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/patterns/04-workflow-agent.md
View on GitHub# 模式 ④ Workflow Agent (工作流型)
> 预定义多步骤 DAG,含条件分支、并行节点和人工审批
> 延迟 1-5s · 可预测性 ★★★★☆ · 适合固定业务流程
## 适用场景
- 退款/售后处理(条件检查 → 自动/手动审批 → 执行)
- 投诉处理(分类 → 升级 → 响应)
- 多步骤审批(申请 → 审核 → 执行 → 通知)
- 数据处理管道(采集 → 清洗 → 分析 → 报告)
## 不适用信号
- 流程是动态的,无法预定义 → ⑤ ReAct
- 只需单步执行 → ① CREX
- 需要跨系统多 Agent 协作 → ⑦ Multi-Agent
## 与 CREX 的关系
Workflow 是 CREX Pipeline 的增强版。CREX Pipeline 是线性串联,
Workflow 支持条件分支、并行节点和人工审批。
## 核心概念
```typescript
// 工作流定义 — 有向无环图 (DAG)
interface WorkflowDefinition {
id: string;
name: string;
trigger: string; // 触发意图
nodes: WorkflowNode[];
edges: WorkflowEdge[];
}
interface WorkflowNode {
id: string;
type: 'skill' // 执行技能
| 'condition' // 条件判断
| 'llm' // LLM 节点(轻量模型)
| 'human_approval' // 人工审批
| 'parallel' // 并行执行
| 'notify'; // 通知(短信/推送/消息)
config: Record<string, any>;
}
interface WorkflowEdge {
from: string;
to: string;
condition?: string; // 条件表达式: "result.amount > 100"
}
```
## 实现示例: 退款工作流
```
[查询订单] ──→ [检查退款条件] ─── 符合 ──→ [计算退款金额]
│ │
│ 金额≤100 ──→ [自动退款] ──→ [通知用户]
│ │
│ 金额>100 ──→ [人工审批] ──→ [退款/拒绝] ──→ [通知用户]
│
└── 不符合 ──→ [返回拒绝原因]
```
### 工作流定义
```typescript
const refundWorkflow: WorkflowDefinition = {
id: 'refund.process',
name: '退款处理',
trigger: 'order.refund',
nodes: [
{ id: 'query', type: 'skill', config: { skillId: 'order.query' } },
{ id: 'check', type: 'condition', config: {
expression: 'order.status === "completed" && order.refundable === true'
}},
{ id: 'calc', type: 'skill', config: { skillId: 'refund.calculate' } },
{ id: 'auto_check', type: 'condition', config: { expression: 'amount <= 100' } },
{ id: 'auto_refund', type: 'skill', config: { skillId: 'refund.execute' } },
{ id: 'manual_review', type: 'human_approval', config: {
assignee: 'finance_team', timeout: '24h',
timeoutAction: 'auto_approve'
}},
{ id: 'notify', type: 'notify', config: { channel: 'wechat_template' } },
{ id: 'reject', type: 'skill', config: { skillId: 'refund.reject_response' } },
],
edges: [
{ from: 'query', to: 'check' },
{ from: 'check', to: 'calc', condition: 'passed' },
{ from: 'check', to: 'reject', condition: '!passed' },
{ from: 'calc', to: 'auto_check' },
{ from: 'auto_check', to: 'auto_refund', condition: 'amount <= 100' },
{ from: 'auto_check', to: 'manual_review', condition: 'amount > 100' },
{ from: 'manual_review', to: 'auto_refund', condition: 'approved' },
{ from: 'manual_review', to: 'reject', condition: '!approved' },
{ from: 'auto_refund', to: 'notify' },
],
};
```
### 工作流引擎
```typescript
class WorkflowEngine {
async execute(
workflow: WorkflowDefinition,
initialParams: Record<string, any>,
context: AgentContext
): Promise<SkillResponse> {
let current = workflow.nodes[0];
let state = { ...initialParams };
while (current) {
const result = await this.executeNode(current, state, context);
state = { ...state, ...result };
// 人工审批节点 → 暂停并返回等待状态
if (current.type === 'human_approval' && !result._approved) {
return {
type: 'text',
content: '您的退款申请已提交,等待审核中',
actions: [{ label: '查看进度', action: 'navigate', target: `/refund/${state.refundId}` }],
followUp: null,
};
}
// 查找下一个节点
const nextEdge = workflow.edges.find(
e => e.from === current!.id && (!e.condition || this.evaluate(e.condition, state))
);
current = nextEdge ? workflow.nodes.find(n => n.id === nextEdge.to)! : null!;
}
return state._lastResponse as SkillResponse;
}
}
```
## Human-in-the-Loop 审批
人工审批节点会暂停工作流,等待人工操作:
```typescript
interface HumanApprovalConfig {
assignee: string; // 审批人/团队
timeout: string; // 超时时间: "24h", "2d"
timeoutAction: 'auto_approve' | 'auto_reject' | 'escalate';
notifyChannels: string[]; // 通知渠道
}
```
审批结果通过 Webhook 或定时轮询回写到工作流状态。
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 |