taskmeld 0.1.2 → 0.1.41

package/README.zh-CN.md

@@ -1,176 +1,176 @@
-<p align="center">
-  <a href="./README.md">English</a>
-  &nbsp;·&nbsp;
-  <strong>简体中文</strong>
-  &nbsp;·&nbsp;
-  <a href="https://taskmeld.com">Website</a>
-</p>
-
-<p align="center">
-  <a href="https://www.npmjs.com/package/taskmeld"><img src="https://img.shields.io/npm/v/taskmeld.svg?style=flat-square&color=cb3837&labelColor=161b22&logo=npm&logoColor=white" alt="npm version"/></a>
-  <a href="./package.json"><img src="https://img.shields.io/node/v/taskmeld.svg?style=flat-square&color=5fa04e&labelColor=161b22&logo=nodedotjs&logoColor=white" alt="node"/></a>
-</p>
-
-<br/>
-
-<h3 align="center">Agent 流水线编排平台</h3>
-<p align="center">将 OpenClaw Agent 编排为可执行流水线——定义、运行、观察、迭代。文件持久化，零外部数据库。</p>
-
-<br/>
-
-> [!TIP]
-> TaskMeld 是 OpenClaw 的流水线运行时。OpenClaw 负责 Agent 执行，TaskMeld 负责将 Agent 串成 DAG 工作流，含路由分流、重试和产物追踪。
-
-<br/>
-
-## 环境要求
-
-- Node ≥ 18
-- OpenClaw ≥ 5.20
-- Windows（macOS 和 Linux 尚未测试验证）
-
-## 安装
-
-~~~bash
-npm install -g taskmeld
-~~~
-
-<br/>
-
-## 快速开始
-
-~~~bash
-# 初始化 — 引导式配置 OpenClaw Gateway 连接
-taskmeld init
-
-# 启动后端守护进程
-taskmeld server start
-
-# 查看可用流水线
-taskmeld pipeline list
-
-# 运行一条流水线
-taskmeld pipeline start <pipelineId>
-
-# 实时监听流水线运行
-taskmeld pipeline watch <pipelineId>
-~~~
-
-| 命令 | 场景 |
-|---|---|
-| `taskmeld pipeline list` | 查看有哪些流水线 |
-| `taskmeld pipeline start <id>` | 启动一条流水线 |
-| `taskmeld pipeline watch <id>` | 通过 WebSocket 实时跟踪运行 |
-| `taskmeld pipeline status <id>` | 一次性状态快照 |
-| `taskmeld pipeline stop <id>` | 停止运行中的流水线 |
-| `taskmeld pipeline retry-node <id> <node>` | 重试失败的节点 |
-| `taskmeld server start` | 启动后端守护进程 |
-| `taskmeld agent list` | 列出已注册的 Agent |
-| `taskmeld artifact list` | 浏览流水线产物 |
-
-完整命令参考：`taskmeld --help` 或 [CLI 文档](docs/cli.md)。
-
-<br/>
-
-## 特性
-
-- **DAG 流水线引擎** — 节点依赖图、并行组、路由分支、节点级重试、状态持久化
-- **CLI 工具** — 全生命周期管理：list, run, status, stop, retry, watch（WebSocket 流式监听）
-- **HTTP + WebSocket API** — REST 端点用于控制面，WS 广播用于实时可观测性
-- **Web 控制台** — React 19 仪表盘，含 DAG 可视化、Agent 会话、产物浏览器、日志查看器
-- **Gateway 集成** — WebSocket 客户端，对接 OpenClaw Gateway 鉴权、事件中继与 Agent 会话委托
