@humanclaw/humanclaw 1.1.5 → 1.2.0

package/README.md

@@ -4,7 +4,6 @@
 
 **碳基节点编排框架 —— 将人类抽象为分布式 Worker 节点**
 
-[![CI](https://github.com/humanclaw/humanclaw/actions/workflows/ci.yml/badge.svg)](https://github.com/humanclaw/humanclaw/actions/workflows/ci.yml)
 [![npm](https://img.shields.io/npm/v/@humanclaw/humanclaw)](https://www.npmjs.com/package/@humanclaw/humanclaw)
 [![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](https://opensource.org/licenses/MIT)
 
@@ -18,7 +17,9 @@
 
 ## 概述
 
-HumanClaw 是一个碳基节点编排框架。系统将真实人类抽象为 Agent（碳基节点），将现实中的任务派发与结果收集抽象为进程的**挂起（Suspend）**与**恢复（Resume）**，最终将碳基节点产生的结构化结果组装并提交给上游数字系统（如 OpenClaw），完成赛博世界与碳基世界的闭环。
+HumanClaw 是一个碳基节点编排框架。系统将真实人类抽象为 Agent（碳基节点），将现实中的任务派发与结果收集抽象为进程的**挂起（Suspend）**与**恢复（Resume）**。
+
+核心流程：输入自然语言需求 → 选人 → AI 自动规划（拆任务 + 生成话术 + 设 DDL）→ 确认分发 → 收交付物 → AI 聚合审查。
 
 ## 核心架构
 
@@ -30,17 +31,17 @@ HumanClaw 是一个碳基节点编排框架。系统将真实人类抽象为 Age
 │             │   Resume + Result │              │
 └─────┬───────┘                   └──────────────┘
       │
-      │  Aggregate
+      │  AI Review
       ▼
 ┌─────────────┐
-│  OpenClaw   │
-│  (数字生命)  │
+│  LLM 审查   │
+│ (Claude/GPT)│
 └─────────────┘
 ```
 
-- **Master 节点**：解析需求，拆解为无依赖的扁平 TODO 列表，分发给碳基节点
+- **Master 节点**：输入需求，AI 自动拆解为独立子任务，分发给碳基节点
 - **Worker 节点 (HumanAgent)**：接收带 `trace_id` 的独立任务，在碳基世界异步执行
-- **状态机与存储**：SQLite 本地持久化，保存上下文快照，释放系统内存
+- **AI 审查**：所有任务完成后，LLM 自动审查交付质量并生成报告
 
 ## 快速开始
 
@@ -65,16 +66,17 @@ humanclaw agent add
 # 交互式录入：节点名称、技能标签
 ```
 
-### 查看算力池
+### AI 规划任务
 
 ```bash
-humanclaw agent list
+humanclaw plan "完成首页重构，包括导航栏和页脚的响应式改版"
+# AI 自动拆解任务、匹配碳基节点、生成话术和 DDL
 ```
 
-### 查看任务状态
+### 查看算力池
 
 ```bash
-humanclaw status
+humanclaw agent list
 ```
 
 ## API 接口
@@ -84,28 +86,21 @@ humanclaw status
 | `GET` | `/api/v1/nodes/status` | 碳基算力池状态 |
 | `POST` | `/api/v1/nodes` | 注册碳基节点 |
 | `PATCH` | `/api/v1/nodes/:id/status` | 更新节点状态 |
+| `POST` | `/api/v1/jobs/plan` | AI 智能规划（不分发） |
 | `POST` | `/api/v1/jobs/create` | 创建并分发任务 |
 | `GET` | `/api/v1/jobs/active` | 获取看板数据 |
 | `POST` | `/api/v1/tasks/resume` | 提交交付物，触发恢复 |
 | `POST` | `/api/v1/tasks/reject` | 打回重做 |
-| `POST` | `/api/v1/jobs/:id/sync` | 聚合结果同步至 OpenClaw |
+| `POST` | `/api/v1/jobs/:id/review` | AI 聚合审查交付质量 |
+| `GET` | `/api/v1/config` | 获取 LLM 配置 |
+| `PUT` | `/api/v1/config` | 更新 LLM 配置 |
 
