tuya-platform-cli 0.1.0

package/README.md

@@ -0,0 +1,218 @@
+# tuya-platform-cli
+
+CLI tool for creating, managing, and publishing AI workflows on the Tuya platform via Chrome CDP.
+
+## 安装
+
+```bash
+# 全局安装
+npm install -g tuya-platform-cli
+
+# 或使用 npx 直接运行
+npx tuya-platform-cli --help
+```
+
+## 运行前提
+
+1. 本机 macOS 已安装 Google Chrome。
+2. Chrome 以 `--remote-debugging-port=9440` 启动。
+3. 已登录 `platform.tuya.com`（或 `us.platform.tuya.com`）。
+
+## CLI 命令
+
+```bash
+tuya-platform-cli --help
+```
+
+也可以通过 `package.json` 脚本运行：
+
+```bash
+npm run tuya -- --help
+```
+
+### 1. 手工录制 8 步
+
+默认命令，行为与之前一致：启动隔离 Chrome，并按步骤提示你手工完成操作。
+
+```bash
+npm run tuya -- \
+  --output-dir artifacts/tuya-platform-cli/manual-run \
+  --base-url https://platform.tuya.com/ai/workflow
+```
+
+等价命令：
+
+```bash
+node src/cli.js manual --output-dir artifacts/tuya-platform-cli/manual-run
+```
+
+运行后脚本会：
+
+1. 复制一份 Chrome `Default` profile 到临时目录，避免锁定现有 Chrome。
+2. 启动带 `--remote-debugging-port` 的专用 Chrome。
+3. 打开平台页面。
+4. 逐步提示你在浏览器中手动完成 8 个操作；每一步结束后回终端按 Enter。
+5. 自动导出：
+   - `report.md`
+   - `static-analysis.md`
+   - 每一步的 `network.json`、`requests/*.json`、`screenshot.png`
+   - `bundles/` 目录中的脚本响应体
+
+### 0. 执行前自检
+
+建议在自动命令前先跑一次 `doctor`，确认：
+
+- CDP 端口可连接
+- 当前页面仍是 workflow 页面
+- 登录态有效
+- `workflow/list` 可调用
+- 可选的 `workflowId` 能通过 `workflow/get` / `run-pre-check` / `test/config`
+- 可选的图片路径存在
+
+```bash
+node src/cli.js doctor \
+  --remote-debugging-port 9440 \
+  --workflow-id 5883 \
+  --region AZ \
+  --image-path artifacts/live-run/add-node-inspect.png \
+  --output-dir artifacts/live-run/doctor
+```
+
+或者：
+
+```bash
+npm run tuya:doctor -- --remote-debugging-port 9440 --workflow-id 5883 --region AZ
+```
+
+输出包括：
+
+- `doctor.json`
+- `doctor.md`
+
+### 2. 自动跑基础 8 步
+
+这个命令不会启动新 Chrome，而是直接连接到已开启 CDP 的 Chrome。
+适合你已经手动登录好平台，并希望自动执行“列表 -> 创建 -> 编辑 -> 加节点 -> 保存 -> 试运行 -> 删除”。
+
+```bash
+node src/cli.js auto-basic \
+  --remote-debugging-port 9440 \
+  --output-dir artifacts/live-run/auto-basic
+```
+
+或者：
+
+```bash
+npm run tuya:auto -- --remote-debugging-port 9440
+```
+
+前提：
+
+- 目标 Chrome 已用 `--remote-debugging-port=9440` 启动
+- 当前账号已登录 `platform.tuya.com`
+
+### 3. 补采扩展节点样本
+
+当前内置 `Loop` 和 `Monitor(定时监听)` 两种额外节点的自动采样。
+命令会连接当前 Chrome，会话内直接创建样本 workflow，并导出节点 schema 样本。
+
+```bash
+node src/cli.js sample-extra-nodes \
+  --remote-debugging-port 9440 \
+  --region AZ \
+  --output-dir artifacts/live-run/extra-nodes
+```
+
+或者：
+
+```bash
+npm run tuya:samples -- --remote-debugging-port 9440 --region AZ
+```
+
+输出包括：
+
+- `extra-node-samples-summary.json`
+- `extra-node-samples-extended.json`
+- `extra-node-samples.md`
+- `loop-node-sample-<workflowId>.json`
+- `monitor-node-sample-<workflowId>.json`
+
+### 4. 补采多分支 edges 样本
+
+这个命令会自动创建一个“开始 -> 选择器 -> 双分支大模型 -> 双结束节点”的工作流，并直接落盘保存 schema。
+
+```bash
+node src/cli.js sample-branch-edges \
+  --remote-debugging-port 9440 \
+  --region AZ \
+  --output-dir artifacts/live-run/branch-edges
+```
+
+或者：
+
+```bash
+npm run tuya:branches -- --remote-debugging-port 9440 --region AZ
+```
+
+输出包括：
+
+- `branch-edges-<workflowId>.json`
+- `branch-edges.md`
+
+### 5. 补采 trial-run 输入样本
+
+这个命令要求提供一个已经可试运行的 `workflowId`。命令会自动执行：
+
+- `run-pre-check`
+- `test/config`
+- `test/config/save`
+- 文本输入执行
+- 可选的图文输入执行
+
+文本 + 图片样本需要显式传入图片路径：
+
+```bash
+node src/cli.js sample-trial-inputs \
+  --remote-debugging-port 9440 \
+  --workflow-id 5883 \
+  --region AZ \
+  --image-path artifacts/live-run/add-node-inspect.png \
+  --output-dir artifacts/live-run/trial-inputs
+```
+
+或者：
+
+```bash
+npm run tuya:trial-inputs -- --remote-debugging-port 9440 --workflow-id 5883 --region AZ
+```
+
+## 通用参数
+
+```text
+--base-url
+--chrome-path
+--chrome-profile-root
+--chrome-profile-name
+--output-dir
+--remote-debugging-port
+--workflow-name
+--workflow-id
+--region
+--image-path
+--headless
+```
+
+## 推荐流程
+
+1. 在专用 Chrome 窗口中先确认登录态仍然有效。
+2. 严格按终端提示逐步执行，避免额外点击造成噪音请求。
+3. 如果某一步明显失败，不要继续误操作；先结束该步，让报告记录失败痕迹。
+4. 试运行步骤结束后等待流式响应完全停止，再回终端确认。
+
+## 局限
+
+- 画布基于 Canvas/WebGL，脚本不会尝试解析节点视觉结构。
+- 若目标请求完全隐藏在系统层无法被 CDP 获取，可在报告基础上结合代理抓包继续补充。
+- 报告中的 schema 摘要来自实测保存 payload 和 bundle 静态分析，不是官方协议文档。
+- `auto-basic` 和 `sample-extra-nodes` 强依赖当前登录态，以及页面文案/布局没有大改。
+- `sample-trial-inputs` 依赖当前账号下已有可运行 workflow，且试运行授权已经完成。