-- **文件持久化** — 所有状态以 JSON 和日志文件默认存储在 `~/.taskmeld/` 下，可用 `TASKMELD_DATA_DIR` 覆盖，无需外部数据库
-
-<br/>
-
-## 架构
-
-```
-CLI (taskmeld)  ·  Web 控制台 (React)
-        │                │
-   HTTP API ─────── WS Broker
-        │                │
-     App Assembly (注册表 + 运行时)
-              │
-     Pipeline Engine (DAG · 调度器 · 状态机)
-              │
-     Gateway Client (OpenClaw — 鉴权、事件、会话)
-```
-
-| 目录 | 说明 |
-|------|------|
-| `src/cli/` | CLI 入口、路由、输出渲染 |
-| `src/pipeline/` | 流水线引擎（运行时、调度器、执行、DAG） |
-| `src/server/` | HTTP API 服务 + 路由模块 |
-| `src/transport/` | WebSocket 广播 |
-| `src/gateway/` | 外部 Gateway WebSocket 客户端 |
-| `src/services/` | 服务层（读写 facade） |
-| `src/app/` | 应用装配（注册表、运行时、上下文） |
-| `src/artifacts/` | 产物存储 |
-| `src/logs/` | 时间线日志 |
-| `web/` | React 管理前端 |
-
-<br/>
-
-## 开发状态
-
-> [!WARNING]
-> TaskMeld 当前处于初始测试阶段。功能正在逐步构建，API 可能在版本之间变化，部分界面仍较为粗糙。生产环境使用请自行评估——欢迎早期用户试用和反馈。
-
-<br/>
-
-## 后续规划
-
-### 现状
-
-- **流水线执行** — 节点驱动模式，每个节点绑定一个 OpenClaw Agent。CLI 已暴露完整命令集，外部 Agent 可通过编程方式调用（`pipeline list`、`pipeline start`、`pipeline status` 等）。
-- **Agent 管理** — 以只读操作为主（对话、编辑核心文件如 `agent.md` / `memory.md` / `soul.md`）。创建 Agent、配置 Skill 等操作仍需切换到 OpenClaw 中完成。
-- **数据存储** — 基于文件持久化（默认 `~/.taskmeld/` 下的 JSON + 日志文件），零外部依赖。
-
-### 计划
-
-- **内置自主 Agent** — 作为一等运行时组件，全权负责流水线的完整生命周期：调度运行、创建与审查流水线定义、故障分类、产物整理。目标是从*你来操作流水线*过渡到*Agent 操作流水线，你来指挥*。
-- **Agent 生命周期管理** — 创建 Agent、配置 Skill、编辑核心文件等操作统一收敛到 TaskMeld 内，由内置 Agent 驱动，不再需要切换到 OpenClaw。
-- **数据库存储层** — 从文件持久化演进为数据库存储，提升查询性能、并发访问和可扩展性，同时保留单节点零依赖的轻量体验。
-
-<br/>
-
-## 开发
-
-```bash
-npm install          # 安装依赖
-npm run build        # 构建
-npm run typecheck    # 仅类型检查
-npm run lint         # 代码检查
-npm test             # 运行测试
-npm run dev:web      # 启动前端开发服务器（Vite HMR）
-```
-
-| 层面 | 技术 |
-|------|------|
-| 语言 | TypeScript（strict, CommonJS） |
-| 运行时 | Node.js |
-| 后端 HTTP | Node.js 内置 `http` |
-| WebSocket | `ws` |
-| 前端 | React 19 + Vite 7 |
-| CSS | Tailwind CSS 4 |
-| 测试 | Vitest |
-| Lint | ESLint 9 |
-
-<br/>
-
-## 文档
-
-- [CLI 参考](docs/cli.md)
-- [后端架构](docs/backend.md)
-- [前端架构](docs/web.md)
-- [流水线引擎](docs/pipeline/)
-- [参与贡献](CONTRIBUTING.md)
-
-<br/>
-
----
-
-<p align="center">
-  <sub>MIT — 详见 <a href="./LICENSE">LICENSE</a></sub>
-</p>
+<p align="center">
+  <a href="./README.md">English</a>
+  &nbsp;·&nbsp;
+  <strong>简体中文</strong>
+  &nbsp;·&nbsp;
+  <a href="https://taskmeld.com">Website</a>
+</p>
+
+<p align="center">
+  <a href="https://www.npmjs.com/package/taskmeld"><img src="https://img.shields.io/npm/v/taskmeld.svg?style=flat-square&color=cb3837&labelColor=161b22&logo=npm&logoColor=white" alt="npm version"/></a>
+  <a href="./package.json"><img src="https://img.shields.io/node/v/taskmeld.svg?style=flat-square&color=5fa04e&labelColor=161b22&logo=nodedotjs&logoColor=white" alt="node"/></a>
+</p>
+
+<br/>
+
+<h3 align="center">Agent 流水线编排平台</h3>
+<p align="center">将 OpenClaw Agent 编排为可执行流水线——定义、运行、观察、迭代。文件持久化，零外部数据库。</p>
+
+<br/>
+
+> [!TIP]
+> TaskMeld 是 OpenClaw 的流水线运行时。OpenClaw 负责 Agent 执行，TaskMeld 负责将 Agent 串成 DAG 工作流，含路由分流、重试和产物追踪。
+
+<br/>
+
+## 环境要求
+
+- Node ≥ 18
+- OpenClaw ≥ 5.20
+- Windows（macOS 和 Linux 尚未测试验证）
+
+## 安装
+
+~~~bash
+npm install -g taskmeld
+~~~
+
+<br/>
+
+## 快速开始
+
+~~~bash
+# 初始化 — 引导式配置 OpenClaw Gateway 连接
+taskmeld init
+
+# 启动后端守护进程
+taskmeld server start
+
+# 查看可用流水线
+taskmeld pipeline list
+
+# 运行一条流水线
+taskmeld pipeline start <pipelineId>
+
+# 实时监听流水线运行
+taskmeld pipeline watch <pipelineId>
+~~~
+
+| 命令 | 场景 |
+|---|---|
+| `taskmeld pipeline list` | 查看有哪些流水线 |
+| `taskmeld pipeline start <id>` | 启动一条流水线 |
+| `taskmeld pipeline watch <id>` | 通过 WebSocket 实时跟踪运行 |
+| `taskmeld pipeline status <id>` | 一次性状态快照 |
+| `taskmeld pipeline stop <id>` | 停止运行中的流水线 |
+| `taskmeld pipeline retry-node <id> <node>` | 重试失败的节点 |
+| `taskmeld server start` | 启动后端守护进程 |
+| `taskmeld agent list` | 列出已注册的 Agent |
+| `taskmeld artifact list` | 浏览流水线产物 |
+
+完整命令参考：`taskmeld --help` 或 [CLI 文档](docs/cli.md)。
+
+<br/>
+
+## 特性
+
+- **DAG 流水线引擎** — 节点依赖图、并行组、路由分支、节点级重试、状态持久化
+- **CLI 工具** — 全生命周期管理：list, run, status, stop, retry, watch（WebSocket 流式监听）
+- **WebSocket API** — 统一 WS 传输层，用于控制面和实时可观测性
+- **Web 控制台** — React 19 仪表盘，含 DAG 可视化、Agent 会话、产物浏览器、日志查看器
+- **Gateway 集成** — WebSocket 客户端，对接 OpenClaw Gateway 鉴权、事件中继与 Agent 会话委托
+- **文件持久化** — 所有状态以 JSON 和日志文件默认存储在 `~/.taskmeld/` 下，可用 `TASKMELD_DATA_DIR` 覆盖，无需外部数据库