-### 创建任务示例
+### AI 规划示例
 
 ```bash
-curl -X POST http://localhost:2026/api/v1/jobs/create \
+curl -X POST http://localhost:2026/api/v1/jobs/plan \
   -H "Content-Type: application/json" \
-  -d '{
-    "original_prompt": "完成首页重构",
-    "openclaw_callback": "",
-    "tasks": [
-      {
-        "assignee_id": "emp_xxxxxxxx",
-        "todo_description": "实现响应式导航栏",
-        "deadline": "2026-03-28T18:00:00Z"
-      }
-    ]
-  }'
+  -d '{ "prompt": "完成首页重构" }'
 ```
 
 ### 提交交付物
@@ -115,7 +110,7 @@ curl -X POST http://localhost:2026/api/v1/tasks/resume \
   -H "Content-Type: application/json" \
   -d '{
     "trace_id": "TK-9527",
-    "result_data": { "text": "导航栏已完成，支持移动端适配" }
+    "result_data": { "text": "https://github.com/org/repo/pull/42" }
   }'
 ```
 
@@ -123,17 +118,34 @@ curl -X POST http://localhost:2026/api/v1/tasks/resume \
 
 Web 看板包含三个核心视图：
 
-- **碳基算力池监控** — 实时查看碳基节点状态（🟢空闲 🟡忙碌 🔴离线 🟣崩溃）
-- **碳基编排大盘** — 任务 Kanban（已分发 / 已超时 / 已交付）+ Job 进度条
+- **碳基算力池** — 实时查看碳基节点状态（🟢空闲 🟡忙碌 🔴离线 🟣崩溃），一键添加/删除节点
+- **碳基编排大盘** — AI 智能规划 + 任务 Kanban + 可交互任务卡片（点击直接提交交付/打回）+ AI 聚合审查
 - **I/O 交付终端** — 输入 trace_id 和交付载荷，触发系统恢复
 
+### AI 功能
+
+- **智能规划** — 输入需求，AI 自动拆任务、匹配碳基节点、生成布置话术、设 DDL（可调）
+- **聚合审查** — 全部交付后，AI 审查每个交付物质量（支持 GitHub PR/Commit/Issue URL），生成评分报告
+- **可配置 LLM** — 支持 Claude / OpenAI，可自定义 Base URL 接入私有模型服务（vLLM / Ollama / Azure）
+
 ## 核心工作流
 
-1. **镜像封装** — 录入物理成员信息，构建碳基算力池
-2. **拆分分发** — Master 拆解任务，绑定 Agent，生成 trace_id
-3. **挂起快照** — 上下文序列化持久化，主控进入休眠
-4. **异步恢复** — 碳基节点提交产出物，系统唤醒 Job
-5. **聚合闭环** — 所有子任务完成后，打包提交 OpenClaw
+1. **镜像封装** — 录入碳基成员信息，构建碳基算力池
+2. **AI 规划** — 输入需求，AI 拆解任务、匹配节点、生成话术和 DDL
+3. **确认分发** — 预览规划结果，调整 DDL，确认后一键分发
+4. **异步恢复** — 碳基节点提交交付物（支持 GitHub URL），系统唤醒 Job
+5. **AI 审查** — 所有子任务完成后，LLM 审查交付质量并生成报告
+
+## 环境变量
+
+| 变量 | 默认值 | 说明 |
+|------|--------|------|
+| `HUMANCLAW_LLM_PROVIDER` | `claude` | LLM 提供商：`claude` 或 `openai` |
+| `HUMANCLAW_LLM_API_KEY` | - | LLM API Key（使用 AI 功能时必填） |
+| `HUMANCLAW_LLM_MODEL` | 按 provider | 可选覆盖模型名 |
+| `HUMANCLAW_LLM_BASE_URL` | 官方地址 | 自定义 API 地址（私有部署） |
+
+> Dashboard 设置面板也可以配置以上参数，优先级高于环境变量。
 
 ## 数据模型
 
@@ -154,12 +166,6 @@ interface HumanTask {
   status: TaskStatus;     // PENDING | DISPATCHED | RESOLVED | OVERDUE
   result_data: unknown;
 }
-
-interface OrchestrationJob {
-  job_id: string;
-  original_prompt: string;
-  openclaw_callback: string;
-}
 ```
 
 ## 开发
@@ -178,9 +184,10 @@ npm run lint       # 类型检查
 - **Runtime**: Node.js 22+, TypeScript (ESM, strict)
 - **API**: Express v5
 - **Storage**: SQLite (better-sqlite3, WAL mode)
+- **LLM**: Claude / OpenAI（原生 fetch，零依赖）
 - **CLI**: Commander.js + @clack/prompts
-- **Dashboard**: Vite + Vanilla TS
-- **Testing**: Vitest (32 tests)
+- **Dashboard**: 内联 HTML（无需构建）
+- **Testing**: Vitest (40 tests)
 
 ## License
 

package/README_EN.md

@@ -4,7 +4,6 @@
 
 **Carbon-Based Node Orchestration Framework — Humans as Distributed Worker Nodes**
 
-[![CI](https://github.com/humanclaw/humanclaw/actions/workflows/ci.yml/badge.svg)](https://github.com/humanclaw/humanclaw/actions/workflows/ci.yml)
 [![npm](https://img.shields.io/npm/v/@humanclaw/humanclaw)](https://www.npmjs.com/package/@humanclaw/humanclaw)
 [![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](https://opensource.org/licenses/MIT)
 
@@ -18,7 +17,9 @@
 
 ## Overview
 
-HumanClaw is a carbon-based node orchestration framework. The system abstracts real humans as Agents (carbon-based nodes), models task dispatch and result collection as process **Suspend** and **Resume**, and ultimately assembles structured results from carbon-based nodes to submit to upstream digital systems (e.g., OpenClaw), closing the loop between the cyber world and the carbon-based world.
+HumanClaw is a carbon-based node orchestration framework. The system abstracts real humans as Agents (carbon-based nodes), models task dispatch and result collection as process **Suspend** and **Resume**.
+
+Core flow: natural language input → select people → AI auto-plans (breaks down tasks + generates briefings + sets deadlines) → confirm dispatch → collect deliverables → AI aggregated review.
 
 ## Core Architecture
 
@@ -30,17 +31,17 @@ HumanClaw is a carbon-based node orchestration framework. The system abstracts r
 │             │   Resume + Result │              │
 └─────┬───────┘                   └──────────────┘
       │
-      │  Aggregate
+      │  AI Review
       ▼
 ┌─────────────┐
-│  OpenClaw   │
-│ (Digital AI)│
+│  LLM Review │
+│ (Claude/GPT)│
 └─────────────┘
 ```
 
-- **Master Node**: Parses requirements, breaks them into flat, independent TODOs, and dispatches to carbon-based nodes
+- **Master Node**: Input requirements, AI auto-breaks them into independent sub-tasks, dispatches to carbon-based nodes
 - **Worker Node (HumanAgent)**: Receives independent tasks with a `trace_id`, executes asynchronously in the carbon-based world
-- **State Machine & Storage**: SQLite local persistence, saves context snapshots, frees system memory
+- **AI Review**: After all tasks complete, LLM reviews deliverable quality and generates a report
 
 ## Quick Start
 
@@ -65,16 +66,17 @@ humanclaw agent add
 # Interactive prompts for: node name, capability tags
 ```
 
-### View Compute Pool
+### AI Task Planning
 
 ```bash
-humanclaw agent list
+humanclaw plan "Rebuild the homepage with responsive navbar and footer"
+# AI auto-breaks tasks, matches nodes, generates briefings and deadlines
 ```
 
-### Check Task Status
+### View Compute Pool
 
 ```bash
-humanclaw status
+humanclaw agent list
 ```
 
 ## API Endpoints
@@ -84,28 +86,21 @@ humanclaw status
 | `GET` | `/api/v1/nodes/status` | Carbon compute pool status |
 | `POST` | `/api/v1/nodes` | Register carbon-based node |
 | `PATCH` | `/api/v1/nodes/:id/status` | Update node status |
+| `POST` | `/api/v1/jobs/plan` | AI task planning (does not dispatch) |
 | `POST` | `/api/v1/jobs/create` | Create and dispatch job |
 | `GET` | `/api/v1/jobs/active` | Get kanban data |
 | `POST` | `/api/v1/tasks/resume` | Submit deliverable, trigger resume |
 | `POST` | `/api/v1/tasks/reject` | Reject and retry |
-| `POST` | `/api/v1/jobs/:id/sync` | Aggregate results and sync to OpenClaw |
+| `POST` | `/api/v1/jobs/:id/review` | AI aggregated review of deliverables |
+| `GET` | `/api/v1/config` | Get LLM configuration |
+| `PUT` | `/api/v1/config` | Update LLM configuration |
 
-### Create Job Example
+### AI Planning Example
 
 ```bash
-curl -X POST http://localhost:2026/api/v1/jobs/create \
+curl -X POST http://localhost:2026/api/v1/jobs/plan \
   -H "Content-Type: application/json" \
-  -d '{
-    "original_prompt": "Rebuild the homepage",
-    "openclaw_callback": "",
-    "tasks": [
-      {
-        "assignee_id": "emp_xxxxxxxx",
-        "todo_description": "Implement responsive navbar",
-        "deadline": "2026-03-28T18:00:00Z"
-      }
-    ]
-  }'
+  -d '{ "prompt": "Rebuild the homepage" }'
 ```
 
 ### Submit Deliverable
@@ -115,7 +110,7 @@ curl -X POST http://localhost:2026/api/v1/tasks/resume \
   -H "Content-Type: application/json" \
   -d '{
     "trace_id": "TK-9527",
-    "result_data": { "text": "Navbar complete, mobile-responsive" }
+    "result_data": { "text": "https://github.com/org/repo/pull/42" }
   }'
 ```
 
@@ -123,17 +118,34 @@ curl -X POST http://localhost:2026/api/v1/tasks/resume \
 
 The web dashboard includes three core views:
 
-- **Carbon Compute Pool Monitor** — Real-time carbon-based node status (🟢Idle 🟡Busy 🔴Offline 🟣OOM)
-- **Carbon Orchestration Pipeline** — Task Kanban (Dispatched / Overdue / Resolved) + Job progress bars
+- **Carbon Compute Pool** — Real-time carbon-based node status (🟢Idle 🟡Busy 🔴Offline 🟣OOM), add/remove nodes
+- **Carbon Orchestration Pipeline** — AI planning + Task Kanban + interactive task cards (click to submit/reject) + AI review
 - **I/O Resolution Terminal** — Input trace_id and payload to trigger system resume
 
+### AI Features
+
+- **Smart Planning** — Input requirements, AI auto-breaks tasks, matches nodes, generates briefings, sets adjustable deadlines
+- **Aggregated Review** — After all deliveries, AI reviews each deliverable (supports GitHub PR/Commit/Issue URLs), generates quality report
+- **Configurable LLM** — Supports Claude / OpenAI, custom Base URL for private deployments (vLLM / Ollama / Azure)
+
 ## Core Workflow
 
 1. **Agent Encapsulation** — Register human members, build the carbon compute pool
-2. **Dispatch** — Master breaks down tasks, binds Agents, generates trace_ids
-3. **Suspend & Snapshot** — Serialize context to persistence, main process hibernates
-4. **Async Resume** — Carbon-based nodes submit deliverables, system wakes up the Job
-5. **Aggregation** — When all sub-tasks complete, package and submit to OpenClaw
+2. **AI Planning** — Input requirements, AI breaks tasks, matches nodes, generates briefings and deadlines
+3. **Confirm Dispatch** — Preview plan, adjust deadlines, one-click dispatch
+4. **Async Resume** — Carbon-based nodes submit deliverables (supports GitHub URLs), system wakes up the Job
+5. **AI Review** — When all sub-tasks complete, LLM reviews deliverable quality and generates a report
+
+## Environment Variables
+
+| Variable | Default | Description |
+|----------|---------|-------------|
+| `HUMANCLAW_LLM_PROVIDER` | `claude` | LLM provider: `claude` or `openai` |
+| `HUMANCLAW_LLM_API_KEY` | - | LLM API Key (required for AI features) |
+| `HUMANCLAW_LLM_MODEL` | per provider | Optional model override |
+| `HUMANCLAW_LLM_BASE_URL` | official | Custom API URL (private deployments) |
+
+> Dashboard settings panel can also configure these, with higher priority than env vars.
 
 ## Data Models
 
@@ -154,12 +166,6 @@ interface HumanTask {
   status: TaskStatus;     // PENDING | DISPATCHED | RESOLVED | OVERDUE
   result_data: unknown;
 }
-
-interface OrchestrationJob {
-  job_id: string;
-  original_prompt: string;
-  openclaw_callback: string;
-}
 ```
 
 ## Development
@@ -178,9 +184,10 @@ npm run lint       # Type check
 - **Runtime**: Node.js 22+, TypeScript (ESM, strict)
 - **API**: Express v5
 - **Storage**: SQLite (better-sqlite3, WAL mode)
+- **LLM**: Claude / OpenAI (native fetch, zero dependencies)
 - **CLI**: Commander.js + @clack/prompts
-- **Dashboard**: Vite + Vanilla TS
-- **Testing**: Vitest (32 tests)
+- **Dashboard**: Inline HTML (no build step)
+- **Testing**: Vitest (40 tests)
 