package/examples/rag-workflow.json

@@ -0,0 +1,63 @@
+{
+  "name": "RAG知识库问答工作流",
+  "description": "用户输入 → 知识库检索 → 大模型生成回答 → 输出",
+  "region": "AZ",
+  "startOutputs": [
+    { "id": "USER_TEXT", "type": "string", "required": true }
+  ],
+  "nodes": [
+    {
+      "id": "sk_1",
+      "type": "SearchKnowledge",
+      "title": "知识库检索",
+      "inputs": [
+        { "id": "q1", "ref": "start.USER_TEXT", "name": "query" }
+      ],
+      "params": {
+        "libResource": [
+          {
+            "canUse": true,
+            "libCategories": [],
+            "libCode": "YOUR_LIB_CODE",
+            "libName": "你的知识库名称",
+            "supportCategory": true
+          }
+        ],
+        "maxRecallNum": 5,
+        "searchStrategy": 1,
+        "similarity": 0.7
+      },
+      "outputs": [
+        { "id": "d1", "name": "data", "type": "string" }
+      ]
+    },
+    {
+      "id": "llm_1",
+      "type": "LLM",
+      "title": "RAG回答生成",
+      "inputs": [
+        { "id": "in1", "ref": "sk_1.d1", "name": "input" }
+      ],
+      "params": {
+        "model": "160",
+        "systemPrompt": "你是一个RAG问答助手。优先依据知识库检索结果回答，不要编造；如果检索信息不足，直接说明信息不足。",
+        "userPrompt": "请结合检索到的知识内容，对用户问题给出简洁准确的回答。"
+      },
+      "outputs": [
+        { "id": "out1", "name": "output", "type": "string" }
+      ]
+    }
+  ],
+  "ends": [
+    {
+      "id": "end_1",
+      "title": "输出",
+      "input": { "id": "in_end", "ref": "llm_1.out1" }
+    }
+  ],
+  "edges": [
+    ["start", "sk_1"],
+    ["sk_1", "llm_1"],
+    ["llm_1", "end_1"]
+  ]
+}