+
+<br/>
+
+## 架构
+
+```
+CLI (taskmeld)  ·  Web 控制台 (React)
+        │                │
+   WS RPC ─────── WS Broker
+        │                │
+     App Assembly (注册表 + 运行时)
+              │
+     Pipeline Engine (DAG · 调度器 · 状态机)
+              │
+     Gateway Client (OpenClaw — 鉴权、事件、会话)
+```
+
+| 目录 | 说明 |
+|------|------|
+| `src/cli/` | CLI 入口、路由、输出渲染 |
+| `src/pipeline/` | 流水线引擎（运行时、调度器、执行、DAG） |
+| `src/server/` | HTTP 服务（健康检查 + 静态文件） |
+| `src/transport/` | WebSocket 传输层（广播 + RPC 方法） |
+| `src/gateway/` | 外部 Gateway WebSocket 客户端 |
+| `src/services/` | 服务层（读写 facade） |
+| `src/app/` | 应用装配（注册表、运行时、上下文） |
+| `src/artifacts/` | 产物存储 |
+| `src/logs/` | 时间线日志 |
+| `web/` | React 管理前端 |
+
+<br/>
+
+## 开发状态
+
+> [!WARNING]
+> TaskMeld 当前处于初始测试阶段。功能正在逐步构建，API 可能在版本之间变化，部分界面仍较为粗糙。生产环境使用请自行评估——欢迎早期用户试用和反馈。
+
+<br/>
+
+## 后续规划
+
+### 现状
+
+- **流水线执行** — 节点驱动模式，每个节点绑定一个 OpenClaw Agent。CLI 已暴露完整命令集，外部 Agent 可通过编程方式调用（`pipeline list`、`pipeline start`、`pipeline status` 等）。
+- **Agent 管理** — 以只读操作为主（对话、编辑核心文件如 `agent.md` / `memory.md` / `soul.md`）。创建 Agent、配置 Skill 等操作仍需切换到 OpenClaw 中完成。
+- **数据存储** — 基于文件持久化（默认 `~/.taskmeld/` 下的 JSON + 日志文件），零外部依赖。
+
+### 计划
+
+- **内置自主 Agent** — 作为一等运行时组件，全权负责流水线的完整生命周期：调度运行、创建与审查流水线定义、故障分类、产物整理。目标是从*你来操作流水线*过渡到*Agent 操作流水线，你来指挥*。
+- **Agent 生命周期管理** — 创建 Agent、配置 Skill、编辑核心文件等操作统一收敛到 TaskMeld 内，由内置 Agent 驱动，不再需要切换到 OpenClaw。
+- **数据库存储层** — 从文件持久化演进为数据库存储，提升查询性能、并发访问和可扩展性，同时保留单节点零依赖的轻量体验。
+
+<br/>
+
+## 开发
+
+```bash
+npm install          # 安装依赖
+npm run build        # 构建
+npm run typecheck    # 仅类型检查
+npm run lint         # 代码检查
+npm test             # 运行测试
+npm run dev:web      # 启动前端开发服务器（Vite HMR）
+```
+
+| 层面 | 技术 |
+|------|------|
+| 语言 | TypeScript（strict, CommonJS） |
+| 运行时 | Node.js |
+| 后端 HTTP | Node.js 内置 `http` |
+| WebSocket | `ws` |
+| 前端 | React 19 + Vite 7 |
+| CSS | Tailwind CSS 4 |
+| 测试 | Vitest |
+| Lint | ESLint 9 |
+
+<br/>
+
+## 文档
+
+- [CLI 参考](docs/cli.md)
+- [后端架构](docs/backend.md)
+- [前端架构](docs/web.md)
+- [流水线引擎](docs/pipeline/)
+- [参与贡献](CONTRIBUTING.md)
+
+<br/>
+
+---
+
+<p align="center">
+  <sub>MIT — 详见 <a href="./LICENSE">LICENSE</a></sub>
+</p>