 ## License
 

package/dist/index.js

@@ -35,12 +35,13 @@ function getDb(dbPath) {
 function initSchema(db2) {
   db2.exec(`
     CREATE TABLE IF NOT EXISTS agents (
-      agent_id   TEXT PRIMARY KEY,
-      name       TEXT NOT NULL,
+      agent_id     TEXT PRIMARY KEY,
+      name         TEXT NOT NULL,
       capabilities TEXT NOT NULL DEFAULT '[]',
-      status     TEXT NOT NULL DEFAULT 'IDLE'
-                 CHECK (status IN ('IDLE', 'BUSY', 'OFFLINE', 'OOM')),
-      created_at TEXT NOT NULL DEFAULT (datetime('now'))
+      relationship TEXT NOT NULL DEFAULT '',
+      status       TEXT NOT NULL DEFAULT 'IDLE'
+                   CHECK (status IN ('IDLE', 'BUSY', 'OFFLINE', 'OOM')),
+      created_at   TEXT NOT NULL DEFAULT (datetime('now'))
     );
 
     CREATE TABLE IF NOT EXISTS jobs (
@@ -73,6 +74,10 @@ function initSchema(db2) {
       value TEXT NOT NULL
     );
   `);
+  const cols = db2.prepare("PRAGMA table_info(agents)").all();
+  if (!cols.some((c) => c.name === "relationship")) {
+    db2.exec(`ALTER TABLE agents ADD COLUMN relationship TEXT NOT NULL DEFAULT ''`);
+  }
 }
 
 // src/routes/nodes.ts
@@ -89,12 +94,13 @@ function createAgent(agent, db2) {
   const conn = db2 ?? getDb();
   const now = (/* @__PURE__ */ new Date()).toISOString();
   conn.prepare(
-    `INSERT INTO agents (agent_id, name, capabilities, status, created_at)
-       VALUES (?, ?, ?, ?, ?)`
+    `INSERT INTO agents (agent_id, name, capabilities, relationship, status, created_at)
+       VALUES (?, ?, ?, ?, ?, ?)`
   ).run(
     agent.agent_id,
     agent.name,
     JSON.stringify(agent.capabilities),
+    agent.relationship || "",
     agent.status,
     now
   );
@@ -167,7 +173,7 @@ router.get("/status", (_req, res) => {
   });
 });
 router.post("/", (req, res) => {
-  const { name, capabilities, status } = req.body;
+  const { name, capabilities, relationship, status } = req.body;
   if (!name || !Array.isArray(capabilities)) {
     res.status(400).json({ error: "name and capabilities[] are required" });
     return;
@@ -176,6 +182,7 @@ router.post("/", (req, res) => {
     agent_id: generateId("emp"),
     name,
     capabilities,
+    relationship: relationship || "",
     status: status ?? "IDLE"
   });
   res.status(201).json(agent);
@@ -536,7 +543,8 @@ function buildUserPrompt(prompt, agents) {
   const agentList = agents.map((a) => {
     const load = a.active_task_count > 0 ? `\u5F53\u524D\u6709 ${a.active_task_count} \u4E2A\u8FDB\u884C\u4E2D\u7684\u4EFB\u52A1` : "\u5F53\u524D\u7A7A\u95F2";
     const speed = a.avg_delivery_hours !== null ? `\u5E73\u5747\u4EA4\u4ED8\u65F6\u95F4 ${a.avg_delivery_hours}h` : "\u6682\u65E0\u5386\u53F2\u6570\u636E";
-    return `- ${a.name} (ID: ${a.agent_id})  \u6280\u80FD: [${a.capabilities.join(", ")}]  ${load}  ${speed}`;
+    const rel = a.relationship ? `  \u5173\u7CFB: ${a.relationship}` : "";
+    return `- ${a.name} (ID: ${a.agent_id})  \u6280\u80FD: [${a.capabilities.join(", ")}]${rel}  ${load}  ${speed}`;
   }).join("\n");
   return `\u5F53\u524D\u65F6\u95F4: ${now.toISOString()}
 
@@ -738,6 +746,61 @@ function rejectTask(traceId, newDeadline, db2) {
   return getTask(traceId, conn);
 }
 
