应该认为它不会），专业的Agent应该干专业的事，而不是要求它或者需要它什么都会

package/dist/index.d.ts

@@ -0,0 +1,85 @@
+import { EventType } from 'mitt';
+
+/**
+ * Agent 主类，负责协调上下文管理、大模型调用及 ReAct 逻辑循环
+ */
+declare class Agent {
+    private llmClient;
+    private contextManager;
+    private toolManager;
+    private maxRetries;
+    private baseURL;
+    private apiKey;
+    private mode;
+    private log;
+    private handleToolCall;
+    private handleFinalAnswer;
+    private handleStrategy;
+    /**
+     * 初始化 Agent
+     * @param options 配置选项，包含 API Key、基础 URL、最大 ReAct 循环次数、运行模式、模型名称
+     */
+    constructor(options?: {
+        baseURL?: string;
+        apiKey?: string;
+        maxTokens?: number;
+        mode?: 'dev' | 'prod';
+        maxRetries?: number;
+        model?: string;
+    });
+    /**
+     * 处理用户提问的主入口
+     * @param userInput 用户输入内容
+     * @param stream 是否启用流式响应
+     * @returns 最终回答内容
+     */
+    chat(userInput: string, stream?: boolean): Promise<string>;
+    /**
+     * 清除当前 Agent 的上下文记忆
+     */
+    clearContext(): void;
+}
+
+/**
+ * 事件系统类，基于 mitt 封装
+ * 提供事件的订阅、发布和取消订阅功能
+ */
+declare class EventSystem {
+    private emitter;
+    constructor();
+    /**
+     * 订阅事件
+     * @param event 事件名称
+     * @param handler 事件处理函数
+     */
+    on<T = any>(event: EventType, handler: (event: T) => void): void;
+    /**
+     * 订阅事件，只触发一次
+     * @param event 事件名称
+     * @param handler 事件处理函数
+     */
+    once<T = any>(event: EventType, handler: (event: T) => void): void;
+    /**
+     * 取消订阅事件
+     * @param event 事件名称
+     * @param handler 事件处理函数
+     */
+    off<T = any>(event: EventType, handler: (event: T) => void): void;
+    /**
+     * 发布事件
+     * @param event 事件名称
+     * @param data 事件数据
+     */
+    emit<T = any>(event: EventType, data?: T): void;
+    /**
+     * 清空所有事件订阅
+     */
+    clear(): void;
+    /**
+     * 获取当前所有事件订阅数量
+     */
+    getEventCount(): number;
+}
+declare const eventSystem: EventSystem;
+
+export { Agent, eventSystem };

package/dist/index.js

@@ -0,0 +1,60 @@
+import I from"axios";var h=class{constructor(e){this.mode="dev";this.mode=e?.mode||"dev",this.instance=I.create({timeout:1e4,headers:{"Content-Type":"application/json"},...e}),this.instance.interceptors.request.use(t=>t,t=>Promise.reject(t)),this.instance.interceptors.response.use(t=>t,t=>(this.mode==="dev"&&(t.response?console.error(`HTTP Error: ${t.response.status} - ${t.response.statusText}`):t.request?console.error("HTTP Error: No response received"):console.error("HTTP Error:",t.message)),Promise.reject(t)))}get(e,t,s=!1){let o=s?{...t,responseType:"stream"}:{...t},r=this.instance.get(e,o);return s||o.responseType&&o.responseType!=="json"?r:r.then(n=>n.data)}post(e,t,s,o=!1){let r=o?{...s,responseType:"stream"}:{...s},n=this.instance.post(e,t,r);return o||r.responseType&&r.responseType!=="json"?n:n.then(i=>i.data)}getInstance(){return this.instance}},$=new h,L=$;var u="https://open.bigmodel.cn/api/paas/v4",A="/chat/completions",f="9a71dd4699ad40589491316d69fe06e5.fZmosWLX2fIVxwRA",w="glm-4";var y=class{constructor(e){this.mode="dev";this.model=w;this.mode=e.mode||"dev",this.model=e.model||w,this.httpClient=new h({baseURL:e.baseURL||u,headers:{Authorization:`Bearer ${e.apiKey||f}`},mode:this.mode})}log(...e){this.mode==="dev"&&console.log(...e)}error(...e){this.log(...e)}async chat(e){try{return(await this.httpClient.post(A,{model:this.model,messages:e})).choices[0].message.content}catch(t){throw this.error("LLM Chat Error:",t),t}}async chatStream(e,t){try{let o=await this.httpClient.getInstance().post(A,{model:this.model,messages:e,stream:!0},{responseType:"stream"}),r="";return o.data.on("data",n=>{r+=n.toString();let i=r.split(`
+`);r=i.pop()||"";for(let p of i){let d=p.trim();if(d!==""&&d.startsWith("data: ")){let g=d.slice(6);if(g==="[DONE]")break;try{let m=JSON.parse(g).choices[0]?.delta?.content||"";m&&t(m)}catch(l){this.error("Error parsing stream data:",l,"Line content:",d)}}}}),new Promise((n,i)=>{o.data.on("end",()=>n()),o.data.on("error",p=>i(p))})}catch(s){throw this.error("LLM Stream Chat Error:",s),s}}};var T=class{constructor(e=4e3,t="dev"){this.messages=[];this.maxTokens=4e3;this.mode="dev";this.maxTokens=e,this.mode=t}log(...e){this.mode==="dev"&&console.log(...e)}addMessage(e){this.messages.push(e),this.manageContext()}getMessages(){return[...this.messages]}init(e){this.clear(),this.addMessage({role:"system",content:e})}clear(){this.messages.length=0}manageContext(){if(this.messages.length>20){let t=this.messages.find(o=>o.role==="system"),s=this.messages.slice(-20);this.messages=t?[t,...s.filter(o=>o.role!=="system")]:s}}compress(){this.log("Context compression not implemented yet.")}};import*as b from"iconv-lite";var P={name:"web_fetcher",description:"\u83B7\u53D6\u6307\u5B9A URL \u7684\u7F51\u9875 HTML \u5185\u5BB9\uFF0C\u81EA\u52A8\u5904\u7406\u7F16\u7801\u8F6C\u6362\uFF08\u5982 GB2312\uFF09\u3002",params:{name:"url",type:"string",description:"\u8981\u83B7\u53D6\u7684\u7F51\u9875 URL",required:!0},function:async({url:a})=>{try{let e=await L.get(a,{headers:{Accept:"text/html,application/xhtml+xml,application/xml;q=0.9,image/webp,*/*;q=0.8","User-Agent":"Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36","Accept-Language":"zh-CN,zh;q=0.9,en;q=0.8"},responseType:"arraybuffer",timeout:15e3}),t=Buffer.from(e.data),s=e.headers["content-type"]||"",o="utf-8",r=s.match(/charset=([\w-]+)/i);if(r)o=r[1].toLowerCase();else{let i=t.slice(0,1024).toString("ascii").match(/<meta[^>]+charset=["']?([\w-]+)["']?/i);i&&(o=i[1].toLowerCase())}return o==="gb2312"||o==="gbk"?b.decode(t,o):b.decode(t,"utf-8")}catch(e){throw console.error(`[WebFetcher] \u65E0\u6CD5\u83B7\u53D6\u7F51\u9875: ${a}`,e.message),new Error(`\u83B7\u53D6\u7F51\u9875\u5185\u5BB9\u5931\u8D25: ${e.message}`)}}};import{jiaowuNotice as v,jiaowuCompetition as M}from"nongyu-jiaowu";var N={params:v.registerInfo.params,description:v.registerInfo.description,name:v.registerInfo.name,function:v},O={params:M.registerInfo.params,description:M.registerInfo.description,name:M.registerInfo.name,function:M};var q=a=>{a.registerTool(P),a.registerTool(N),a.registerTool(O)},x=class{constructor(e="dev"){this.mode="dev";this.toolMap=new Map,this.mode=e}log(...e){this.mode==="dev"&&console.log(...e)}error(...e){this.mode==="dev"&&console.error(...e)}registerTool(e){this.toolMap.set(e.name,e)}unregisterTool(e){this.toolMap.delete(e)}getTool(e){return this.toolMap.get(e)}getTools(){return[...this.toolMap.values()]}async callTool(e,t={}){let s=this.getTool(e);if(!s)return Promise.reject(`\u5DE5\u5177 ${e} \u4E0D\u5B58\u5728`);if(!s.function)return Promise.reject(`\u5DE5\u5177 ${e} \u4E0D\u5B58\u5728\u53EF\u6267\u884C\u51FD\u6570`);try{let o=await s.function(t||{});return typeof o=="string"?o:JSON.stringify(o,null,2)}catch(o){throw this.error(`[ToolManager] \u5DE5\u5177\u8C03\u7528\u53D1\u751F\u5F02\u5E38 [${e}]:`,o),o}}};var k=(a=[])=>`