package/examples/simple-llm.json

@@ -0,0 +1,37 @@
+{
+  "name": "简单LLM工作流",
+  "description": "最小示例：用户输入 → 大模型回答 → 输出",
+  "region": "AZ",
+  "startOutputs": [
+    { "id": "USER_TEXT", "type": "string", "required": true }
+  ],
+  "nodes": [
+    {
+      "id": "llm_1",
+      "type": "LLM",
+      "title": "大模型",
+      "inputs": [
+        { "id": "in1", "ref": "start.USER_TEXT", "name": "input" }
+      ],
+      "params": {
+        "model": "160",
+        "systemPrompt": "你是一个有帮助的助手。",
+        "userPrompt": "请回答用户的问题。"
+      },
+      "outputs": [
+        { "id": "out1", "name": "output", "type": "string" }
+      ]
+    }
+  ],
+  "ends": [
+    {
+      "id": "end_1",
+      "title": "输出",
+      "input": { "id": "in_end", "ref": "llm_1.out1" }
+    }
+  ],
+  "edges": [
+    ["start", "llm_1"],
+    ["llm_1", "end_1"]
+  ]
+}

package/package.json

@@ -0,0 +1,40 @@
+{
+  "name": "tuya-platform-cli",
+  "version": "0.1.0",
+  "description": "CLI tool for creating, managing, and publishing AI workflows on the Tuya platform via Chrome CDP.",
+  "type": "module",
+  "bin": {
+    "tuya-platform-cli": "./src/cli.js"
+  },
+  "files": [
+    "src/",
+    "examples/",
+    "README.md"
+  ],
+  "engines": {
+    "node": ">=18.0.0"
+  },
+  "keywords": [
+    "tuya",
+    "workflow",
+    "ai",
+    "cdp",
+    "cli"
+  ],
+  "scripts": {
+    "tuya:doctor": "node src/cli.js doctor",
+    "tuya": "node src/cli.js manual",
+    "tuya:auto": "node src/cli.js auto-basic",
+    "tuya:samples": "node src/cli.js sample-extra-nodes",
+    "tuya:branches": "node src/cli.js sample-branch-edges",
+    "tuya:trial-inputs": "node src/cli.js sample-trial-inputs",
+    "tuya:configure": "node src/cli.js configure",
+    "tuya:publish": "node src/cli.js publish",
+    "tuya:list-libraries": "node src/cli.js list-libraries",
+    "test": "node --test"
+  },
+  "dependencies": {
+    "playwright": "^1.58.2",
+    "ws": "^8.18.3"
+  }
+}