+// src/services/simulator.ts
+function buildSimulatePrompt(agentName, relationship, capabilities, taskDescription, deadline) {
+  return `\u4F60\u73B0\u5728\u626E\u6F14\u4E00\u4E2A\u540D\u53EB\u300C${agentName}\u300D\u7684\u4EBA\uFF0C\u4F60\u7684\u6280\u80FD\u6807\u7B7E\u662F [${capabilities.join(", ")}]\u3002
+${relationship ? `\u4F60\u548C\u5E03\u7F6E\u4EFB\u52A1\u7684\u4EBA\u7684\u5173\u7CFB\u662F\uFF1A${relationship}\u3002` : ""}
+
+\u4F60\u6536\u5230\u4E86\u4E00\u4E2A\u4EFB\u52A1\uFF1A
+${taskDescription}
+
+\u622A\u6B62\u65F6\u95F4\uFF1A${new Date(deadline).toLocaleString("zh-CN")}
+
+\u8BF7\u7AD9\u5728\u300C${agentName}\u300D\u7684\u89C6\u89D2\uFF0C\u7528\u8FD9\u4E2A\u4EBA\u7269\u81EA\u7136\u7684\u8BED\u6C14\u548C\u53E3\u543B\uFF0C\u5199\u4E00\u6BB5\u4EFB\u52A1\u4EA4\u4ED8\u6C47\u62A5\u3002
+\u8981\u6C42\uFF1A
+1. \u5185\u5BB9\u8981\u8D34\u5408\u4EFB\u52A1\u63CF\u8FF0\uFF0C\u4F53\u73B0\u4E13\u4E1A\u80FD\u529B
+2. \u8BED\u6C14\u7B26\u5408\u4EBA\u7269\u8EAB\u4EFD${relationship ? "\u548C\u4E0E\u4E0A\u7EA7\u7684\u5173\u7CFB" : ""}
+3. \u6C47\u62A5\u5185\u5BB9\u5177\u4F53\u3001\u6709\u7EC6\u8282\uFF0C\u5305\u542B\u505A\u4E86\u4EC0\u4E48\u3001\u9047\u5230\u4E86\u4EC0\u4E48\u95EE\u9898\u3001\u6700\u7EC8\u7ED3\u679C\u5982\u4F55
+4. \u5B57\u6570 200-400 \u5B57
+5. \u76F4\u63A5\u8F93\u51FA\u6C47\u62A5\u5185\u5BB9\uFF0C\u4E0D\u8981\u52A0\u4EFB\u4F55\u683C\u5F0F\u524D\u7F00\u6216\u8BF4\u660E`;
+}
+async function simulateDelivery(traceId, provider, db2) {
+  const conn = db2 ?? getDb();
+  const task = getTask(traceId, conn);
+  if (!task) {
+    throw new Error(`Task not found: ${traceId}`);
+  }
+  const agent = getAgent(task.assignee_id, conn);
+  if (!agent) {
+    throw new Error(`Agent not found: ${task.assignee_id}`);
+  }
+  const llm = provider ?? createLlmProvider();
+  const response = await llm.complete({
+    messages: [
+      {
+        role: "system",
+        content: "\u4F60\u662F\u4E00\u4E2A\u89D2\u8272\u626E\u6F14\u4E13\u5BB6\uFF0C\u64C5\u957F\u6A21\u62DF\u4E0D\u540C\u4EBA\u7269\u7684\u8BED\u6C14\u548C\u6C47\u62A5\u98CE\u683C\u3002"
+      },
+      {
+        role: "user",
+        content: buildSimulatePrompt(
+          agent.name,
+          agent.relationship,
+          agent.capabilities,
+          task.todo_description,
+          task.deadline
+        )
+      }
+    ],
+    temperature: 0.7,
+    max_tokens: 1024
+  });
+  return {
+    trace_id: traceId,
+    simulated_delivery: response.content
+  };
+}
+
 // src/routes/tasks.ts
 var router3 = Router3();
 router3.post("/resume", (req, res) => {
@@ -776,6 +839,21 @@ router3.post("/reject", (req, res) => {
     res.status(400).json({ error: message });
   }
 });
+router3.post("/simulate", async (req, res) => {
+  const { trace_id } = req.body;
+  if (!trace_id) {
+    res.status(400).json({ error: "trace_id is required" });
+    return;
+  }
+  try {
+    const result = await simulateDelivery(trace_id);
+    res.json(result);
+  } catch (error) {
+    const message = error instanceof Error ? error.message : "Unknown error";
+    const status = message.includes("API Key") || message.includes("API key") ? 503 : 400;
+    res.status(status).json({ error: message });
+  }
+});
 var tasks_default = router3;
 