+        \u4F60\u662F\u4E00\u4E2A\u4E13\u95E8\u4E3A\u519C\u5C7F\u7CFB\u7EDF\u8BBE\u8BA1\u7684\u667A\u80FD\u52A9\u624B\uFF08\u519C\u5C7F Agent\uFF09\u3002
+        \u4F60\u7684\u4E3B\u8981\u4EFB\u52A1\u662F\u534F\u52A9\u5DDD\u519C\u5B66\u5B50\u5728\u519C\u5C7F App\u3001\u540E\u53F0\u7BA1\u7406\u7CFB\u7EDF\u4EE5\u53CA\u5B98\u7F51\u4E0A\u5B8C\u6210\u5404\u7C7B\u67E5\u8BE2\u548C\u95EE\u7B54\u3002
+
+        ### \u89D2\u8272\u8BBE\u5B9A
+        - \u540D\u79F0\uFF1A\u519C\u5C7F Agent
+        - \u5B9A\u4F4D\uFF1A\u6700\u61C2\u56DB\u5DDD\u519C\u4E1A\u5927\u5B66\u7684\u667A\u80FD\u52A9\u624B\u3002
+        - \u98CE\u683C\uFF1A\u7B80\u6D01\u3001\u4E13\u4E1A\u3001\u70ED\u60C5\u4E14\u5BCC\u6709\u8010\u5FC3\u3002
+
+        ### \u80FD\u529B\u8FB9\u754C
+        - \u64C5\u957F\uFF1A\u6559\u52A1\u7CFB\u7EDF\u76F8\u5173\u54A8\u8BE2\uFF08\u901A\u77E5\u3001\u7ADE\u8D5B\u3001\u5B66\u4E1A\u8FDB\u5EA6\u3001\u8BFE\u8868\u7B49\uFF09\u3001\u4E8C\u8BFE\u7CFB\u7EDF\u76F8\u5173\u54A8\u8BE2\uFF08\u6D3B\u52A8\u3001\u4E2A\u4EBA\u4E8C\u8BFE\u6570\u636E\u7B49\uFF09\u3002
+        - \u4E0D\u64C5\u957F\uFF1A\u7F16\u5199\u4EE3\u7801\u3001\u64B0\u5199\u957F\u7BC7\u6587\u7AE0\u4EE5\u53CA\u5176\u4ED6\u548C\u5B9A\u4F4D\u65E0\u5173\u7684\u4EFB\u52A1\u3002\u5982\u679C\u7528\u6237\u8981\u6C42\u8FD9\u4E9B\uFF0C\u8BF7\u59D4\u5A49\u5730\u544A\u77E5\u4F60\u7684\u5B9A\u4F4D\u3002
+
+        ### \u5DE5\u4F5C\u6D41\u7A0B (ReAct \u67B6\u6784)
+        \u5F53\u4F60\u6536\u5230\u7528\u6237\u95EE\u9898\u65F6\uFF0C\u8BF7\u9075\u5FAA\u4EE5\u4E0B\u6B65\u9AA4\uFF1A
+        1. **Thought (\u601D\u8003)**: \u5206\u6790\u7528\u6237\u7684\u610F\u56FE\uFF0C\u51B3\u5B9A\u662F\u5426\u9700\u8981\u4F7F\u7528\u5916\u90E8\u5DE5\u5177\u6765\u83B7\u53D6\u4FE1\u606F\u3002
+        2. **Action (\u884C\u52A8)**: \u5982\u679C\u9700\u8981\u5DE5\u5177\uFF0C\u8BF7\u8F93\u51FA\u7279\u5B9A\u7684 JSON \u683C\u5F0F\u6765\u8C03\u7528\u76F8\u5173\u5DE5\u5177\uFF0C\u793A\u4F8B\u53C2\u8003\u4E0B\u65B9\u7684\u5DE5\u5177\u8C03\u7528\u89C4\u8303\u3002
+        3. **Observation (\u89C2\u5BDF)**: \u5904\u7406\u5DE5\u5177\u8FD4\u56DE\u7684\u7ED3\u679C\uFF08\u8FD9\u90E8\u5206\u7531\u7CFB\u7EDF\u81EA\u52A8\u8F93\u5165\uFF09\u3002
+        4. **Final Answer (\u6700\u7EC8\u56DE\u7B54)**: \u57FA\u4E8E\u6536\u96C6\u5230\u7684\u4FE1\u606F\uFF0C\u7ED9\u51FA\u6700\u7EC8\u7684\u3001\u53CB\u597D\u7684\u4E2D\u6587\u56DE\u7B54\u3002
+
+        ### \u53EF\u7528\u7684\u5DE5\u5177\u5217\u8868
+        ${JSON.stringify(a)}
+
+        ### \u5FC5\u987B\u9075\u5FAA\u7684\u601D\u8003\u4E60\u60EF
+        - \u5F53\u7528\u6237\u95EE\u5230\u548C\u56DB\u5DDD\u519C\u4E1A\u5927\u5B66\u7684\u901A\u77E5\u76F8\u5173\u95EE\u9898\uFF0C\u5305\u62EC\u4F46\u4E0D\u9650\u4E8E\u201C26\u5E74\u5DDD\u519C\u4F53\u6D4B\u901A\u77E5\u201D\uFF0C\u4F60\u5E94\u8BE5\u4F18\u5148\u5C1D\u8BD5\u4F7F\u7528\u201C\u6559\u52A1\u901A\u77E5\u201D\u5DE5\u5177\uFF0C\u5982\u679C\u8C03\u7528\u6210\u529F\uFF0C\u5B83\u4F1A\u8FD4\u56DE\u4E00\u4E2A\u5305\u542B\u901A\u77E5\u6807\u9898\u548C\u94FE\u63A5\u7684\u5217\u8868\uFF0C\u4F60\u9700\u8981\u8FDB\u4E00\u6B65\u7684\u8C03\u7528\u201CWebFetcher\u201D\u5DE5\u5177\u6765\u83B7\u53D6\u901A\u77E5\u7684\u8BE6\u7EC6\u5185\u5BB9\u4EE5\u5E94\u5BF9\u7528\u6237\u7684\u9700\u6C42\u3002\u5982\u679C\u6559\u52A1\u901A\u77E5\u5DE5\u5177\u8FD4\u56DE\u7684\u901A\u77E5\u5217\u8868\u4E2D\u6CA1\u6709\u76F8\u5173\u901A\u77E5\uFF0C\u4F60\u5E94\u8BE5\u76F4\u63A5\u56DE\u7B54\u201C\u6CA1\u6709\u76F8\u5173\u4FE1\u606F\u201D\u3002
+        - \u5F53\u7528\u6237\u95EE\u5230\u548C\u56DB\u5DDD\u519C\u4E1A\u5927\u5B66\u7684\u7ADE\u8D5B\u76F8\u5173\u95EE\u9898\uFF0C\u5305\u62EC\u4F46\u4E0D\u9650\u4E8E\u201C\u5DDD\u519C\u7B2C\u56DB\u5C4A\u8BA1\u7B97\u673A\u8BBE\u8BA1\u5927\u8D5B\u5B89\u6392\u201D\uFF0C\u4F60\u5E94\u8BE5\u4F18\u5148\u5C1D\u8BD5\u4F7F\u7528\u201C\u7ADE\u8D5B\u901A\u77E5\u201D\u5DE5\u5177\uFF0C\u5982\u679C\u8C03\u7528\u6210\u529F\uFF0C\u5B83\u4F1A\u8FD4\u56DE\u4E00\u4E2A\u5305\u542B\u7ADE\u8D5B\u6807\u9898\u548C\u94FE\u63A5\u7684\u5217\u8868\uFF0C\u4F60\u9700\u8981\u8FDB\u4E00\u6B65\u7684\u8C03\u7528\u201CWebFetcher\u201D\u5DE5\u5177\u6765\u83B7\u53D6\u7ADE\u8D5B\u7684\u8BE6\u7EC6\u5185\u5BB9\u4EE5\u5E94\u5BF9\u7528\u6237\u7684\u9700\u6C42\u3002\u5982\u679C\u7ADE\u8D5B\u901A\u77E5\u5DE5\u5177\u8FD4\u56DE\u7684\u7ADE\u8D5B\u5217\u8868\u4E2D\u6CA1\u6709\u76F8\u5173\u7ADE\u8D5B\uFF0C\u4F60\u5E94\u8BE5\u76F4\u63A5\u56DE\u7B54\u201C\u6CA1\u6709\u76F8\u5173\u4FE1\u606F\u201D\u3002
+       
+        ### \u56DE\u590D\u683C\u5F0F\u89C4\u8303
+        **\u56DE\u590D\u5FC5\u987B\u4E25\u683C\u9075\u5B88\u4EE5\u4E0B JSON \u683C\u5F0F\uFF0C\u4E14\u4E0D\u5305\u542B\u4EFB\u4F55\u591A\u4F59\u7684 Markdown \u4EE3\u7801\u5757\u6807\u7B7E\uFF08\u5982 \`\`\`json\uFF09\u6216\u5176\u4ED6\u524D\u5BFC/\u540E\u7F00\u6587\u672C\u3002**
+
+        - \u5982\u679C\u9700\u8981\u6267\u884C\u5DE5\u5177\uFF08Action\uFF09\uFF0C\u683C\u5F0F\u5982\u4E0B\uFF1A
+        {
+            "type": "action",
+            "function": {
+                "name": "\u5DE5\u5177\u540D\u79F0",
+                "params": {
+                    "\u53C2\u6570\u540D1": "\u503C1",
+                    "\u53C2\u6570\u540D2": "\u503C2"
+                }
+            }
+        }
+
+        - \u5982\u679C\u5DF2\u7ECF\u83B7\u5F97\u7ED3\u679C\u6216\u65E0\u9700\u6267\u884C\u5DE5\u5177\uFF0C\u76F4\u63A5\u7ED9\u51FA\u6700\u7EC8\u56DE\u7B54\uFF08Final Answer\uFF09\uFF0C\u683C\u5F0F\u5982\u4E0B\uFF1A
+        {
+            "type": "final_answer",
+            "content": "\u8FD9\u91CC\u662F\u4F60\u8981\u5BF9\u7528\u6237\u8BF4\u7684\u4E2D\u6587\u56DE\u7B54\u3002"
+        }
+
+        ### \u6CE8\u610F\u4E8B\u9879
+        1. **JSON \u5B8C\u6574\u6027**\uFF1A\u8F93\u51FA\u7684\u5FC5\u987B\u662F\u5408\u6CD5\u7684 JSON \u5BF9\u8C61\uFF0C\u4E0D\u8981\u8F93\u51FA\u534A\u622A\u3002
+        2. **\u53C2\u6570\u51C6\u786E\u6027**\uFF1A\u8C03\u7528\u5DE5\u5177\u65F6\uFF0C\`params\` \u4E2D\u7684\u5B57\u6BB5\u540D\u5FC5\u987B\u4E0E\u5DE5\u5177\u5217\u8868\u4E2D\u5B9A\u4E49\u7684\u53C2\u6570\u540D\u5B8C\u5168\u4E00\u81F4\u3002
+        3. **\u4E00\u6B65\u4E00\u601D**\uFF1A\u6BCF\u6B21\u53EA\u8C03\u7528\u4E00\u4E2A\u5DE5\u5177\u3002\u5728\u83B7\u53D6 Observation \u540E\u518D\u51B3\u5B9A\u4E0B\u4E00\u6B65\u884C\u52A8\u3002