package/dist/src/app/app-context-env.js

@@ -1,7 +1,7 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.resolveAppContextConfig = exports.DEFAULT_ITEM_KEYS = exports.DEFAULT_GATEWAY_SCOPES = exports.DEFAULT_WEB_ORIGIN = exports.DEFAULT_API_HOST = exports.DEFAULT_API_PORT = void 0;
-exports.DEFAULT_API_PORT = 3100;
+exports.DEFAULT_API_PORT = 54320;
 exports.DEFAULT_API_HOST = "0.0.0.0";
 exports.DEFAULT_WEB_ORIGIN = "*";
 exports.DEFAULT_GATEWAY_SCOPES = ["operator.read", "operator.write", "operator.admin"];

package/dist/src/app/create-app-context.js

@@ -35,8 +35,8 @@ const createAppContext = (options = {}) => {
     const ensureGatewayClient = () => {
         if (clientRef)
             return clientRef;
-        // dev/server 进程允许先在“网关未配置/未就绪”状态下启动 HTTP/WS，
-        // 真正触发 connect/sendReq 时再校验凭据，才能兼容本地调试与需要网关的命令两类路径。
+        // dev/server processes allow HTTP/WS to start even before gateway is configured/ready,
+        // deferring credential validation to when connect/sendReq actually fires, to support both local debugging and gateway-required commands.
         const credentials = resolveGatewayCredentials(options, env);
         clientRef = (0, gateway_1.createGatewayClient)({
             gatewayUrl: credentials.url,
@@ -83,7 +83,7 @@ const createAppContext = (options = {}) => {
     });
     appRef = app;
     const appServices = (0, services_1.createAppServices)(app);
-    // 网关握手成功后需要把 hello 回灌到 registry，保证运行态上下文完整。
+    // After the gateway handshake succeeds, feed the hello payload back into the registry so runtime context stays complete.
     const connect = async () => {
         const hello = await client.connect();
         app.onGatewayReady(hello);

package/dist/src/app/data-dir.js

@@ -1,9 +1,11 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.resolveTaskMeldDataPath = exports.getTaskMeldDataDir = exports.isTaskMeldTestRuntime = void 0;
+exports.resolveTaskMeldDataPath = exports.getTaskMeldDataDir = exports.isTaskMeldDevRuntime = exports.isTaskMeldTestRuntime = void 0;
 const node_os_1 = require("node:os");
 const node_path_1 = require("node:path");
 const TEST_ARG_PATTERN = /(^|[\\/])(test|dist[\\/]test)[\\/]/;
+// Detect tsx / ts-node etc. dev-mode run (e.g. `npx tsx src/index.ts`)
+const DEV_ARG_PATTERN = /(^|[\\/])tsx[\\/]|(^|[\\/])(src[\\/]index\.ts)$/;
 const isTaskMeldTestRuntime = (options = {}) => {
     const env = options.env ?? process.env;
     const argv = options.argv ?? process.argv;
@@ -13,13 +15,21 @@ const isTaskMeldTestRuntime = (options = {}) => {
     return argv.some((arg) => TEST_ARG_PATTERN.test(arg));
 };
 exports.isTaskMeldTestRuntime = isTaskMeldTestRuntime;
+const isTaskMeldDevRuntime = (options = {}) => {
+    const env = options.env ?? process.env;
+    const argv = options.argv ?? process.argv;
+    if (env.NODE_ENV === "development")
+        return true;
+    return argv.some((arg) => DEV_ARG_PATTERN.test(arg));
+};
+exports.isTaskMeldDevRuntime = isTaskMeldDevRuntime;
 const getTaskMeldDataDir = (options = {}) => {
     const env = options.env ?? process.env;
     const override = env.TASKMELD_DATA_DIR?.trim();
     if (override)
         return override;
-    // 测试用例默认隔离到当前仓库，避免污染用户真实的 ~/.taskmeld 数据。
-    if ((0, exports.isTaskMeldTestRuntime)({ env, argv: options.argv })) {
+    // Test/dev cases default to isolation in the current repo to avoid polluting the user's real ~/.taskmeld data.
+    if ((0, exports.isTaskMeldTestRuntime)({ env, argv: options.argv }) || (0, exports.isTaskMeldDevRuntime)({ env, argv: options.argv })) {
         return (0, node_path_1.join)(options.cwd ?? process.cwd(), ".data");
     }
     return (0, node_path_1.join)(options.homeDir ?? (0, node_os_1.homedir)(), ".taskmeld");

package/dist/src/app/pipeline-config.js

@@ -15,7 +15,7 @@ const createPipelineDefinition = (id, title) => {
     const baseDir = (0, node_path_1.join)(PIPELINE_ROOT_DIR, id);
     return {
         id,
-        title: title?.trim() || `流水线 DAG-${id}`,
+        title: title?.trim() || `Pipeline ${id}`,
         workflowFilePath: (0, node_path_1.join)(baseDir, "workflow.json"),
         runStateFile: (0, node_path_1.join)(baseDir, "run-state.json"),
         artifactDir: (0, node_path_1.join)(baseDir, "artifacts"),
@@ -27,7 +27,7 @@ const createDefaultDefinitionsDocument = () => ({
     defaultPipelineId: DEFAULT_PIPELINE_ID_FALLBACK,
     items: DEFAULT_PIPELINE_IDS.map((pipelineId) => ({
         id: pipelineId,
-        title: `流水线 DAG-${pipelineId}`,
+        title: `Pipeline ${pipelineId}`,
     })),
 });
 const isValidPipelineId = (value) => typeof value === "string" && PIPELINE_ID_PATTERN.test(value.trim());
@@ -43,7 +43,7 @@ const normalizeDefinitionItems = (items) => {
         if (!(0, exports.isValidPipelineId)(record.id))
             continue;
         const id = record.id.trim();
-        const title = typeof record.title === "string" && record.title.trim() ? record.title.trim() : `流水线 DAG-${id}`;
+        const title = typeof record.title === "string" && record.title.trim() ? record.title.trim() : `Pipeline ${id}`;
         deduped.set(id, { id, title });
     }
     return [...deduped.values()];
@@ -82,7 +82,7 @@ exports.savePipelineDefinitions = savePipelineDefinitions;
 const ensurePipelineDefinitions = () => {
     const fromDisk = readDefinitionsDocumentFromDisk();
     if (fromDisk) {
-        // 每次启动都按当前约束回写一遍，避免 defaultPipelineId 指向失效项或 title 为空。
+        // Rewrite on every startup according to current constraints, to prevent defaultPipelineId pointing to an invalid entry or title being empty.
         return (0, exports.savePipelineDefinitions)(fromDisk);
     }
     return (0, exports.savePipelineDefinitions)(createDefaultDefinitionsDocument());

package/dist/src/app/pipeline-registry.js

@@ -30,7 +30,7 @@ const buildArchivedPipelineDirPath = (pipelineId) => {
     (0, node_fs_1.mkdirSync)(deletedRootDir, { recursive: true });
     let archiveDirPath = (0, node_path_1.join)(deletedRootDir, createArchivedPipelineDirName(pipelineId));
     let suffix = 1;
-    // 归档目录需要保证唯一，避免同秒内重复删除时互相覆盖。
+    // Archive directory must be unique to avoid collisions when multiple deletes happen within the same second.
     while ((0, node_fs_1.existsSync)(archiveDirPath)) {
         archiveDirPath = (0, node_path_1.join)(deletedRootDir, `${createArchivedPipelineDirName(pipelineId)}-${suffix}`);
         suffix += 1;
@@ -184,7 +184,7 @@ const createPipelineRegistry = (options) => {
                 });
                 return;
             }
-            // 网关状态/握手是全局共享事件，只透传主流水线的一份，避免前端收到重复广播。
+            // Gateway status/handshake is a globally shared event; only relay one copy from the primary pipeline to avoid duplicate broadcasts to frontends.
             if ((event.type === "gateway.status" || event.type === "gateway.ready" || event.type === "gateway.frame") &&
                 definition.id !== getPrimaryPipelineId()) {
                 return;
@@ -241,7 +241,7 @@ const createPipelineRegistry = (options) => {
             timelineHasMore: combined.length > MAX_BOOTSTRAP_TIMELINE,
             status: getPrimaryRuntime().gateway.getLatestStatus() ?? options.client.getStatus(),
             hello: getPrimaryRuntime().gateway.getLatestHello(),
-            // 保留旧字段兜底，避免前端分阶段改造时直接失效。
+            // Keep the old top-level fields as a fallback to avoid breaking frontends during phased migration.
             run: primary?.run,
             pipeline: primary?.pipeline,
             runId: primary?.runId,
@@ -249,8 +249,8 @@ const createPipelineRegistry = (options) => {
         };
     };
     const broadcastBootstrapPayload = () => {
-        // 流水线资源发生新增、删除、重命名时，直接广播完整 bootstrap；
-        // 这样已连接前端可以一次性拿到最新 definitions + 运行态快照，避免自己拼补丁遗漏字段。
+        // When pipeline resources are added, deleted, or renamed, broadcast a full bootstrap immediately;
+        // this way connected frontends get the latest definitions + runtime snapshot in one message, avoiding hand-rolled patch merging with missing fields.
         broadcast({
             type: "bootstrap",
             payload: getBootstrapPayload(),
@@ -285,7 +285,7 @@ const createPipelineRegistry = (options) => {
     };
     const initializePipelineWorkflowFile = (definition, cloneFrom) => {
         if (cloneFrom) {
-            // 克隆只复制 workflow 定义，运行态和产物目录仍由新流水线独立初始化。
+            // Clone only copies the workflow definition; runtime state and artifact directories are independently initialized for the new pipeline.
             const sourceWorkflow = (0, template_1.loadWorkflowDefinitionWithStorage)({ workflowFilePath: cloneFrom.workflowFilePath });
             (0, template_1.saveWorkflowDefinitionWithStorage)(sourceWorkflow, { workflowFilePath: definition.workflowFilePath });
             return;
@@ -309,7 +309,7 @@ const createPipelineRegistry = (options) => {
                     continue;
                 await runtime.initialize();
             }
-            // 启动时扫描所有流水线，对已有 pending job 的队列触发 drain
+            // On startup, scan all pipelines and trigger drain for any queue that already has pending jobs.
             for (const definition of pipelineDefinitions) {
                 const pendingCount = inboundQueue.getPendingCount(definition.id);
                 if (pendingCount > 0) {
@@ -356,8 +356,8 @@ const createPipelineRegistry = (options) => {
             persistDefinitions([...currentDocument.items, { id: definition.id, title: definition.title }], currentDocument.defaultPipelineId);
             let runtime = null;
             try {
-                // 先把目标流水线的 workflow 文件初始化到位，再创建 runtime；
-                // 否则 runtime 构造阶段会先读到默认 workflow，导致“磁盘已克隆、内存仍是默认”的假克隆问题。
+                // Initialize the target pipeline's workflow file on disk first, then create the runtime;
+                // otherwise the runtime constructor would read the default workflow, causing a "disk already cloned, memory still default" phantom-clone issue.
                 initializePipelineWorkflowFile(definition, cloneSourceDefinition ?? undefined);
                 runtime = createRuntimeForDefinition(definition);
                 bindRuntimeBroadcast(definition, runtime);
@@ -419,7 +419,7 @@ const createPipelineRegistry = (options) => {
                 archivePipelineDirectory(pipelineDefinitions[definitionIndex]);
             }
             catch (error) {
-                // 归档失败时立即回滚 definitions，避免页面列表先把流水线删掉但目录仍保持原位。
+                // If archiving fails, immediately rollback definitions so the page listing doesn't remove the pipeline while the directory stays in place.
                 (0, pipeline_config_1.savePipelineDefinitions)(currentDocument);
                 throw error;
             }
@@ -484,7 +484,7 @@ const createPipelineRegistry = (options) => {
                 if (routed)
                     return;
             }
-            // 无法判定 sessionKey 或无匹配 runtime 时，投递全部
+            // When sessionKey can't be determined or no runtime matches, deliver to all
             for (const runtime of runtimeById.values()) {
                 runtime.onGatewayRawFrame(rawFrame);
             }