package/src/cli.js

@@ -0,0 +1,78 @@
+#!/usr/bin/env node
+import { buildHelpText, parseCliArgs } from "./lib/config.js";
+import { runManualRecord } from "./lib/commands/manual-record.js";
+import { runAutoBasicCapture } from "./lib/commands/auto-basic.js";
+import { runSampleExtraNodes } from "./lib/commands/sample-extra-nodes.js";
+import { runSampleBranchEdges } from "./lib/commands/sample-branch-edges.js";
+import { runSampleTrialInputs } from "./lib/commands/sample-trial-inputs.js";
+import { runDoctor } from "./lib/commands/doctor.js";
+import { runConfigure } from "./lib/commands/configure.js";
+import { runPublish } from "./lib/commands/publish.js";
+import { runListLibraries } from "./lib/commands/list-libraries.js";
+
+const COMMANDS = {
+  doctor: runDoctor,
+  manual: runManualRecord,
+  "auto-basic": runAutoBasicCapture,
+  "sample-extra-nodes": runSampleExtraNodes,
+  "sample-branch-edges": runSampleBranchEdges,
+  "sample-trial-inputs": runSampleTrialInputs,
+  configure: runConfigure,
+  publish: runPublish,
+  "list-libraries": runListLibraries,
+};
+
+const cleanupCallbacks = [];
+
+export function registerCleanup(fn) {
+  cleanupCallbacks.push(fn);
+}
+
+async function runCleanup() {
+  for (const fn of cleanupCallbacks) {
+    try {
+      await fn();
+    } catch {
+      // Best-effort cleanup
+    }
+  }
+  cleanupCallbacks.length = 0;
+}
+
+let shuttingDown = false;
+process.on("SIGINT", async () => {
+  if (shuttingDown) {
+    process.exit(1);
+  }
+  shuttingDown = true;
+  console.error("\nReceived SIGINT, cleaning up...");
+  await runCleanup();
+  process.exit(130);
+});
+
+process.on("SIGTERM", async () => {
+  console.error("\nReceived SIGTERM, cleaning up...");
+  await runCleanup();
+  process.exit(143);
+});
+
+async function main() {
+  const config = parseCliArgs(process.argv.slice(2));
+
+  if (config.help) {
+    console.log(buildHelpText());
+    return;
+  }
+
+  const handler = COMMANDS[config.command];
+  if (!handler) {
+    throw new Error(`Unknown command: ${config.command}`);
+  }
+
+  await handler(config);
+}
+
+main().catch((error) => {
+  console.error(error.stack || error.message);
+  process.exitCode = 1;
+});