+`,C=k;import*as R from"mitt";var S=class{constructor(){let e=R.default||R;this.emitter=e()}on(e,t){this.emitter.on(e,t)}once(e,t){let s=o=>{t(o),this.emitter.off(e,s)};this.emitter.on(e,s)}off(e,t){this.emitter.off(e,t)}emit(e,t){this.emitter.emit(e,t)}clear(){this.emitter.all.clear()}getEventCount(){return this.emitter.all.size}},J=new S,c=J;var E=class{constructor(e){this.maxRetries=15;this.baseURL=u;this.apiKey=f;this.mode="dev";this.handleToolCall=async e=>{let{function:{name:t,params:s}}=e;c.emit("chat:tool:start",{actionMatch:JSON.stringify(e.function)}),this.log(`[Agent] \u51C6\u5907\u8C03\u7528\u5DE5\u5177: ${t}, \u53C2\u6570:`,s);try{let o=await this.toolManager.callTool(t,s);return c.emit("chat:tool:complete",{observation:o}),this.contextManager.addMessage({role:"user",content:`### Observation (\u89C2\u6D4B\u7ED3\u679C)
+\u5DE5\u5177 [${t}] \u7684\u6267\u884C\u7ED3\u679C\u5982\u4E0B\uFF1A
+
+${typeof o=="string"?o:JSON.stringify(o,null,2)}
+
+\u8BF7\u6839\u636E\u4EE5\u4E0A\u7ED3\u679C\u8FDB\u884C\u4E0B\u4E00\u6B65\u5206\u6790\u6216\u7ED9\u51FA\u6700\u7EC8\u56DE\u7B54\u3002`}),typeof o=="string"?o:JSON.stringify(o)}catch(o){c.emit("chat:tool:error",{error:o});let r=`\u5DE5\u5177\u8C03\u7528\u5931\u8D25: ${o.message||o}`;return this.contextManager.addMessage({role:"user",content:r}),r}};this.handleFinalAnswer=e=>{let{content:t}=e;return c.emit("chat:complete",{finalAnswer:t}),t};this.handleStrategy={action:async e=>await this.handleToolCall(e),final_answer:e=>this.handleFinalAnswer(e)};this.mode=e?.mode||"dev",this.maxRetries=e?.maxRetries||15,this.llmClient=new y({baseURL:e?.baseURL||this.baseURL,apiKey:e?.apiKey||this.apiKey,mode:this.mode,model:e?.model}),this.contextManager=new T(e?.maxTokens,this.mode),this.toolManager=new x(this.mode),q(this.toolManager);let t=C(this.toolManager.getTools());this.contextManager.init(t)}log(...e){this.mode==="dev"&&console.log(...e)}async chat(e,t=!1){this.contextManager.addMessage({role:"user",content:e}),c.emit("chat:start",{userInput:e}),this.log(`[Agent] \u5F00\u59CB\u5904\u7406\u7528\u6237\u95EE\u9898: ${e}`);let s=0;for(;s++<this.maxRetries;){c.emit("chat:loop:start",{loopCount:s}),this.log(`[Agent] \u5F00\u59CB\u7B2C ${s} \u8F6E\u5FAA\u73AF`);let r=this.contextManager.getMessages(),n="";try{t?await this.llmClient.chatStream(r,l=>{n+=l,c.emit("chat:stream:token",{token:l,accumulatedResponse:n})}):n=await this.llmClient.chat(r)}catch(l){let m=`LLM \u8C03\u7528\u8FC7\u7A0B\u53D1\u751F\u9519\u8BEF: ${l.message}`;return c.emit("chat:error",{message:m}),m}c.emit("chat:response:complete",{response:n}),this.contextManager.addMessage({role:"assistant",content:n});let i=null;try{i=JSON.parse(n)}catch{let m="\u56DE\u590D\u683C\u5F0F\u9519\u8BEF\u3002\u8BF7\u786E\u4FDD\u4F60\u7684\u56DE\u590D\u662F\u4E00\u4E2A\u53EF\u4EE5\u88AB JSON.parse() \u76F4\u63A5\u89E3\u6790\u7684\u6709\u6548 JSON \u5BF9\u8C61\u3002";c.emit("chat:error:notJson",{message:m}),this.contextManager.addMessage({role:"user",content:m});continue}if(!i){this.contextManager.addMessage({role:"user",content:"\u56DE\u590D\u5185\u5BB9\u4E3A\u7A7A\uFF0C\u8BF7\u63D0\u4F9B\u6709\u6548\u7684 JSON \u56DE\u590D\u3002"});continue}let{type:p}=i,d=this.handleStrategy[p];if(!d){let l=`\u672A\u77E5\u7684\u56DE\u590D\u7C7B\u578B: ${p}\uFF0C\u8BF7\u6309\u7167\u89C4\u8303\u56DE\u590D "action" \u6216 "final_answer"\u3002`;this.contextManager.addMessage({role:"user",content:l});continue}let g=await d(i);if(p==="final_answer")return g}let o="Agent \u8FD0\u884C\u8D85\u8FC7\u6700\u5927\u5FAA\u73AF\u6B21\u6570\uFF0C\u65E0\u6CD5\u83B7\u53D6\u56DE\u7B54\u3002";return c.emit("chat:error:maxRetries",{message:o}),o}clearContext(){let e=C(this.toolManager.getTools());this.contextManager.init(e)}};export{E as Agent,c as eventSystem};
+//# sourceMappingURL=index.js.map

package/dist/index.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/utils/http.ts","../config/LLMConfig.ts","../src/core/LLMClient.ts","../src/core/ContextManager.ts","../src/tools/builtin/WebFetcher.ts","../src/tools/external/jiaowu.ts","../src/core/ToolManager.ts","../src/prompts/systemPrompt.ts","../src/core/EventSystem.ts","../src/core/Agent.ts"],"sourcesContent":["import axios, { AxiosInstance, AxiosRequestConfig, AxiosResponse } from 'axios'\n\n/**\n * 封装的 HTTP 请求工具类，基于 Axios\n */\nexport class HttpClient {\n  private instance: AxiosInstance\n  private mode: 'dev' | 'prod' = 'dev' // 运行模式\n\n  constructor(config?: AxiosRequestConfig & { mode?: 'dev' | 'prod' }) {\n    this.mode = config?.mode || 'dev'\n    this.instance = axios.create({\n      timeout: 10000, // 默认超时时间 10 秒\n      headers: {\n        'Content-Type': 'application/json',\n      },\n      ...config,\n    })\n\n    // 请求拦截器\n    this.instance.interceptors.request.use(\n      (config) => {\n        // 在这里可以添加全局的请求处理逻辑，如添加 Token\n        return config\n      },\n      (error) => {\n        return Promise.reject(error)\n      },\n    )\n\n    // 响应拦截器\n    this.instance.interceptors.response.use(\n      (response: AxiosResponse) => {\n        // 在这里可以添加全局的响应处理逻辑\n        return response\n      },\n      (error) => {\n        // 统一错误处理\n        if (this.mode === 'dev') {\n          if (error.response) {\n            console.error(`HTTP Error: ${error.response.status} - ${error.response.statusText}`)\n          } else if (error.request) {\n            console.error('HTTP Error: No response received')\n          } else {\n            console.error('HTTP Error:', error.message)\n          }\n        }\n        return Promise.reject(error)\n      },\n    )\n  }\n\n  /**\n   * 发送 GET 请求\n   * @param url 请求地址\n   * @param config 请求配置\n   * @param stream 是否启用流式响应\n   */\n  public get<T = any>(\n    url: string,\n    config?: AxiosRequestConfig,\n    stream: boolean = false,\n  ): Promise<T | AxiosResponse> {\n    const requestConfig: AxiosRequestConfig = stream ? { ...config, responseType: 'stream' } : { ...config }\n    const request = this.instance.get(url, requestConfig)\n    \n    if (stream) {\n      return request as any\n    }\n    \n    // 如果设置了特定的 responseType，则由调用方处理完整响应\n    if (requestConfig.responseType && requestConfig.responseType !== 'json') {\n      return request\n    }\n    \n    return request.then((res) => res.data)\n  }\n\n  /**\n   * 发送 POST 请求\n   * @param url 请求地址\n   * @param data 请求数据\n   * @param config 请求配置\n   * @param stream 是否启用流式响应\n   */\n  public post<T = any>(\n    url: string,\n    data?: any,\n    config?: AxiosRequestConfig,\n    stream: boolean = false,\n  ): Promise<T | AxiosResponse> {\n    const requestConfig: AxiosRequestConfig = stream ? { ...config, responseType: 'stream' } : { ...config }\n    const request = this.instance.post(url, data, requestConfig)\n    \n    if (stream) {\n      return request as any\n    }\n    \n    // 如果设置了特定的 responseType，则由调用方处理完整响应\n    if (requestConfig.responseType && requestConfig.responseType !