xingyue
5970456a54
CI / check (pull_request) Failing after 8m30s
Rename packages/ subdirectories to match their @united-workforce/* scope: cli-workflow → cli workflow-agent-builtin → agent-builtin workflow-agent-claude-code → agent-claude-code workflow-agent-hermes → agent-hermes workflow-dashboard → dashboard workflow-protocol → protocol workflow-util-agent → util-agent workflow-util → util Updated all tsconfig references, scripts, and active docs. Historical docs (docs/plans/, docs/superpowers/) left as-is. Closes #21
192 lines
6.3 KiB
Markdown
192 lines
6.3 KiB
Markdown
# workflow-agent-react — ReAct Agent Package
|
|
|
|
**Status**: RFC v3
|
|
**Author**: 小橘 🍊
|
|
|
|
## Problem
|
|
|
|
现有的 agent 包都依赖外部 CLI 进程：
|
|
|
|
| Package | 机制 | 能力 |
|
|
|---------|------|------|
|
|
| `agent-hermes` | spawn `hermes chat` | 完整工具链（文件、终端、浏览器…） |
|
|
| `workflow-agent-cursor` | spawn `cursor-agent` | IDE 级别代码编辑 |
|
|
| `workflow-agent-llm` | 单轮 chat completion | 纯文本，无工具 |
|
|
|
|
缺少一个 **内置 ReAct agent**：用 LLM + tool calling 循环执行任务，不依赖外部 CLI，工具集由调用方注入。
|
|
|
|
## 核心设计变更：AdapterFn 替代 AgentFn
|
|
|
|
### 现状的问题
|
|
|
|
当前 `AgentFn` 返回 `string`，engine 再用额外一轮 LLM 调用 extract meta：
|
|
|
|
```
|
|
Agent(ctx) → string → Extract(string, schema) → meta // 浪费一轮 LLM
|
|
```
|
|
|
|
### 新抽象：AdapterFn
|
|
|
|
```typescript
|
|
type RoleFn<T> = (ctx: ThreadContext) => Promise<T>;
|
|
|
|
type AdapterFn = <T>(prompt: string, schema: z.ZodType<T>) => RoleFn<T>;
|
|
```
|
|
|
|
- **`prompt`** — role 的 system prompt，描述角色职责和输出要求
|
|
- **`schema`** — role 的 meta schema，定义输出格式
|
|
- **`ThreadContext`** — threadId, depth, bundleHash, start, steps
|
|
|
|
prompt 和 schema 是一对：prompt 说"你要输出什么"，schema 定义"输出的格式"。它们属于 role definition，由 `createWorkflow` 在每个 role 执行时传给 adapter。
|
|
|
|
### AgentContext 不再需要
|
|
|
|
`AgentContext` 在 `ThreadContext` 上扩展了 `currentRole: { name, systemPrompt }`。prompt 现在直接传给 adapter，`AgentContext` 可以删除。
|
|
|
|
### createWorkflow 签名变更
|
|
|
|
```typescript
|
|
// Before
|
|
type AgentBinding = {
|
|
agent: AgentFn;
|
|
overrides: Partial<Record<string, AgentFn>> | null;
|
|
};
|
|
|
|
// After
|
|
type AdapterBinding = {
|
|
adapter: AdapterFn;
|
|
overrides: Partial<Record<string, AdapterFn>> | null;
|
|
};
|
|
```
|
|
|
|
engine 对每个 role 的执行逻辑：
|
|
|
|
```typescript
|
|
// Before
|
|
const result = await agent({ ...threadCtx, currentRole: { name, systemPrompt } });
|
|
const meta = await extract(result, role.metaSchema, provider); // 额外一轮 LLM
|
|
|
|
// After
|
|
const roleFn = adapter(role.systemPrompt, role.metaSchema);
|
|
const meta = await roleFn(threadCtx); // 直接拿到类型安全的 T
|
|
```
|
|
|
|
## `createReactAdapter` — 复用 workflow-reactor
|
|
|
|
AdapterFn 的终止条件是"拿到符合 schema 的 T"——和 `workflow-reactor` 的 `ThreadReactorFn` 完全一致。因此 react adapter 是对 reactor 的**薄包装**，不需要自己实现 ReAct 循环。
|
|
|
|
```typescript
|
|
import { createLlmFn, createThreadReactor } from "@uncaged/workflow-reactor";
|
|
import type { ThreadContext, LlmProvider } from "@uncaged/protocol";
|
|
import type { ToolDefinition } from "@uncaged/workflow-reactor";
|
|
|
|
type ReactToolHandler = (name: string, args: string) => Promise<string>;
|
|
|
|
type ReactAdapterConfig = {
|
|
provider: LlmProvider;
|
|
tools: readonly ToolDefinition[];
|
|
toolHandler: ReactToolHandler;
|
|
maxRounds: number;
|
|
};
|
|
|
|