 // src/routes/sync.ts
@@ -803,11 +881,13 @@ function buildReviewUserPrompt(originalPrompt, tasks) {
     const t = tasks[i];
     let resultText = "";
     if (t.result_data) {
-      try {
-        const rd = JSON.parse(t.result_data);
-        resultText = rd.text || JSON.stringify(rd, null, 2);
-      } catch {
+      if (typeof t.result_data === "string") {
         resultText = t.result_data;
+      } else if (typeof t.result_data === "object") {
+        const rd = t.result_data;
+        resultText = rd.text || JSON.stringify(rd, null, 2);
+      } else {
+        resultText = String(t.result_data);
       }
     }
     prompt += `--- \u5B50\u4EFB\u52A1 ${i + 1} ---
@@ -1131,6 +1211,7 @@ async function loadFleet(el){
     for(const a of d.agents){
       h+='<div class="card"><div class="agent-header"><span class="dot '+a.status+'"></span><span class="agent-name">'+esc(a.name)+'</span></div>';
       h+='<div class="agent-id">'+a.agent_id+'</div>';
+      if(a.relationship)h+='<div style="font-size:11px;color:var(--purple);margin-bottom:4px">&#128101; '+esc(a.relationship)+'</div>';
       h+='<div class="caps">';for(const c of a.capabilities)h+='<span class="cap">'+esc(c)+'</span>';h+='</div>';
       h+='<div class="agent-meta"><span>\u4EFB\u52A1: '+a.active_task_count+'</span>';
       if(a.avg_delivery_hours!==null)h+='<span>\u5E73\u5747\u4EA4\u4ED8: '+a.avg_delivery_hours+'h</span>';
@@ -1152,6 +1233,7 @@ window.showAddAgent=function(){
   ov.innerHTML='<div class="form-card"><h3>+ \u6DFB\u52A0\u78B3\u57FA\u8282\u70B9</h3>'
     +'<div class="fg"><label>\u8282\u70B9\u540D\u79F0</label><input id="aa-name" placeholder="\u4F8B: \u524D\u7AEF\u8001\u674E"/></div>'
     +'<div class="fg"><label>\u6280\u80FD\u6807\u7B7E</label><input id="aa-caps" placeholder="\u4F8B: UI/UX, \u524D\u7AEF\u5F00\u53D1, \u6297\u538B\u80FD\u529B\u5F3A"/><div class="hint">\u591A\u4E2A\u6807\u7B7E\u7528\u9017\u53F7\u5206\u9694</div></div>'
+    +'<div class="fg"><label>\u4E0E\u4F60\u7684\u5173\u7CFB <span style="color:var(--text-dim);font-weight:400">(\u53EF\u9009)</span></label><input id="aa-rel" placeholder="\u4F8B: \u76F4\u5C5E\u4E0B\u5C5E / \u5B9E\u4E60\u751F / \u5916\u5305\u540C\u4E8B / \u4E49\u5F1F"/><div class="hint">\u63CF\u8FF0\u6B64\u4EBA\u4E0E\u4F60\u7684\u5173\u7CFB\uFF0CAI \u89C4\u5212\u548C\u6A21\u62DF\u4EA4\u4ED8\u65F6\u4F1A\u53C2\u8003</div></div>'
     +'<div class="btn-group"><button class="btn btn-primary" onclick="submitAgent()">\u6CE8\u518C\u8282\u70B9</button><button class="btn btn-ghost" onclick="document.getElementById(\\'overlay\\').remove()">\u53D6\u6D88</button></div></div>';
   document.body.appendChild(ov);
   document.getElementById('aa-name').focus();
@@ -1159,10 +1241,11 @@ window.showAddAgent=function(){
 window.submitAgent=async function(){
   const name=document.getElementById('aa-name').value.trim();
   const caps=document.getElementById('aa-caps').value.trim();
+  const rel=document.getElementById('aa-rel').value.trim();
   if(!name){toast('\u8BF7\u8F93\u5165\u8282\u70B9\u540D\u79F0',false);return}
   if(!caps){toast('\u8BF7\u8F93\u5165\u81F3\u5C11\u4E00\u4E2A\u6280\u80FD\u6807\u7B7E',false);return}
   try{
-    const r=await fetch(API+'/nodes',{method:'POST',headers:{'Content-Type':'application/json'},body:JSON.stringify({name,capabilities:caps.split(',').map(s=>s.trim()).filter(Boolean)})});
+    const r=await fetch(API+'/nodes',{method:'POST',headers:{'Content-Type':'application/json'},body:JSON.stringify({name,capabilities:caps.split(',').map(s=>s.trim()).filter(Boolean),relationship:rel})});
     const d=await r.json();
     if(!r.ok){toast(d.error||'\u6CE8\u518C\u5931\u8D25',false);return}
     toast('\u8282\u70B9 '+d.agent_id+' \u6CE8\u518C\u6210\u529F\uFF01',true);
@@ -1439,13 +1522,14 @@ window.showTaskDetail=function(traceId){
   if(t.status==='RESOLVED'&&t.result_data){
     // Show result
     let resultText='';
-    try{const rd=JSON.parse(t.result_data);resultText=rd.text||JSON.stringify(rd,null,2)}catch{resultText=t.result_data}
+    if(typeof t.result_data==='object'&&t.result_data){resultText=t.result_data.text||JSON.stringify(t.result_data,null,2)}else{resultText=String(t.result_data||'')}
     h+='<div class="detail-row"><div class="detail-label">\u4EA4\u4ED8\u7ED3\u679C</div><div class="result-display">'+esc(resultText)+'</div></div>';
     h+='<div class="btn-group"><button class="btn btn-ghost" onclick="document.getElementById(\\'overlay\\').remove()">\u5173\u95ED</button></div>';
   }else{
     // Input for resume/reject
     h+='<div class="fg" style="margin-top:16px"><label>\u63D0\u4EA4 Human \u4EA4\u4ED8\u7269</label><textarea id="td-payload" rows="4" placeholder="\u7C98\u8D34\u4EA4\u4ED8\u7269\u5185\u5BB9\u3001GitHub PR/Commit URL\u3001\u5DE5\u4F5C\u6C47\u62A5\u7B49..."></textarea><div class="hint">\u652F\u6301\u8D34 GitHub URL\uFF08PR\u3001Commit\u3001Issue\uFF09\uFF0CAI \u5BA1\u67E5\u65F6\u4F1A\u5206\u6790</div></div>';
     h+='<div class="btn-group">';
+    h+='<button class="btn btn-primary btn-sm" onclick="simulateDelivery(\\''+t.trace_id+'\\')">&#129302; \u6A21\u62DF\u4EA4\u4ED8</button>';
     h+='<button class="btn btn-green" onclick="taskResume(\\''+t.trace_id+'\\')">\u63D0\u4EA4\u4EA4\u4ED8 (Resume)</button>';
     h+='<button class="btn btn-danger" onclick="taskReject(\\''+t.trace_id+'\\')">\u6253\u56DE\u91CD\u505A (Reject)</button>';
     h+='<button class="btn btn-ghost" onclick="document.getElementById(\\'overlay\\').remove()">\u53D6\u6D88</button>';
@@ -1481,6 +1565,21 @@ window.taskReject=async function(traceId){
   }catch{toast('\u7F51\u7EDC\u9519\u8BEF',false)}
 };
 