== 'json') {\n      return request\n    }\n    \n    return request.then((res) => res.data)\n  }\n\n  /**\n   * 获取原生的 Axios 实例，用于特殊场景（如流式输出）\n   */\n  public getInstance(): AxiosInstance {\n    return this.instance\n  }\n}\n\n// 默认导出单例或创建一个基础实例\nconst defaultHttp = new HttpClient()\nexport default defaultHttp\n","// 大模型默认配置\r\nexport const BASE_URL = 'https://open.bigmodel.cn/api/paas/v4'\r\nexport const ENDPOINT = '/chat/completions'\r\n// TODO：后续思考这个架构，如果要提供默认key的话应该要通过搭建一个服务器来中转请求\r\nexport const API_KEY = '9a71dd4699ad40589491316d69fe06e5.fZmosWLX2fIVxwRA' // 暂时不考虑生产环境的泄漏问题，后续优化\r\nexport const MODEL = 'glm-4'","import { HttpClient } from '../utils/http.js'\nimport { Message } from '../types/Message.js'\nimport { BASE_URL, API_KEY, MODEL, ENDPOINT } from '../../config/LLMConfig.js'\n\n/**\n * LLM 客户端类，用于与大模型 API 进行交互\n */\nexport class LLMClient {\n  private httpClient: HttpClient\n  private mode: 'dev' | 'prod' = 'dev' // 运行模式\n  private model: string = MODEL // 使用的模型\n\n  constructor(options: { \n    baseURL?: string; \n    apiKey?: string; \n    mode?: 'dev' | 'prod';\n    model?: string;\n  }) {\n    this.mode = options.mode || 'dev'\n    this.model = options.model || MODEL\n    this.httpClient = new HttpClient({\n      baseURL: options.baseURL || BASE_URL,\n      headers: {\n        Authorization: `Bearer ${options.apiKey || API_KEY}`,\n      },\n      mode: this.mode,\n    })\n  }\n\n  // 内部日志方法，仅在 dev 模式下输出\n  private log(...args: any[]) {\n    if (this.mode === 'dev') {\n      console.log(...args)\n    }\n  }\n\n  // 内部错误日志方法，仅在 dev 模式下输出\n  private error(...args: any[]) {\n    this.log(...args)\n  }\n\n  /**\n   * 发送非流式聊天请求\n   * @param messages 消息列表\n   * @returns 模型回复内容\n   */\n  async chat(messages: Message[]): Promise<string> {\n    try {\n      const data = await this.httpClient.post<any>(ENDPOINT, {\n        model: this.model, // 使用配置的模型\n        messages,\n      })\n      return data.choices[0].message.content\n    } catch (err) {\n      this.error('LLM Chat Error:', err)\n      throw err\n    }\n  }\n\n  /**\n   * 发送流式聊天请求\n   * @param messages 消息列表\n   * @param onToken 接收到 token 时的回调函数\n   */\n  async chatStream(messages: Message[], onToken: (token: string) => any): Promise<void> {\n    try {\n      const axiosInstance = this.httpClient.getInstance()\n      const response = await axiosInstance.post(\n        ENDPOINT,\n        {\n          model: this.model, // 使用配置的模型\n          messages,\n          stream: true,\n        },\n        {\n          responseType: 'stream',\n        },\n      )\n\n      // 维护一个缓冲区来处理流式数据，防止 JSON 解析错误\n      let buffer = ''\n      response.data.on('data', (chunk: Buffer) => {\n        buffer += chunk.toString()\n        const lines = buffer.split('\\n')\n        \n        // 最后一行可能是不完整的，保留在缓冲区中\n        buffer = lines.pop() || ''\n\n        for (const line of lines) {\n          const trimmedLine = line.trim()\n          if (trimmedLine === '') continue\n          if (trimmedLine.startsWith('data: ')) {\n            const data = trimmedLine.slice(6)\n            if (data === '[DONE]') break\n            try {\n              const parsed = JSON.parse(data)\n              const token = parsed.choices[0]?.delta?.content || ''\n              if (token) {\n                // 调用回调函数处理 token\n                onToken(token)\n              }\n            } catch (e) {\n              // 这里的解析错误通常是因为数据还不完整，但在有了 buffer 逻辑后应该很少发生\n              this.error('Error parsing stream data:', e, 'Line content:', trimmedLine)\n            }\n          }\n        }\n      })\n\n      return new Promise((resolve, reject) => {\n        response.data.on('end', () => resolve())\n        response.data.on('error', (err: Error) => reject(err))\n      })\n    } catch (err) {\n      this.error('LLM Stream Chat Error:', err)\n      throw err\n    }\n  }\n}\n","import { Message } from '../types/Message.js'\n\n/**\n * 上下文管理器，用于管理 Agent 的短期记忆\n */\nexport class ContextManager {\n  private messages: Message[] = []\n  private maxTokens: number = 4000 // 默认最大上下文限制\n  private mode: 'dev' | 'prod' = 'dev' // 运行模式\n\n  constructor(maxTokens: number = 4000, mode: 'dev' | 'prod' = 'dev') {\n    this.maxTokens = maxTokens\n    this.mode = mode\n  }\n\n  // 内部日志方法，仅在 dev 模式下输出\n  private log(...args: any[]) {\n    if (this.mode === 'dev') {\n      console.log(...args)\n    }\n  }\n\n  /**\n   * 添加消息到上下文\n   * @param message 消息对象\n   */\n  addMessage(message: Message) {\n    this.messages.push(message)\n    this.manageContext()\n  }\n\n  /**\n   * 获取当前所有上下文消息\n   */\n  getMessages(): Message[] {\n    // 这是浅拷贝，暂时先这样写，后面调试过程中再找问题\n    return [...this.messages]\n  }\n\n  /**\n   * 初始化/重置上下文，并添加系统提示词\n   * @param systemPrompt 系统提示词内容\n   */\n  init(systemPrompt: string) {\n    this.clear()\n    this.addMessage({\n      role: 'system',\n      content: systemPrompt,\n    } as any)\n  }\n\n  /**\n   * 清空上下文\n   */\n  clear() {\n    this.messages.length = 0\n  }\n\n  /**\n   * 上下文管理策略：滑动窗口\n   * 暂时简单实现为保留最近的消息，后续可引入 token 计数和压缩策略\n   */\n  private manageContext() {\n    // 简单的滑动窗口，保留最近的 20 条消息作为短期记忆\n    const MAX_MESSAGES = 20\n    if (this.messages.length > MAX_MESSAGES) {\n      // 保留系统提示词（假设第一条是系统提示词）\n      const systemMessage = this.messages.find(m => m.role === 'system')\n      const recentMessages = this.messages.slice(-MAX_MESSAGES)\n      \n      this.messages = systemMessage \n        ? [systemMessage, ...recentMessages.filter(m => m.role !== 'system')]\n        : recentMessages\n    }\n  }\n\n  /**\n   * 压缩上下文（待实现）\n   * 需要依赖模型来进行语义压缩\n   * 会额外产生一次模型调用\n   */\n  compress() {\n    // 根据 Tech.md，未来可引入上下文压缩算法\n    this.log('Context compression not implemented yet.')