package/dist/src/app/pipeline-runtime.js

@@ -1,7 +1,6 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.createPipelineRuntime = void 0;
-const http_utils_1 = require("../server/http-utils");
 const template_1 = require("../pipeline/template");
 const execution_timeout_1 = require("../pipeline/execution-timeout");
 const runtime_model_1 = require("../pipeline/runtime-model");
@@ -45,7 +44,6 @@ const createPipelineRuntime = (options) => {
     getBatchRunId = () => schedulerService.getBatchRunState().batchRunId;
     schedulerStateAccessor = schedulerService.getSchedulerState;
     batchRunStateGetter = () => schedulerService.getBatchRunState();
-    const sendJson = (res, code, data) => (0, http_utils_1.sendJson)(res, code, data, options.webOrigin);
     const setRun = (nextRun) => {
         runtimeStore.setRun(nextRun);
         graph.syncRunGroupsFromWorkflow(nextRun);
@@ -64,7 +62,7 @@ const createPipelineRuntime = (options) => {
     };
     const onGatewayStatus = (status) => {
         runtimeStore.setLatestStatus(status);
-        runtimeStore.pushTimeline(`网关状态: ${status.status}`, status.status.includes("failed") ? "error" : "info");
+        runtimeStore.pushTimeline(`Gateway status: ${status.status}`, status.status.includes("failed") ? "error" : "info");
         runtimeStore.broadcast({
             type: "gateway.status",
             payload: status,
@@ -76,7 +74,7 @@ const createPipelineRuntime = (options) => {
         const isSilentEvent = isHealthEvent || isTickEvent;
         runtimeStore.setLastFrame(frame);
         if (frame.type === "event" && !isSilentEvent) {
-            runtimeStore.pushTimeline(`事件: ${frame.event}`, "info", {
+            runtimeStore.pushTimeline(`Event: ${frame.event}`, "info", {
                 type: "event",
                 event: frame.event,
                 seq: frame.seq ?? null,
@@ -84,8 +82,8 @@ const createPipelineRuntime = (options) => {
                 payload: frame.payload ?? null,
             });
         }
-        // health / tick 都是高频事件，只保留为 lastFrame，不推送到 timeline / ws，
-        // 避免前端与日志被刷屏，同时不影响依赖最后一帧做诊断的调试能力。
+        // health / tick are high-frequency events; only store as lastFrame, don't push to timeline/WS,
+        // to avoid flooding the frontend and logs, while still allowing diagnostic debugging that depends on the last frame.
         if (!isSilentEvent) {
             runtimeStore.broadcast({
                 type: "gateway.frame",
@@ -97,7 +95,7 @@ const createPipelineRuntime = (options) => {
         executionService.onGatewayFrame(rawFrame);
     };
     const onGatewayError = (error) => {
-        runtimeStore.pushTimeline(`网关错误: ${String(error)}`, "error");
+        runtimeStore.pushTimeline(`Gateway error: ${String(error)}`, "error");
         runtimeStore.broadcast({
             type: "gateway.error",
             payload: { message: String(error) },
@@ -105,7 +103,7 @@ const createPipelineRuntime = (options) => {
     };
     const onGatewayReady = (hello) => {
         runtimeStore.setLatestHello(hello);
-        runtimeStore.pushTimeline("网关握手完成");
+        runtimeStore.pushTimeline("Gateway handshake completed");
         runtimeStore.emitPipeline();
         runtimeStore.broadcast({
             type: "gateway.ready",
@@ -135,7 +133,6 @@ const createPipelineRuntime = (options) => {
     return {
         initialize,
         dispose: () => executionService.dispose(),
-        sendJson,
         getBootstrapPayload,
         onGatewayStatus,
         onGatewayFrame,

package/dist/src/app/runtime-store.js

@@ -31,7 +31,7 @@ const createRuntimeStore = (options) => {
             }
             catch (error) {
                 // Persistence failures should not break pipeline execution.
-                pushTimeline(`持久化运行状态失败: ${error instanceof Error ? error.message : String(error)}`, "warn");
+                pushTimeline(`Failed to persist run state: ${error instanceof Error ? error.message : String(error)}`, "warn");
             }
         })().finally(() => {
             persistRunStateInFlight = null;
@@ -69,7 +69,7 @@ const createRuntimeStore = (options) => {
             return savedRun;
         }
         catch (error) {
-            pushTimeline(`加载持久化运行状态失败: ${error instanceof Error ? error.message : String(error)}`, "warn");
+            pushTimeline(`Failed to load persisted run state: ${error instanceof Error ? error.message : String(error)}`, "warn");
             return null;
         }
     };
@@ -107,7 +107,7 @@ const createRuntimeStore = (options) => {
         options.graph.syncRunGroupsFromWorkflow(run);
         (0, runtime_model_1.syncRunNodeStatusFromItemRuns)(run);
         (0, runtime_model_1.touchRun)(run);
-        pushTimeline(`已恢复上次运行状态: ${run.id}`);
+        pushTimeline(`Restored previous run state: ${run.id}`);
         emitPipeline();
     };
     const bootstrapRun = () => {