mirror of https://github.com/claude-code-best/claude-code.git synced 2026-06-23 16:55:51 +00:00

Files

claude-code-best 8f6d4f88dd chore(workflow-engine): 封包发布到 npm

- 移除 private，补全 exports/types/files/publishConfig/license/repository 等
- 添加 LICENSE (MIT) 与 README
- 添加 scripts/build.ts + tsconfig.build.json，用 tsc emit 输出 dist/**/*.js + .d.ts
  （Bun bundle + external zod 会丢失 createWorkflowTool/workflowInputSchema/persistInlineScript 符号，改用 tsc emit）
- 修 src/index.ts 的 WORKFLOW_TOOL_NAME 重复 export；tool/* 的 named re-export 改为 import + 再 export

Co-Authored-By: glm-5.2 <zai-org@claude-code-best.win>

2026-06-22 20:09:24 +08:00

2.4 KiB

Raw Blame History

@claude-code-best/workflow-engine

Deterministic JS script orchestration engine for multi-agent workflows. The core layer has zero runtime dependencies and talks to the outside world exclusively through port adapters — you bring your own agent backend, journal store, and progress sink.

Why

When you orchestrate multiple LLM agents, you want the orchestration itself to be deterministic, replayable, and testable. This engine runs a plain JS script (compiled by Bun's transpiler) with primitives like agent(), phase(), parallel() and pipeline(). The non-deterministic parts (the LLM, the file system, the clock) are isolated behind ports, so the same script produces the same journal on every replay.

Installation

bun add @claude-code-best/workflow-engine
# or
npm install @claude-code-best/workflow-engine

Runtime peer requirements: ajv and zod are pulled in automatically as dependencies.

Minimal example

import {
  createFileJournalStore,
  createHostHandle,
  runWorkflow,
  type WorkflowPorts,
} from '@claude-code-best/workflow-engine'

const script = `
export const meta = { name: 'hello', description: 'minimal demo' }
phase('Greet')
const reply = await agent({ prompt: 'Say hi in one short sentence.' })
emit('result', { reply })
`

const ports: WorkflowPorts = {
  // Provide your own agent runner + journal + progress emitter.
  // See examples/smoke.ts for a complete Anthropic SDK wiring.
} as WorkflowPorts

const handle = createHostHandle()
await runWorkflow({
  script,
  ports,
  workflowDir: '.wfe/runs/hello',
  hostHandle: handle,
})

For a fully wired end-to-end example with the Anthropic SDK, see examples/smoke.ts.

Core primitives

agent(params) — call the configured AgentRunner; supports structured-output via JSON Schema.
phase(name) — declare a logical phase (display + progress grouping).
parallel([...]) — barrier-style fan-out with bounded concurrency.
pipeline(stream, fn) — streaming pipeline with per-item hooks.
emit(type, payload) — emit a progress event to the host.
log.* / hooks / budgets — see the TypeScript definitions for the full surface.

Building from source

bun install            # from the repo root
bun run build          # outputs dist/index.js + dist/**/*.d.ts
bun test               # 178 tests

2.4 KiB Raw Blame History