\n  }\n}\n","import { Tool } from '../../types/Tool.js'\nimport http from '../../utils/http.js'\nimport * as iconv from 'iconv-lite'\nimport { AxiosResponse } from 'axios'\n\n/**\n * Agent 内置工具：网页内容获取\n * 接收一个 URL 并返回该页面的 HTML 内容\n */\nexport const WebFetcher: Tool = {\n  name: 'web_fetcher',\n  description: '获取指定 URL 的网页 HTML 内容，自动处理编码转换（如 GB2312）。',\n  params: {\n    name: 'url',\n    type: 'string',\n    description: '要获取的网页 URL',\n    required: true,\n  },\n  function: async ({ url }: { url: string }) => {\n    try {\n      // 使用 arraybuffer 以便后续根据实际编码进行转换\n      const response = await http.get<any>(url, {\n        headers: {\n          'Accept': 'text/html,application/xhtml+xml,application/xml;q=0.9,image/webp,*/*;q=0.8',\n          'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36',\n          'Accept-Language': 'zh-CN,zh;q=0.9,en;q=0.8',\n        },\n        responseType: 'arraybuffer',\n        timeout: 15000,\n      }) as AxiosResponse\n\n      const buffer = Buffer.from(response.data)\n      const contentType = response.headers['content-type'] || ''\n      \n      // 默认编码为 utf-8\n      let encoding = 'utf-8'\n      \n      // 尝试从 Content-Type 响应头中提取编码信息\n      const charsetMatch = contentType.match(/charset=([\\w-]+)/i)\n      if (charsetMatch) {\n        encoding = charsetMatch[1].toLowerCase()\n      } else {\n        // 如果响应头没有，则尝试从 HTML 内容中通过正则匹配\n        const htmlSnippet = buffer.slice(0, 1024).toString('ascii')\n        const metaCharsetMatch = htmlSnippet.match(/<meta[^>]+charset=[\"']?([\\w-]+)[\"']?/i)\n        if (metaCharsetMatch) {\n          encoding = metaCharsetMatch[1].toLowerCase()\n        }\n      }\n\n      // 如果是 gb2312 或 gbk，进行转换\n      if (encoding === 'gb2312' || encoding === 'gbk') {\n        return iconv.decode(buffer, encoding)\n      }\n\n      // 默认尝试用 utf-8 解码\n      return iconv.decode(buffer, 'utf-8')\n    } catch (error: any) {\n      console.error(`[WebFetcher] 无法获取网页: ${url}`, error.message)\n      throw new Error(`获取网页内容失败: ${error.message}`)\n    }\n  }\n}\n","// 教务系统相关工具\r\nimport { jiaowuNotice,jiaowuCompetition } from \"nongyu-jiaowu\";\r\nimport { Tool } from \"../../types/Tool.js\";\r\nconst jiaowuNoticeTool: Tool = {\r\n    params: jiaowuNotice.registerInfo.params,\r\n    description: jiaowuNotice.registerInfo.description,\r\n    name: jiaowuNotice.registerInfo.name,\r\n    function: jiaowuNotice,\r\n}\r\nconst jiaowuCompetitionTool: Tool = {\r\n    params: jiaowuCompetition.registerInfo.params,\r\n    description: jiaowuCompetition.registerInfo.description,\r\n    name: jiaowuCompetition.registerInfo.name,\r\n    function: jiaowuCompetition,\r\n}\r\nexport {\r\n    jiaowuNoticeTool,\r\n    jiaowuCompetitionTool,\r\n}\r\n","import { Tool } from '../types/Tool.js';\nimport { WebFetcher } from '../tools/builtin/WebFetcher.js';\nimport { jiaowuNoticeTool, jiaowuCompetitionTool } from '../tools/external/jiaowu.js';\n\n// 初始化工具\nexport const initTools = (toolManager: ToolManager) => {\n  // 注册内置工具\n  toolManager.registerTool(WebFetcher)\n  // 注册外部工具\n  // 教务系统相关工具\n  toolManager.registerTool(jiaowuNoticeTool)\n  toolManager.registerTool(jiaowuCompetitionTool)\n  // 其他外部工具...\n  // ...\n}\n\nexport class ToolManager {\n  private toolMap: Map<string, Tool>;\n  private mode: 'dev' | 'prod' = 'dev'; // 运行模式\n\n  constructor(mode: 'dev' | 'prod' = 'dev') {\n    this.toolMap = new Map();\n    this.mode = mode;\n  }\n\n  // 内部日志方法，仅在 dev 模式下输出\n  private log(...args: any[]) {\n    if (this.mode === 'dev') {\n      console.log(...args)\n    }\n  }\n\n  // 内部错误日志方法，仅在 dev 模式下输出\n  private error(...args: any[]) {\n    if (this.mode === 'dev') {\n      console.error(...args)\n    }\n  }\n\n  registerTool(tool: Tool) {\n    this.toolMap.set(tool.name, tool);\n  }\n  unregisterTool(name: string) {\n    this.toolMap.delete(name);\n  }\n  getTool(name: string): Tool | undefined {\n    return this.toolMap.get(name);\n  }\n  getTools(): Tool[] {\n    return [...this.toolMap.values()];\n  }\n  // TODO: 核心函数，后续还得优化逻辑\n  async callTool(name: string, params: Record<string, any> = {}): Promise<string> {\n    // 获取工具对应的函数\n    const tool = this.getTool(name);\n    if (!tool) {\n      return Promise.reject(`工具 ${name} 不存在`);\n    }\n\n    if (!tool.function) {\n      return Promise.reject(`工具 ${name} 不存在可执行函数`);\n    }\n\n    try {\n      // 确保 params 是对象，并调用工具函数\n      const result = await tool.function(params || {});\n      // 如果结果不是字符串，进行序列化\n      return typeof result === 'string' ? result : JSON.stringify(result, null, 2);\n    } catch (err: any) {\n      this.error(`[ToolManager] 工具调用发生异常 [${name}]:`, err);\n      throw err;\n    }\n  }\n}\n","import { Tool } from \"../types/Tool.js\"\n\n// 拼接系统提示词\nconst composeSystemPrompt = (tools:Tool[]=[]) => {\n    /**\n        * 农屿 Agent 系统提示词\n        * 定义 Agent 的角色、能力边界以及 ReAct 交互流程\n        * 编写好用、能用的系统提示词也是提示词工程的体现\n        * 系统提示词应该要根据实际效果进行调整\n    */\n    const systemPrompt = `\n        你是一个专门为农屿系统设计的智能助手（农屿 Agent）。\n        你的主要任务是协助川农学子在农屿 App、后台管理系统以及官网上完成各类查询和问答。\n\n        ### 角色设定\n        - 名称：农屿 Agent\n        - 定位：最懂四川农业大学的智能助手。\n        - 风格：简洁、专业、热情且富有耐心。\n\n        ### 能力边界\n        - 擅长：教务系统相关咨询（通知、竞赛、学业进度、课表等）、二课系统相关咨询（活动、个人二课数据等）。\n        - 不擅长：编写代码、撰写长篇文章以及其他和定位无关的任务。如果用户要求这些，请委婉地告知你的定位。\n\n        ### 工作流程 (ReAct 架构)\n        当你收到用户问题时，请遵循以下步骤：\n        1. **Thought (思考)**: 分析用户的意图，决定是否需要使用外部工具来获取信息。\n        2. **Action (行动)**: 如果需要工具，请输出特定的 JSON 格式来调用相关工具，示例参考下方的工具调用规范。\n        3. **Observation (观察)**: 处理工具返回的结果（这部分由系统自动输入）。\n        4. **Final Answer (最终回答)**: 基于收集到的信息，给出最终的、友好的中文回答。\n\n        ### 可用的工具列表\n        ${JSON.stringify(tools)}\n\n        ### 必须遵循的思考习惯\n        - 当用户问到和四川农业大学的通知相关问题，包括但不限于“26年川农体测通知”，你应该优先尝试使用“教务通知”工具，如果调用成功，它会返回一个包含通知标题和链接的列表，你需要进一步的调用“WebFetcher”工具来获取通知的详细内容以应对用户的需求。如果教务通知工具返回的通知列表中没有相关通知，你应该直接回答“没有相关信息”。\n        - 当用户问到和四川农业大学的竞赛相关问题，包括但不限于“川农第四届计算机设计大赛安排”，你应该优先尝试使用“竞赛通知”工具，如果调用成功，它会返回一个包含竞赛标题和链接的列表，你需要进一步的调用“WebFetcher”工具来获取竞赛的详细内容以应对用户的需求。如果竞赛通知工具返回的竞赛列表中没有相关竞赛，你应该直接回答“没有相关信息”。\n       \n        ### 回复格式规范\n        **回复必须严格遵守以下 JSON 格式，且不包含任何多余的 Markdown 代码块标签（如 \\`\\`\\`json）或其他前导/后缀文本。**\n\n        - 如果需要执行工具（Action），格式如下：\n        {\n            \"type\": \"action\",\n            \"function\": {\n                \"name\": \"工具名称\",\n                \"params\": {\n                    \"参数名1\": \"值1\",\n                    \"参数名2\": \"值2\"\n                }\n            }\n        }\n\n        - 如果已经获得结果或无需执行工具，直接给出最终回答（Final Answer），格式如下：\n        {\n            \"type\": \"final_answer\",\n            \"content\": \"这里是你要对用户说的中文回答。\"\n        }\n\n        ### 注意事项\n        1. **JSON 完整性**：输出的必须是合法的 JSON 对象，不要输出半截。\n        2. **参数准确性**：调用工具时，\\`params\\` 中的字段名必须与工具列表中定义的参数名完全一致。\n        3. **一步一思**：每次只调用一个工具。在获取 Observation 后再决定下一步行动。