+window.simulateDelivery=async function(traceId){
+  const btn=event.target;
+  const origText=btn.innerHTML;
+  btn.disabled=true;btn.innerHTML='&#8987; AI \u751F\u6210\u4E2D...';
+  try{
+    const r=await fetch(API+'/tasks/simulate',{method:'POST',headers:{'Content-Type':'application/json'},body:JSON.stringify({trace_id:traceId})});
+    const d=await r.json();
+    if(!r.ok){toast(d.error||'\u6A21\u62DF\u5931\u8D25',false);btn.disabled=false;btn.innerHTML=origText;return}
+    const ta=document.getElementById('td-payload');
+    if(ta)ta.value=d.simulated_delivery;
+    toast('\u5DF2\u751F\u6210\u6A21\u62DF\u4EA4\u4ED8\u5185\u5BB9',true);
+    btn.disabled=false;btn.innerHTML=origText;
+  }catch(e){toast('\u7F51\u7EDC\u9519\u8BEF: '+e.message,false);btn.disabled=false;btn.innerHTML=origText}
+};
+
 // \u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550
 // TERMINAL
 // \u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550
@@ -1660,10 +1759,20 @@ agentCmd.command("add").description("Register a new carbon-based node").action(a
     process.exit(0);
   }
   const capabilities = capInput.split(",").map((s) => s.trim()).filter(Boolean);
+  const relInput = await p.text({
+    message: "Relationship with you (optional):",
+    placeholder: "e.g. direct report / intern / contractor",
+    defaultValue: ""
+  });
+  if (p.isCancel(relInput)) {
+    p.cancel("Cancelled.");
+    process.exit(0);
+  }
   const agent = createAgent({
     agent_id: generateId("emp"),
     name,
     capabilities,
+    relationship: relInput || "",
     status: "IDLE"
   });
   p.outro(