function createReactAdapter(config: ReactAdapterConfig): AdapterFn {
|
|
return <T>(prompt: string, schema: z.ZodType<T>) => {
|
|
const reactor = createThreadReactor<ThreadContext>({
|
|
llm: createLlmFn(config.provider),
|
|
staticTools: config.tools,
|
|
structuredToolFromSchema: (s) => buildStructuredTool(s),
|
|
systemPromptForStructuredTool: () => prompt,
|
|
toolHandler: (call, ctx) =>
|
|
config.toolHandler(call.function.name, call.function.arguments),
|
|
maxRounds: config.maxRounds,
|
|
});
|
|
|
|
return async (ctx: ThreadContext): Promise<T> => {
|
|
const input = buildThreadInput(ctx);
|
|
const result = await reactor({ thread: ctx, input, schema });
|
|
if (!result.ok) throw new Error(result.error);
|
|
return result.value;
|
|
};
|
|
};
|
|
}
|
|
```
|
|
|
|
整个包就是：**一个工厂函数 + 类型定义 + thread 输入构造**。
|
|
|
|
## `agentToAdapter` — 向后兼容
|
|
|
|
把现有 `AgentFn`（hermes/cursor）包装成 `AdapterFn`：
|
|
|
|
```typescript
|
|
function agentToAdapter(agent: AgentFn, extractProvider: LlmProvider): AdapterFn {
|
|
return <T>(prompt: string, schema: z.ZodType<T>): RoleFn<T> => {
|
|
return async (ctx: ThreadContext): Promise<T> => {
|
|
const agentCtx = { ...ctx, currentRole: { name: "agent", systemPrompt: prompt } };
|
|
const result = await agent(agentCtx);
|
|
const output = typeof result === "string" ? result : result.output;
|
|
return extract(output, schema, extractProvider);
|
|
};
|
|
};
|
|
}
|
|
```
|
|
|
|
hermes/cursor agent 内部不改，bundle-entry 层多包一层即可。
|
|
|
|
## 包结构
|
|
|
|
```
|
|
packages/workflow-agent-react/
|
|
src/
|
|
types.ts # ReactAdapterConfig, ReactToolHandler
|
|
create-react-adapter.ts # AdapterFn 工厂（包装 reactor）
|
|
thread-input.ts # ThreadContext → user message string
|
|
index.ts
|
|
__tests__/
|
|
create-react-adapter.test.ts
|
|
package.json
|
|
```
|
|
|
|
依赖：
|
|
- `@uncaged/protocol` — `ThreadContext`, `LlmProvider`
|
|
- `@uncaged/workflow-reactor` — `createLlmFn`, `createThreadReactor`, types
|
|
|
|
## 影响范围
|
|
|
|
### Breaking Changes
|
|
|
|
| 改动 | 影响 |
|
|
|------|------|
|
|
| `AgentBinding` → `AdapterBinding` | `createWorkflow` 调用方（所有 bundle-entry） |
|
|
| `AgentContext` 删除 | `buildAgentPrompt`（util-agent）改为接收 `ThreadContext` |
|
|
| extract 从 engine 下沉到 adapter | `workflow-execute` 简化 |
|
|
|
|
### 需修改的包
|
|
|
|
1. `protocol` — 删除 `AgentContext`/`AgentFn`/`AgentFnResult`/`AgentBinding`，新增 `AdapterFn`/`RoleFn`/`AdapterBinding`
|
|
2. `workflow-runtime` — 更新 re-export
|
|
3. `workflow-execute` — engine 调用 `adapter(prompt, schema)` 替代 `agent(ctx) + extract`
|
|
4. `util-agent` — `buildAgentPrompt` → `buildThreadInput`，接收 `ThreadContext`
|
|
5. 所有 bundle-entry — `agent:` → `adapter:`
|
|
|
|
### 不受影响
|
|
|
|
- `workflow-cas` / `workflow-register` / `workflow-reactor` / `dashboard`
|
|
- `agent-hermes` / `workflow-agent-cursor`（内部不改，外部用 `agentToAdapter` 包装）
|
|
|
|
## Phases
|
|
|
|
1. **Phase 1**: protocol 类型 + `createWorkflow` 签名变更 + `agentToAdapter`
|
|
2. **Phase 2**: `workflow-agent-react` 包（包装 reactor）
|
|
3. **Phase 3**: 工具集实现（read/write/patch/shell） + smoke test 闭环
|
|
|
|
## 工具集（后续讨论）
|
|
|
|
| 工具 | 说明 | 优先级 |
|
|
|------|------|--------|
|
|
| `read_file` | 读文件 | P0 |
|
|
| `write_file` | 写文件 | P0 |
|
|
| `patch_file` | find-and-replace 编辑 | P0 |
|
|
| `shell_exec` | 执行 shell 命令 | P0 |
|
|
| `search_files` | grep / find | P1 |
|
|
| `list_files` | ls | P1 |
|