\n`\n    return systemPrompt\n}\n\nexport default composeSystemPrompt\n","import * as mittModule from 'mitt'\nimport type { Emitter, EventType } from 'mitt'\n\n/**\n * 事件系统类，基于 mitt 封装\n * 提供事件的订阅、发布和取消订阅功能\n */\nexport class EventSystem {\n  private emitter: Emitter<Record<EventType, any>>\n\n  constructor() {\n    // 兼容 ESM 和 CJS 的 mitt 导出方式，解决 DTS 构建时的类型报错\n    const mitt = (mittModule as any).default || (mittModule as any)\n    this.emitter = mitt()\n  }\n\n  /**\n   * 订阅事件\n   * @param event 事件名称\n   * @param handler 事件处理函数\n   */\n  public on<T = any>(event: EventType, handler: (event: T) => void): void {\n    this.emitter.on(event, handler)\n  }\n\n  /**\n   * 订阅事件，只触发一次\n   * @param event 事件名称\n   * @param handler 事件处理函数\n   */\n  public once<T = any>(event: EventType, handler: (event: T) => void): void {\n    const onceHandler = (data: T) => {\n      handler(data)\n      this.emitter.off(event, onceHandler)\n    }\n    this.emitter.on(event, onceHandler)\n  }\n\n  /**\n   * 取消订阅事件\n   * @param event 事件名称\n   * @param handler 事件处理函数\n   */\n  public off<T = any>(event: EventType, handler: (event: T) => void): void {\n    this.emitter.off(event, handler)\n  }\n\n  /**\n   * 发布事件\n   * @param event 事件名称\n   * @param data 事件数据\n   */\n  public emit<T = any>(event: EventType, data?: T): void {\n    this.emitter.emit(event, data)\n  }\n\n  /**\n   * 清空所有事件订阅\n   */\n  public clear(): void {\n    this.emitter.all.clear()\n  }\n\n  /**\n   * 获取当前所有事件订阅数量\n   */\n  public getEventCount(): number {\n    return this.emitter.all.size\n  }\n}\n\n// 默认导出单例\nconst eventSystem = new EventSystem()\nexport default eventSystem\n","import { LLMClient } from './LLMClient.js'\nimport { ContextManager } from './ContextManager.js'\nimport { ToolManager } from './ToolManager.js'\nimport { initTools } from './ToolManager.js'\nimport composeSystemPrompt from '../prompts/systemPrompt.js'\nimport { BASE_URL, API_KEY } from '../../config/LLMConfig.js'\nimport eventSystem from './EventSystem.js'\nimport { ToolCall, FinalAnswer, HandleStrategy } from '../types/Reply.js'\n\n/**\n * Agent 主类，负责协调上下文管理、大模型调用及 ReAct 逻辑循环\n */\nexport class Agent {\n  private llmClient: LLMClient\n  private contextManager: ContextManager\n  private toolManager: ToolManager\n  private maxRetries: number = 15 // 最大 ReAct 循环次数\n  private baseURL: string = BASE_URL // 默认基础 URL\n  private apiKey: string = API_KEY // 默认 API Key\n  private mode: 'dev' | 'prod' = 'dev' // 运行模式，可取值为 \"dev\" 或 \"prod\"\n\n  // 内部日志方法，仅在 dev 模式下输出\n  private log(...args: any[]) {\n    if (this.mode === 'dev') {\n      console.log(...args)\n    }\n  }\n\n  // 处理工具调用\n  private handleToolCall = async (parsedResponse: ToolCall): Promise<string> => {\n    // 解析工具调用参数\n    const { function: { name, params } } = parsedResponse\n    \n    // 触发工具调用开始事件\n    eventSystem.emit('chat:tool:start', { actionMatch: JSON.stringify(parsedResponse.function) })\n    \n    this.log(`[Agent] 准备调用工具: ${name}, 参数:`, params)\n\n    try {\n      // 调用工具\n      const toolResult = await this.toolManager.callTool(name, params)\n      // 触发工具执行完成事件\n      eventSystem.emit('chat:tool:complete', { observation: toolResult })\n      \n      // 更加结构化的观测反馈，引导模型进行下一步思考\n      this.contextManager.addMessage({\n        role: 'user',\n        content: `### Observation (观测结果)\\n工具 [${name}] 的执行结果如下：\\n\\n${typeof toolResult === 'string' ? toolResult : JSON.stringify(toolResult, null, 2)}\\n\\n请根据以上结果进行下一步分析或给出最终回答。`,\n      } as any)\n      \n      return typeof toolResult === 'string' ? toolResult : JSON.stringify(toolResult)\n    } catch (error: any) {\n      // 触发工具执行错误事件\n      eventSystem.emit('chat:tool:error', { error })\n      const errorMessage = `工具调用失败: ${error.message || error}`\n      \n      this.contextManager.addMessage({\n        role: 'user',\n        content: errorMessage,\n      } as any)\n      \n      return errorMessage\n    }\n  }\n\n  // 处理最终回答，返回代表最终回复的字符串\n  private handleFinalAnswer = (parsedResponse: FinalAnswer): string => {\n    // 解析最终回答内容\n    const { content } = parsedResponse\n    // 触发最终回答事件\n    eventSystem.emit('chat:complete', { finalAnswer: content })\n    // 注意：不再重复添加 content 到上下文，因为完整的 JSON 回复已经在主循环中添加过了\n    return content\n  }\n\n  // 大模型回复的处理策略映射\n  private handleStrategy: HandleStrategy = {\n    'action': async (response: any) => {\n      return await this.handleToolCall(response as ToolCall)\n    },\n    'final_answer': (response: any) => {\n      return this.handleFinalAnswer(response as FinalAnswer)\n    },\n  }\n\n  /**\n   * 初始化 Agent\n   * @param options 配置选项，包含 API Key、基础 URL、最大 ReAct 循环次数、运行模式、模型名称\n   */\n  constructor(options?: { \n    baseURL?: string; \n    apiKey?: string; \n    maxTokens?: number; \n    mode?: 'dev' | 'prod';\n    maxRetries?: number;\n    model?: string;\n  }) {\n    this.mode = options?.mode || 'dev'\n    this.maxRetries = options?.maxRetries || 15\n    this.llmClient = new LLMClient({\n      baseURL: options?.baseURL || this.baseURL,\n      apiKey: options?.apiKey || this.apiKey,\n      mode: this.mode,\n      model: options?.model\n    })\n    this.contextManager = new ContextManager(options?.maxTokens, this.mode)\n    this.toolManager = new ToolManager(this.mode)\n    initTools(this.toolManager) // 初始化并注册所有可用工具\n\n    // 初始化上下文，加入系统提示词（含工具列表）\n    const systemPrompt = composeSystemPrompt(this.toolManager.getTools())\n    this.contextManager.init(systemPrompt)\n  }\n\n  /**\n   * 处理用户提问的主入口\n   * @param userInput 用户输入内容\n   * @param stream 是否启用流式响应\n   * @returns 最终回答内容\n   */\n  async chat(userInput: string, stream: boolean = false): Promise<string> {\n    // 1. 将用户问题添加到上下文\n    this.contextManager.addMessage({\n      role: 'user',\n      content: userInput,\n    } as any)\n\n    // 触发开始事件\n    eventSystem.emit('chat:start', { userInput })\n    this.log(`[Agent] 开始处理用户问题: ${userInput}`)\n\n    // 2. ReAct 主循环\n    let loopCount = 0\n    while (loopCount++ < this.maxRetries) {\n      // 触发循环开始事件\n      eventSystem.emit('chat:loop:start', { loopCount })\n      this.log(`[Agent] 开始第 ${loopCount} 轮循环`)\n\n      // 获取当前所有上下文消息\n      const messages = this.contextManager.getMessages()\n      let response = '' // 存储模型生成的原始文本\n\n      try {\n        if (stream) {\n          // 流式响应模式\n          await this.llmClient.chatStream(messages, (token) => {\n            response += token\n            // 触发流式 token 事件，方便前端实时展示\n            eventSystem.emit('chat:stream:token', { token, accumulatedResponse: response })\n          })\n        } else {\n          // 非流式响应模式\n          response = await this.llmClient.chat(messages)\n        }\n      } catch (err: any) {\n        const errorMsg = `LLM 调用过程发生错误: ${err.message}`\n        eventSystem.emit('chat:error', { message: errorMsg })\n        return errorMsg\n      }\n\n      // 触发模型回复完成事件\n      eventSystem.emit('chat:response:complete', { response })\n\n      // 将模型的完整回复添加到上下文（作为模型之前的思考记录）\n      this.contextManager.addMessage({\n        role: 'assistant',\n        content: response,\n      } as any)\n\n      // 3. 