northseadl/norix-skills / llm-agent-dev
llm-agent-dev
UnverifiedLLM Agent engineering: pattern selection (12-mode matrix), data simulation, convergence iteration. Covers intent routing, function calling, ReAct, MCP, prompt chaining, guardrails, evaluation.
references/intent-patterns.md
View on GitHub# 意图模式与分类策略 (Intent Patterns & Classification)
> IntentSchema 设计、行业垂直意图模板、参数归一化与消歧策略
## IntentSchema 设计
LLM-Native 模式下,意图定义是**语义化注册表**,作为 System Prompt 传给 LLM:
```typescript
interface IntentSchema {
intents: IntentDefinition[];
}
interface IntentDefinition {
id: string; // "booking.query"
description: string; // 自然语言描述(给 LLM 理解用)
examples: string[]; // 2-5 个用户输入示例
params: ParamDefinition[];
priority: number; // 消歧优先级
}
interface ParamDefinition {
name: string; // "orderId"
type: 'string' | 'date' | 'number' | 'enum';
description: string; // "用户提到的订单号"
required: boolean;
enumValues?: string[];
}
```
### 设计原则
1. **description 决定质量**: LLM 靠 description 理解意图语义,务必精准
2. **examples 是正向锚点**: 提供典型用户表达,帮助 LLM 建立模式识别
3. **params 驱动结构化输出**: 明确告诉 LLM 应该提取什么参数
4. **数量控制**: 单次请求中意图 ≤ 30,超过则按领域分组路由
---
## 通用意图模板
几乎所有 Agent 系统都需要的基础意图:
```json
{
"intents": [
{ "id": "nav.home", "description": "用户想回到首页", "examples": ["回首页", "回到主页"], "params": [] },
{ "id": "nav.page", "description": "用户想导航到特定页面", "examples": ["去订单页", "打开设置"],
"params": [{ "name": "page", "type": "string", "description": "目标页面", "required": true }] },
{ "id": "help.faq", "description": "用户询问常见问题", "examples": ["怎么退款", "营业时间"], "params": [] },
{ "id": "help.contact", "description": "用户要联系人工客服", "examples": ["转人工", "找客服"], "params": [] },
{ "id": "user.profile", "description": "用户查看或修改个人信息", "examples": ["看看我的信息", "改手机号"],
"params": [{ "name": "field", "type": "enum", "description": "要修改的字段", "required": false,
"enumValues": ["name", "phone", "avatar", "address"] }] },
{ "id": "greeting", "description": "用户打招呼或闲聊", "examples": ["你好", "在吗"], "params": [] }
]
}
```
---
## 行业垂直意图
### 电商/零售
| 意图 ID | 描述 | 关键参数 |
|---------|------|---------|
| `product.search` | 搜索商品 | keyword, category, priceRange |
| `cart.view` | 查看购物车 | — |
| `order.query` | 查询订单 | orderId, status |
| `order.refund` | 申请退款 | orderId, reason |
### 预约/服务
| 意图 ID | 描述 | 关键参数 |
|---------|------|---------|
| `booking.create` | 创建预约 | date, timeSlot, serviceType |
| `booking.query` | 查询预约 | bookingId |
| `booking.cancel` | 取消预约 | bookingId, reason |
| `availability.check` | 查看可用时间 | dateHint |
### 内容/教育
| 意图 ID | 描述 | 关键参数 |
|---------|------|---------|
| `content.search` | 搜索内容 | keyword, contentType |
| `course.enroll` | 报名课程 | courseId |
| `progress.check` | 查看学习进度 | courseId |
---
## 参数归一化
LLM 提取的参数需要**后处理校验和归一化**:
### 日期归一化
```typescript
function normalizeDate(raw: string): string | null {
const relativeMap: Record<string, number> = {
'今天': 0, '明天': 1, '后天': 2, '大后天': 3,
};
for (const [word, offset] of Object.entries(relativeMap)) {
if (raw.includes(word)) {
const d = new Date(); d.setDate(d.getDate() + offset);
return d.toISOString().split('T')[0];
}
}
// "下周三" → 计算实际日期
const weekdayMatch = raw.match(/下周([一二三四五六日天])/);
if (weekdayMatch) { /* 计算下周几的日期 */ }
return null;
}
```
### 参数校验
```typescript
function validateParams(
params: Record<string, any>,
schema: ParamDefinition[],
): { valid: Record<string, any>; missing: string[] } {
const valid: Record<string, any> = {};
const missing: string[] = [];
for (const def of schema) {
const raw = params[def.name];
if (raw == null && def.required) { missing.push(def.name); continue; }
if (raw == null) continue;
switch (def.type) {
case 'date': valid[def.name] = normalizeDate(String(raw)); break;
case 'number': valid[def.name] = parseFloat(String(raw)) || null; break;
case 'enum': valid[def.name] = def.enumValues?.includes(raw) ? raw : null; break;
default: valid[def.name] = raw;
}
}
return { valid, missing };
}
```
---
## 规则短路模式
规则在 LLM-Native 体系中是**高频精确匹配的缓存加速层**,非主力分类器。
### 适用条件
1. **极高频**: 占总请求量 > 10%
2. **无歧义**: 输入模式唯一映射到该意图
3. **数量 ≤ 20 条**: 超过说明在重复造 LLM 的轮子
```typescript
const SHORTCUTS: ShortcutRule[] = [
{ intent: 'order.query', pattern: /^查?订单/ },
{ intent: 'booking.query', pattern: /^我的预约/ },
{ intent: 'help.contact', pattern: /^转?人工/ },
{ intent: 'nav.home', pattern: /^回?首页/ },
];
```
---
## 多意图消歧
用户输入可能包含多个意图: "查一下订单,顺便推荐点新品"
### System Prompt 消歧指令
```
当用户输入包含多个意图时:
1. 选择最主要的意图作为 intent
2. 将次要意图放入 params._pendingIntents 数组
3. 判断标准:动作词最先出现的为主意图
```
LLM 输出:
```json
{ "intent": "order.query", "params": { "_pendingIntents": ["product.recommend"] } }
```
主意图执行完毕后通过 `followUp` 引导进入次要意图:
```typescript
if (params._pendingIntents?.length) {
response.followUp = `还需要我帮您${describeIntent(params._pendingIntents[0])}吗?`;
}
```
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |