@agenticforge/agents

AgenticFORGE Agent 实现包，内置 ReAct、Plan-and-Solve、Reflection、FunctionCall、Simple、SkillAgent、WorkflowAgent 七种 Agent 工作流。

安装

npm install @agenticforge/agents

内置 Agent

| Agent | 适用场景 | |-------|----------| | SimpleAgent | 简单对话 Agent，支持多轮上下文 | | FunctionCallAgent | 工具调用驱动，适合任务型场景 | | ReActAgent | 推理-行动-观察循环，适合复杂推理 | | PlanSolveAgent | 先规划后逐步执行，适合多步骤任务 | | ReflectionAgent | 带自我反思与批评机制，适合高质量生成 | | SkillAgent | 自动路由到最合适的 Skill，适合多能力切换场景 | | WorkflowAgent | DAG 工作流编排，支持并发节点执行 |

SkillAgent

SkillAgent 将用户请求自动路由到最合适的 Skill。Skill 可以是 Markdown 文件或 TypeScript 类，详见 @agenticforge/skills。

import { SkillAgent } from "@agenticforge/agents";
import { SkillLoader } from "@agenticforge/skills";

const mdSkills = await SkillLoader.fromDirectory(".cursor/skills");

const agent = new SkillAgent({
  name: "assistant",
  llm,
  skills: [...mdSkills, new StockSkill()],
});

// 自动路由到最合适的 Skill
const reply = await agent.run("东京今天下雨吗？");

// 直接调用指定 Skill
const result = await agent.runSkill("stock-query", "苹果股票现在多少？");
console.log(result.output);

使用示例

import {FunctionCallAgent, LLMClient} from "@agenticforge/agents";
import {Tool, toolAction} from "@agenticforge/tools";
import {z} from "zod";

const calcTool = new Tool({
  name: "calculator",
  description: "执行数学计算",
  parameters: [{name: "expr", type: "string", required: true}],
  action: toolAction(z.object({expr: z.string()}), async ({expr}) => String(eval(expr))),
});

const agent = new FunctionCallAgent({
  llm: new LLMClient({provider: "openai", model: "gpt-4o"}),
  tools: [calcTool],
});

const result = await agent.run("计算 (123 + 456) * 2");
console.log(result);

WorkflowAgent

WorkflowAgent 按 DAG 拓扑顺序执行节点，无相互依赖的节点在同一波次内并发执行。每个节点的输出以 nodeId 为 key 写入共享 context，后续节点可通过 {nodeId} 插值引用。

支持节点类型：tool · llm · fn · passthrough

import { WorkflowAgent, LLMClient } from "@agenticforge/agents";
import type { WorkflowDefinition } from "@agenticforge/workflow"; // 类型定义现已独立至 @agenticforge/workflow

const agent = new WorkflowAgent({
  name: "report-workflow",
  llm: new LLMClient({ provider: "openai", model: "gpt-4o" }),
  verbose: true,
});

const definition: WorkflowDefinition = {
  name: "fan-out-report",
  nodes: [
    { id: "fetch",     type: "tool", toolName: "search", inputTemplate: "{input}",                          depends: [] },
    { id: "analyze",   type: "llm",  promptTemplate: "分析以下内容：\n{fetch}",                              depends: ["fetch"] },
    { id: "translate", type: "llm",  promptTemplate: "将以下内容翻译成英文：\n{fetch}",                      depends: ["fetch"] },
    { id: "report",    type: "llm",  promptTemplate: "综合分析与翻译写出双语简报：\n{analyze}\n{translate}",  depends: ["analyze", "translate"] },
  ],
};

// analyze 与 translate 在 fetch 完成后并发执行
const result = await agent.runWorkflow(definition, "2024年AI行业发展趋势");
console.log(result.output);
console.log(result.nodeResults); // 每个节点的耗时与状态

链接

• GitHub
• npm
• 主项目 README