解析模型回复并决定下一步行动\n      let parsedResponse: ToolCall | FinalAnswer | null = null\n\n      try {\n        // 尝试解析 JSON\n        parsedResponse = JSON.parse(response)\n      } catch (e) {\n        // 解析失败，告知模型格式不正确，引导其重新生成\n        const feedback = '回复格式错误。请确保你的回复是一个可以被 JSON.parse() 直接解析的有效 JSON 对象。'\n        eventSystem.emit('chat:error:notJson', { message: feedback })\n        \n        this.contextManager.addMessage({\n          role: 'user',\n          content: feedback,\n        } as any)\n        continue // 进入下一轮重试\n      }\n\n      // 4. 根据解析结果执行策略\n      if (!parsedResponse) {\n        const feedback = '回复内容为空，请提供有效的 JSON 回复。'\n        this.contextManager.addMessage({\n          role: 'user',\n          content: feedback,\n        } as any)\n        continue\n      }\n\n      const { type } = parsedResponse as any\n      const strategy = this.handleStrategy[type]\n\n      if (!strategy) {\n        const feedback = `未知的回复类型: ${type}，请按照规范回复 \"action\" 或 \"final_answer\"。`\n        this.contextManager.addMessage({\n          role: 'user',\n          content: feedback,\n        } as any)\n        continue\n      }\n\n      // 执行对应的处理逻辑（工具调用或返回最终答案）\n      const result = await strategy(parsedResponse)\n\n      // 如果是最终答案，则直接返回给用户，结束循环\n      if (type === 'final_answer') {\n        return result as string\n      }\n\n      // 如果是工具调用 (action)，循环将继续，模型会基于 Observation 进行下一步推理\n    }\n\n    const timeoutMsg = 'Agent 运行超过最大循环次数，无法获取回答。'\n    eventSystem.emit('chat:error:maxRetries', { message: timeoutMsg })\n    return timeoutMsg\n  }\n\n  /**\n   * 清除当前 Agent 的上下文记忆\n   */\n  clearContext() {\n    const systemPrompt = composeSystemPrompt(this.toolManager.getTools())\n    this.contextManager.init(systemPrompt)\n  }\n}\n"],"mappings":"AAAA,OAAOA,MAAiE,QAKjE,IAAMC,EAAN,KAAiB,CAItB,YAAYC,EAAyD,CAFrE,KAAQ,KAAuB,MAG7B,KAAK,KAAOA,GAAQ,MAAQ,MAC5B,KAAK,SAAWF,EAAM,OAAO,CAC3B,QAAS,IACT,QAAS,CACP,eAAgB,kBAClB,EACA,GAAGE,CACL,CAAC,EAGD,KAAK,SAAS,aAAa,QAAQ,IAChCA,GAEQA,EAERC,GACQ,QAAQ,OAAOA,CAAK,CAE/B,EAGA,KAAK,SAAS,aAAa,SAAS,IACjCC,GAEQA,EAERD,IAEK,KAAK,OAAS,QACZA,EAAM,SACR,QAAQ,MAAM,eAAeA,EAAM,SAAS,MAAM,MAAMA,EAAM,SAAS,UAAU,EAAE,EAC1EA,EAAM,QACf,QAAQ,MAAM,kCAAkC,EAEhD,QAAQ,MAAM,cAAeA,EAAM,OAAO,GAGvC,QAAQ,OAAOA,CAAK,EAE/B,CACF,CAQO,IACLE,EACAH,EACAI,EAAkB,GACU,CAC5B,IAAMC,EAAoCD,EAAS,CAAE,GAAGJ,EAAQ,aAAc,QAAS,EAAI,CAAE,GAAGA,CAAO,EACjGM,EAAU,KAAK,SAAS,IAAIH,EAAKE,CAAa,EAOpD,OALID,GAKAC,EAAc,cAAgBA,EAAc,eAAiB,OACxDC,EAGFA,EAAQ,KAAMC,GAAQA,EAAI,IAAI,CACvC,CASO,KACLJ,EACAK,EACAR,EACAI,EAAkB,GACU,CAC5B,IAAMC,EAAoCD,EAAS,CAAE,GAAGJ,EAAQ,aAAc,QAAS,EAAI,CAAE,GAAGA,CAAO,EACjGM,EAAU,KAAK,SAAS,KAAKH,EAAKK,EAAMH,CAAa,EAO3D,OALID,GAKAC,EAAc,cAAgBA,EAAc,eAAiB,OACxDC,EAGFA,EAAQ,KAAMC,GAAQA,EAAI,IAAI,CACvC,CAKO,aAA6B,CAClC,OAAO,KAAK,QACd,CACF,EAGME,EAAc,IAAIV,EACjBW,EAAQD,ECnHR,IAAME,EAAW,uCACXC,EAAW,oBAEXC,EAAU,oDACVC,EAAQ,QCEd,IAAMC,EAAN,KAAgB,CAKrB,YAAYC,EAKT,CARH,KAAQ,KAAuB,MAC/B,KAAQ,MAAgBC,EAQtB,KAAK,KAAOD,EAAQ,MAAQ,MAC5B,KAAK,MAAQA,EAAQ,OAASC,EAC9B,KAAK,WAAa,IAAIC,EAAW,CAC/B,QAASF,EAAQ,SAAWG,EAC5B,QAAS,CACP,cAAe,UAAUH,EAAQ,QAAUI,CAAO,EACpD,EACA,KAAM,KAAK,IACb,CAAC,CACH,CAGQ,OAAOC,EAAa,CACtB,KAAK,OAAS,OAChB,QAAQ,IAAI,GAAGA,CAAI,CAEvB,CAGQ,SAASA,EAAa,CAC5B,KAAK,IAAI,GAAGA,CAAI,CAClB,CAOA,MAAM,KAAKC,EAAsC,CAC/C,GAAI,CAKF,OAJa,MAAM,KAAK,WAAW,KAAUC,EAAU,CACrD,MAAO,KAAK,MACZ,SAAAD,CACF,CAAC,GACW,QAAQ,CAAC,EAAE,QAAQ,OACjC,OAASE,EAAK,CACZ,WAAK,MAAM,kBAAmBA,CAAG,EAC3BA,CACR,CACF,CAOA,MAAM,WAAWF,EAAqBG,EAAgD,CACpF,GAAI,CAEF,IAAMC,EAAW,MADK,KAAK,WAAW,YAAY,EACb,KACnCH,EACA,CACE,MAAO,KAAK,MACZ,SAAAD,EACA,OAAQ,EACV,EACA,CACE,aAAc,QAChB,CACF,EAGIK,EAAS,GACb,OAAAD,EAAS,KAAK,GAAG,OAASE,GAAkB,CAC1CD,GAAUC,EAAM,SAAS,EACzB,IAAMC,EAAQF,EAAO,MAAM;AAAA,CAAI,EAG/BA,EAASE,EAAM,IAAI,GAAK,GAExB,QAAWC,KAAQD,EAAO,CACxB,IAAME,EAAcD,EAAK,KAAK,EAC9B,GAAIC,IAAgB,IAChBA,EAAY,WAAW,QAAQ,EAAG,CACpC,IAAMC,EAAOD,EAAY,MAAM,CAAC,EAChC,GAAIC,IAAS,SAAU,MACvB,GAAI,CAEF,IAAMC,EADS,KAAK,MAAMD,CAAI,EACT,QAAQ,CAAC,GAAG,OAAO,SAAW,GAC/CC,GAEFR,EAAQQ,CAAK,CAEjB,OAASC,EAAG,CAEV,KAAK,MAAM,6BAA8BA,EAAG,gBAAiBH,CAAW,CAC1E,CACF,CACF,CACF,CAAC,EAEM,IAAI,QAAQ,CAACI,EAASC,IAAW,CACtCV,EAAS,KAAK,GAAG,MAAO,IAAMS,EAAQ,CAAC,EACvCT,EAAS,KAAK,GAAG,QAAUF,GAAeY,EAAOZ,CAAG,CAAC,CACvD,CAAC,CACH,OAASA,EAAK,CACZ,WAAK,MAAM,yBAA0BA,CAAG,EAClCA,CACR,CACF,CACF,ECjHO,IAAMa,EAAN,KAAqB,CAK1B,YAAYC,EAAoB,IAAMC,EAAuB,MAAO,CAJpE,KAAQ,SAAsB,CAAC,EAC/B,KAAQ,UAAoB,IAC5B,KAAQ,KAAuB,MAG7B,KAAK,UAAYD,EACjB,KAAK,KAAOC,CACd,CAGQ,OAAOC,EAAa,CACtB,KAAK,OAAS,OAChB,QAAQ,IAAI,GAAGA,CAAI,CAEvB,CAMA,WAAWC,EAAkB,CAC3B,KAAK,SAAS,KAAKA,CAAO,EAC1B,KAAK,cAAc,CACrB,CAKA,aAAyB,CAEvB,MAAO,CAAC,GAAG,KAAK,QAAQ,CAC1B,CAMA,KAAKC,EAAsB,CACzB,KAAK,MAAM,EACX,KAAK,WAAW,CACd,KAAM,SACN,QAASA,CACX,CAAQ,CACV,CAKA,OAAQ,CACN,KAAK,SAAS,OAAS,CACzB,CAMQ,eAAgB,CAGtB,GAAI,KAAK,SAAS,OAAS,GAAc,CAEvC,IAAMC,EAAgB,KAAK,SAAS,KAAKC,GAAKA,EAAE,OAAS,QAAQ,EAC3DC,EAAiB,KAAK,SAAS,MAAM,GAAa,EAExD,KAAK,SAAWF,EACZ,CAACA,EAAe,GAAGE,EAAe,OAAOD,GAAKA,EAAE,OAAS,QAAQ,CAAC,EAClEC,CACN,CACF,CAOA,UAAW,CAET,KAAK,IAAI,0CAA0C,CACrD,CACF,ECnFA,UAAYC,MAAW,aAOhB,IAAMC,EAAmB,CAC9B,KAAM,cACN,YAAa,yJACb,OAAQ,CACN,KAAM,MACN,KAAM,SACN,YAAa,2CACb,SAAU,EACZ,EACA,SAAU,MAAO,CAAE,IAAAC,CAAI,IAAuB,CAC5C,GAAI,CAEF,IAAMC,EAAW,MAAMC,EAAK,IAASF,EAAK,CACxC,QAAS,CACP,OAAU,6EACV,aAAc,sHACd,kBAAmB,yBACrB,EACA,aAAc,cACd,QAAS,IACX,CAAC,EAEKG,EAAS,OAAO,KAAKF,EAAS,IAAI,EAClCG,EAAcH,EAAS,QAAQ,cAAc,GAAK,GAGpDI,EAAW,QAGTC,EAAeF,EAAY,MAAM,mBAAmB,EAC1D,GAAIE,EACFD,EAAWC,EAAa,CAAC,EAAE,YAAY,MAClC,CAGL,IAAMC,EADcJ,EAAO,MAAM,EAAG,IAAI,EAAE,SAAS,OAAO,EACrB,MAAM,uCAAuC,EAC9EI,IACFF,EAAWE,EAAiB,CAAC,EAAE,YAAY,EAE/C,CAGA,OAAIF,IAAa,UAAYA,IAAa,MAC3B,SAAOF,EAAQE,CAAQ,EAIzB,SAAOF,EAAQ,OAAO,CACrC,OAASK,EAAY,CACnB,cAAQ,MAAM,sDAAwBR,CAAG,GAAIQ,EAAM,OAAO,EACpD,IAAI,MAAM,qDAAaA,EAAM,OAAO,EAAE,CAC9C,CACF,CACF,EC7DA,OAAS,gBAAAC,EAAa,qBAAAC,MAAyB,gBAE/C,IAAMC,EAAyB,CAC3B,OAAQF,EAAa,aAAa,OAClC,YAAaA,EAAa,aAAa,YACvC,KAAMA,EAAa,aAAa,KAChC,SAAUA,CACd,EACMG,EAA8B,CAChC,OAAQF,EAAkB,aAAa,OACvC,YAAaA,EAAkB,aAAa,YAC5C,KAAMA,EAAkB,aAAa,KACrC,SAAUA,CACd,ECTO,IAAMG,EAAaC,GAA6B,CAErDA,EAAY,aAAaC,CAAU,EAGnCD,EAAY,aAAaE,CAAgB,EACzCF,EAAY,aAAaG,CAAqB,CAGhD,EAEaC,EAAN,KAAkB,CAIvB,YAAYC,EAAuB,MAAO,CAF1C,KAAQ,KAAuB,MAG7B,KAAK,QAAU,IAAI,IACnB,KAAK,KAAOA,CACd,CAGQ,OAAOC,EAAa,CACtB,KAAK,OAAS,OAChB,QAAQ,IAAI,GAAGA,CAAI,CAEvB,CAGQ,SAASA,EAAa,CACxB,KAAK,OAAS,OAChB,QAAQ,MAAM,GAAGA,CAAI,CAEzB,CAEA,aAAaC,EAAY,CACvB,KAAK,QAAQ,IAAIA,EAAK,KAAMA,CAAI,CAClC,CACA,eAAeC,EAAc,CAC3B,KAAK,QAAQ,OAAOA,CAAI,CAC1B,CACA,QAAQA,EAAgC,CACtC,OAAO,KAAK,QAAQ,IAAIA,CAAI,CAC9B,CACA,UAAmB,CACjB,MAAO,CAAC,GAAG,KAAK,QAAQ,OAAO,