package/src/lib/analyze.js

@@ -0,0 +1,213 @@
+import path from "node:path";
+import fs from "node:fs/promises";
+import { safeJsonParse, writeText } from "./util.js";
+
+const KEYWORDS = [
+  "workflow",
+  "graph",
+  "node",
+  "edge",
+  "save",
+  "run",
+  "debug",
+  "intent",
+  "llm",
+  "model",
+  "dataCenter",
+  "publish",
+];
+
+function summarizeStream(request) {
+  const headerEntries = Object.entries(request.responseHeaders ?? {}).map(
+    ([key, value]) => [String(key).toLowerCase(), value],
+  );
+  const contentType = headerEntries.find(([key]) => key === "content-type")?.[1];
+
+  if (String(contentType).includes("text/event-stream")) {
+    return "SSE";
+  }
+
+  if (
+    request.protocol?.toLowerCase().includes("http") &&
+    typeof request.responseBody === "string" &&
+    request.responseBody.includes("\n\n")
+  ) {
+    return "Streaming HTTP chunks";
+  }
+
+  if (request.method === "WEBSOCKET") {
+    return "WebSocket";
+  }
+
+  return null;
+}
+
+function unwrapSchemaContainer(payload) {
+  if (!payload || typeof payload !== "object") {
+    return null;
+  }
+
+  const candidates = [
+    payload.params?.schemaJson,
+    payload.schemaJson,
+    payload.result?.schemaJson,
+    payload.result?.data?.schemaJson,
+  ];
+
+  for (const candidate of candidates) {
+    const parsed = safeJsonParse(candidate);
+    if (parsed && typeof parsed === "object") {
+      return parsed;
+    }
+  }
+
+  if (Array.isArray(payload.nodes) || Array.isArray(payload.edges)) {
+    return payload;
+  }
+
+  if (payload.graph && typeof payload.graph === "object") {
+    return payload.graph;
+  }
+
+  if (payload.workflowGraph && typeof payload.workflowGraph === "object") {
+    return payload.workflowGraph;
+  }
+
+  if (payload.data && typeof payload.data === "object") {
+    return unwrapSchemaContainer(payload.data);
+  }
+
+  return null;
+}
+
+function typeName(value) {
+  if (Array.isArray(value)) {
+    return "array";
+  }
+
+  if (value === null) {
+    return "null";
+  }
+
+  return typeof value;
+}
+
+function summarizeFieldTypes(record) {
+  if (!record || typeof record !== "object" || Array.isArray(record)) {
+    return {};
+  }
+
+  return Object.fromEntries(
+    Object.entries(record).map(([key, value]) => [key, typeName(value)]),
+  );
+}
+
+export function summarizeWorkflowSchema(request) {
+  const payload =
+    safeJsonParse(request.requestBody) ??
+    request.responseBodyJson ??
+    null;
+
+  if (!payload || typeof payload !== "object") {
+    return null;
+  }
+
+  const graph = unwrapSchemaContainer(payload) ?? payload.canvas ?? payload;
+
+  const nodes = Array.isArray(graph.nodes)
+    ? graph.nodes.map((node) => ({
+        id: typeof node.id,
+        type: typeof node.type,
+        position: typeof (node.position ?? node.meta?.position),
+        data: typeof node.data,
+      }))
+    : [];
+
+  const edges = Array.isArray(graph.edges)
+    ? graph.edges.map((edge) => summarizeFieldTypes(edge))
+    : [];
+
+  return {
+    workflowId:
+      payload.workflowId ??
+      payload.params?.workflowId ??
+      payload.id ??
+      payload.data?.workflowId ??
+      null,
+    nodeCount: nodes.length,
+    edgeCount: edges.length,
+    nodeTypes: [...new Set((graph.nodes ?? []).map((node) => node.type).filter(Boolean))],
+    nodes: nodes.slice(0, 10),
+    edges: edges.slice(0, 10),
+    edgeSample: Array.isArray(graph.edges) ? graph.edges[0] ?? null : null,
+    streamType: summarizeStream(request),
+  };
+}
+
+export function buildStaticAnalysis(stepNetworks) {
+  const findings = [];
+
+  for (const step of stepNetworks) {
+    for (const request of step.network) {
+      const url = request.url ?? "";
+      const body = `${request.requestBody ?? ""}\n${request.responseBody ?? ""}`;
+      const matches = KEYWORDS.filter(
+        (keyword) => url.includes(keyword) || body.includes(keyword),
+      );
+
+      if (matches.length === 0) {
+        continue;
+      }
+
+      findings.push({
+        stepId: step.step.id,
+        stepTitle: step.step.title,
+        method: request.method,
+        url,
+        matches,
+      });
+    }
+  }
+
+  return findings;
+}
+
+export async function writeStaticAnalysis(outputDir, stepNetworks) {
+  const findings = buildStaticAnalysis(stepNetworks);
+  const lines = ["# Static Analysis", ""];
+
+  if (findings.length === 0) {
+    lines.push("No keyword hits were found in captured request/response bodies.");
+  } else {
+    for (const finding of findings) {
+      lines.push(`## ${finding.stepTitle}`);
+      lines.push(`- Method: ${finding.method}`);
+      lines.push(`- URL: ${finding.url}`);
+      lines.push(`- Keyword hits: ${finding.matches.join(", ")}`);
+      lines.push("");
+    }
+  }
+
+  const bundleDir = path.join(outputDir, "bundles");
+
+  try {
+    const bundleFiles = await fs.readdir(bundleDir);
+    if (bundleFiles.length > 0) {
+      lines.push("## Saved Bundles");
+      for (const bundleFile of bundleFiles) {
+        const bundlePath = path.join(bundleDir, bundleFile);
+        const content = await fs.readFile(bundlePath, "utf8");
+        const matches = KEYWORDS.filter((keyword) => content.includes(keyword));
+        if (matches.length === 0) {
+          continue;
+        }
+
+        lines.push(`- ${bundleFile}: ${matches.join(", ")}`);
+      }
+    }
+  } catch {
+    // No bundles saved.
+  }
+
+  await writeText(path.join(outputDir, "static-analysis.md"), `${lines.join("\n")}\n`);
+}