CAAC,CAClC,CAEA,MAAM,SAASA,EAAcC,EAA8B,CAAC,EAAoB,CAE9E,IAAMF,EAAO,KAAK,QAAQC,CAAI,EAC9B,GAAI,CAACD,EACH,OAAO,QAAQ,OAAO,gBAAMC,CAAI,qBAAM,EAGxC,GAAI,CAACD,EAAK,SACR,OAAO,QAAQ,OAAO,gBAAMC,CAAI,mDAAW,EAG7C,GAAI,CAEF,IAAME,EAAS,MAAMH,EAAK,SAASE,GAAU,CAAC,CAAC,EAE/C,OAAO,OAAOC,GAAW,SAAWA,EAAS,KAAK,UAAUA,EAAQ,KAAM,CAAC,CAC7E,OAASC,EAAU,CACjB,WAAK,MAAM,mEAA2BH,CAAI,KAAMG,CAAG,EAC7CA,CACR,CACF,CACF,ECtEA,IAAMC,EAAsB,CAACC,EAAa,CAAC,IAOlB;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,UAqBf,KAAK,UAAUA,CAAK,CAAC;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAmCxBC,EAAQF,EClEf,UAAYG,MAAgB,OAOrB,IAAMC,EAAN,KAAkB,CAGvB,aAAc,CAEZ,IAAMC,EAA2B,WAAYF,EAC7C,KAAK,QAAUE,EAAK,CACtB,CAOO,GAAYC,EAAkBC,EAAmC,CACtE,KAAK,QAAQ,GAAGD,EAAOC,CAAO,CAChC,CAOO,KAAcD,EAAkBC,EAAmC,CACxE,IAAMC,EAAeC,GAAY,CAC/BF,EAAQE,CAAI,EACZ,KAAK,QAAQ,IAAIH,EAAOE,CAAW,CACrC,EACA,KAAK,QAAQ,GAAGF,EAAOE,CAAW,CACpC,CAOO,IAAaF,EAAkBC,EAAmC,CACvE,KAAK,QAAQ,IAAID,EAAOC,CAAO,CACjC,CAOO,KAAcD,EAAkBG,EAAgB,CACrD,KAAK,QAAQ,KAAKH,EAAOG,CAAI,CAC/B,CAKO,OAAc,CACnB,KAAK,QAAQ,IAAI,MAAM,CACzB,CAKO,eAAwB,CAC7B,OAAO,KAAK,QAAQ,IAAI,IAC1B,CACF,EAGMC,EAAc,IAAIN,EACjBO,EAAQD,EC7DR,IAAME,EAAN,KAAY,CA6EjB,YAAYC,EAOT,CAhFH,KAAQ,WAAqB,GAC7B,KAAQ,QAAkBC,EAC1B,KAAQ,OAAiBC,EACzB,KAAQ,KAAuB,MAU/B,KAAQ,eAAiB,MAAOC,GAA8C,CAE5E,GAAM,CAAE,SAAU,CAAE,KAAAC,EAAM,OAAAC,CAAO,CAAE,EAAIF,EAGvCG,EAAY,KAAK,kBAAmB,CAAE,YAAa,KAAK,UAAUH,EAAe,QAAQ,CAAE,CAAC,EAE5F,KAAK,IAAI,iDAAmBC,CAAI,kBAASC,CAAM,EAE/C,GAAI,CAEF,IAAME,EAAa,MAAM,KAAK,YAAY,SAASH,EAAMC,CAAM,EAE/D,OAAAC,EAAY,KAAK,qBAAsB,CAAE,YAAaC,CAAW,CAAC,EAGlE,KAAK,eAAe,WAAW,CAC7B,KAAM,OACN,QAAS;AAAA,gBAA+BH,CAAI;AAAA;AAAA,EAAiB,OAAOG,GAAe,SAAWA,EAAa,KAAK,UAAUA,EAAY,KAAM,CAAC,CAAC;AAAA;AAAA,qIAChJ,CAAQ,EAED,OAAOA,GAAe,SAAWA,EAAa,KAAK,UAAUA,CAAU,CAChF,OAASC,EAAY,CAEnBF,EAAY,KAAK,kBAAmB,CAAE,MAAAE,CAAM,CAAC,EAC7C,IAAMC,EAAe,yCAAWD,EAAM,SAAWA,CAAK,GAEtD,YAAK,eAAe,WAAW,CAC7B,KAAM,OACN,QAASC,CACX,CAAQ,EAEDA,CACT,CACF,EAGA,KAAQ,kBAAqBN,GAAwC,CAEnE,GAAM,CAAE,QAAAO,CAAQ,EAAIP,EAEpB,OAAAG,EAAY,KAAK,gBAAiB,CAAE,YAAaI,CAAQ,CAAC,EAEnDA,CACT,EAGA,KAAQ,eAAiC,CACvC,OAAU,MAAOC,GACR,MAAM,KAAK,eAAeA,CAAoB,EAEvD,aAAiBA,GACR,KAAK,kBAAkBA,CAAuB,CAEzD,EAcE,KAAK,KAAOX,GAAS,MAAQ,MAC7B,KAAK,WAAaA,GAAS,YAAc,GACzC,KAAK,UAAY,IAAIY,EAAU,CAC7B,QAASZ,GAAS,SAAW,KAAK,QAClC,OAAQA,GAAS,QAAU,KAAK,OAChC,KAAM,KAAK,KACX,MAAOA,GAAS,KAClB,CAAC,EACD,KAAK,eAAiB,IAAIa,EAAeb,GAAS,UAAW,KAAK,IAAI,EACtE,KAAK,YAAc,IAAIc,EAAY,KAAK,IAAI,EAC5CC,EAAU,KAAK,WAAW,EAG1B,IAAMC,EAAeC,EAAoB,KAAK,YAAY,SAAS,CAAC,EACpE,KAAK,eAAe,KAAKD,CAAY,CACvC,CA1FQ,OAAOE,EAAa,CACtB,KAAK,OAAS,OAChB,QAAQ,IAAI,GAAGA,CAAI,CAEvB,CA8FA,MAAM,KAAKC,EAAmBC,EAAkB,GAAwB,CAEtE,KAAK,eAAe,WAAW,CAC7B,KAAM,OACN,QAASD,CACX,CAAQ,EAGRb,EAAY,KAAK,aAAc,CAAE,UAAAa,CAAU,CAAC,EAC5C,KAAK,IAAI,6DAAqBA,CAAS,EAAE,EAGzC,IAAIE,EAAY,EAChB,KAAOA,IAAc,KAAK,YAAY,CAEpCf,EAAY,KAAK,kBAAmB,CAAE,UAAAe,CAAU,CAAC,EACjD,KAAK,IAAI,8BAAeA,CAAS,qBAAM,EAGvC,IAAMC,EAAW,KAAK,eAAe,YAAY,EAC7CX,EAAW,GAEf,GAAI,CACES,EAEF,MAAM,KAAK,UAAU,WAAWE,EAAWC,GAAU,CACnDZ,GAAYY,EAEZjB,EAAY,KAAK,oBAAqB,CAAE,MAAAiB,EAAO,oBAAqBZ,CAAS,CAAC,CAChF,CAAC,EAGDA,EAAW,MAAM,KAAK,UAAU,KAAKW,CAAQ,CAEjD,OAASE,EAAU,CACjB,IAAMC,EAAW,yDAAiBD,EAAI,OAAO,GAC7C,OAAAlB,EAAY,KAAK,aAAc,CAAE,QAASmB,CAAS,CAAC,EAC7CA,CACT,CAGAnB,EAAY,KAAK,yBAA0B,CAAE,SAAAK,CAAS,CAAC,EAGvD,KAAK,eAAe,WAAW,CAC7B,KAAM,YACN,QAASA,CACX,CAAQ,EAGR,IAAIR,EAAgD,KAEpD,GAAI,CAEFA,EAAiB,KAAK,MAAMQ,CAAQ,CACtC,MAAY,CAEV,IAAMe,EAAW,2MACjBpB,EAAY,KAAK,qBAAsB,CAAE,QAASoB,CAAS,CAAC,EAE5D,KAAK,eAAe,WAAW,CAC7B,KAAM,OACN,QAASA,CACX,CAAQ,EACR,QACF,CAGA,GAAI,CAACvB,EAAgB,CAEnB,KAAK,eAAe,WAAW,CAC7B,KAAM,OACN,QAHe,wGAIjB,CAAQ,EACR,QACF,CAEA,GAAM,CAAE,KAAAwB,CAAK,EAAIxB,EACXyB,EAAW,KAAK,eAAeD,CAAI,EAEzC,GAAI,CAACC,EAAU,CACb,IAAMF,EAAW,+CAAYC,CAAI,wFACjC,KAAK,eAAe,WAAW,CAC7B,KAAM,OACN,QAASD,CACX,CAAQ,EACR,QACF,CAGA,IAAMG,EAAS,MAAMD,EAASzB,CAAc,EAG5C,GAAIwB,IAAS,eACX,OAAOE,CAIX,CAEA,IAAMC,EAAa,qHACnB,OAAAxB,EAAY,KAAK,wBAAyB,CAAE,QAASwB,CAAW,CAAC,EAC1DA,CACT,CAKA,cAAe,CACb,IAAMd,EAAeC,EAAoB,KAAK,YAAY,SAAS,CAAC,EACpE,KAAK,eAAe,KAAKD,CAAY,CACvC,CACF","names":["axios","HttpClient","config","error","response","url","stream","requestConfig","request","res","data","defaultHttp","http_default","BASE_URL","ENDPOINT","API_KEY","MODEL","LLMClient","options","MODEL","HttpClient","BASE_URL","API_KEY","args","messages","ENDPOINT","err","onToken","response","buffer","chunk","lines","line","trimmedLine","data","token","e","resolve","reject","ContextManager","maxTokens","mode","args","message","systemPrompt","systemMessage","m","recentMessages","iconv","WebFetcher","url","response","http_default","buffer","contentType","encoding","charsetMatch","metaCharsetMatch","error","jiaowuNotice","jiaowuCompetition","jiaowuNoticeTool","jiaowuCompetitionTool","initTools","toolManager","WebFetcher","jiaowuNoticeTool","jiaowuCompetitionTool","ToolManager","mode","args","tool","name","params","result","err","composeSystemPrompt","tools","systemPrompt_default","mittModule","EventSystem","mitt","event","handler","onceHandler","data","eventSystem","EventSystem_default","Agent","options","BASE_URL","API_KEY","parsedResponse","name","params","EventSystem_default","toolResult","error","errorMessage","content","response","LLMClient","ContextManager","ToolManager","initTools","systemPrompt","systemPrompt_default","args","userInput","stream","loopCount","messages","token","err","errorMsg","feedback","type","strategy","result","timeoutMsg"]}

package/package.json

@@ -0,0 +1,53 @@
+{
+  "name": "nongyu-agent-sdk",
+  "version": "1.0.0",
+  "description": "农屿系统内部专用的Agent SDK",
+  "type": "module",
+  "main": "dist/index.mjs",
+  "module": "dist/index.mjs",
+  "types": "dist/index.d.ts",
+  "exports": {
+    ".": {
+      "import": "./dist/index.mjs",
+      "types": "./dist/index.d.ts"
+    }
+  },
+  "files": [
+    "dist",
+    "README.md",
+    "LICENSE"
+  ],
+  "scripts": {
+    "build": "tsup",
+    "dev": "tsup --watch",
+    "format": "prettier --write \"src/**/*.ts\"",
+    "lint": "eslint \"src/**/*.ts\"",
+    "test": "tsx test/test.ts"
+  },
+  "keywords": [
+    "agent",
+    "sdk",
+    "nongyu"
+  ],
+  "author": "TangLei",
+  "license": "ISC",
+  "dependencies": {
+    "axios": "^1.6.0",
+    "iconv-lite": "^0.6.3",
+    "mitt": "^3.0.1",
+    "nongyu-jiaowu": "^1.0.0"
+  },
+  "devDependencies": {
+    "@types/node": "^20.0.0",
+    "@typescript-eslint/eslint-plugin": "^6.9.0",
+    "@typescript-eslint/parser": "^6.9.0",
+    "eslint": "^8.52.0",
+    "prettier": "^3.1.0",
+    "rimraf": "^5.0.0",
+    "terser": "^5.24.0",
+    "ts-node": "^10.9.1",
+    "tsup": "^8.0.0",
+    "tsx": "^4.0.0",
+    "typescript": "^5.0.0"
+  }
+}