autocrew 0.1.0

package/README.md

@@ -0,0 +1,190 @@
+# AutoCrew
+
+**One-person content studio powered by AI — from trending topics to published posts.**
+
+English | [中文](README_CN.md)
+
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE)
+[![npm version](https://img.shields.io/npm/v/autocrew.svg)](https://www.npmjs.com/package/autocrew)
+[![Node.js](https://img.shields.io/badge/node-%3E%3D18-brightgreen.svg)](https://nodejs.org)
+
+---
+
+AutoCrew is an AI-powered content operations pipeline for Chinese social media. It handles the entire workflow — research trending topics, write platform-native drafts, review for sensitivity and AI-sounding language, and publish to Xiaohongshu, Douyin, WeChat, and more.
+
+Works standalone as a CLI, or as a plugin for [Claude Code](https://claude.ai/code) and [OpenClaw](https://openclaw.dev).
+
+```
+You:       Find me topics for this week
+AutoCrew:  (searches trending lists + scores by your niche → recommends 5 topics)
+
+You:       Write #2 as a Xiaohongshu post
+AutoCrew:  (generates draft in your voice → checks for sensitive words → outputs title options + hashtags)
+
+You:       Too AI-sounding, make it more conversational
+AutoCrew:  (de-AI pass → remembers your preference → applies automatically next time)
+
+You:       Publish
+AutoCrew:  (6-point pre-publish check → publishes)
+```
+
+---
+
+## Quick Start
+
+```bash
+npm install -g autocrew
+autocrew init
+autocrew research
+```
+
+That's it. `init` creates your local data directory (`~/.autocrew/`), and `research` finds trending topics based on public sources (Zhihu, Weibo, industry keywords).
+
+---
+
+## Features
+
+### Research
+- Trending topic discovery from public sources (Zhihu Hot List, Weibo Trending, industry keywords)
+- Topic scoring based on your niche and audience
+- Intel pipeline for monitoring inspiration sources
+
+### Write
+- Style-calibrated drafts — 4-stage deep calibration learns your voice
+- Hook-Body-CTA-Title structure optimized for Chinese social media
+- Multi-platform rewrites (same topic, platform-native format)
+
+### Review
+- Sensitive word scanning (built-in + platform-specific throttled terms)
+- De-AI-ification (去 AI 味) — removes robotic patterns from Chinese text
+- Quality scoring: information density, hook strength, CTA clarity
+
+### Publish
+- Multi-platform: Xiaohongshu, Douyin, WeChat Video, WeChat Articles, Bilibili
+- 6-point pre-publish checklist
+- Cover image generation and A/B/C review
+
+### Learn
+- Auto-captures your edits as feedback
+- After 5 similar corrections, distills them into writing rules
+- Rules are applied automatically in future drafts
+
+### Web Dashboard
+- Built-in web UI for visual content management
+- Script editor with timeline-based video production
+- Jianying (剪映) export for video editing
+
+---
+
+## How It Works
+
+AutoCrew runs entirely on your machine. All content, drafts, and creator profiles are stored locally in `~/.autocrew/`. No data is uploaded anywhere.
+
+The core is a standalone Node.js CLI with 20+ tools and 15+ skills. A thin plugin layer lets it work as a Claude Code MCP server or OpenClaw extension — but it doesn't depend on either.
+
+```
+~/.autocrew/
+├── creator-profile.json    # Your niche, audience, writing rules
+├── STYLE.md                # Writing voice profile
+├── topics/                 # Topic library
+├── contents/               # Content projects
+│   └── content-xxx/
+│       ├── meta.json       # Metadata + status + asset index
+│       ├── draft.md        # Current draft
+│       ├── timeline.json   # Video timeline (if generated)
+│       └── versions/       # Version history
+└── learnings/              # Edit history + distilled rules
+```
+
+---
+
+## Use with Claude Code
+
+Add AutoCrew as an MCP server in your Claude Code settings:
+
+```json
+{
+  "mcpServers": {
+    "autocrew": {
+      "command": "npx",
+      "args": ["tsx", "/path/to/AutoCrew/mcp/server.ts"]
+    }
+  }
+}
+```
+
+Then just talk to Claude naturally — "find me topics", "write this as a post", "make it less AI-sounding".
+
+## Use with OpenClaw
+
+```bash
+openclaw plugins install autocrew
+```
+
+Works out of the box. All skills and tools are available in OpenClaw conversations.
+
+---
+
+## CLI Reference
+
+| Command | Description |
+|---------|-------------|
+| `autocrew init` | Initialize data directory and creator profile |
+| `autocrew status` | Pipeline overview |
+| `autocrew research` | Discover trending topics |
+| `autocrew topics` | List saved topics |
+| `autocrew start <topic>` | Start a new content project |
+| `autocrew contents` | List content items |
+| `autocrew humanize <id>` | De-AI pass on a draft |
+| `autocrew adapt <id> <platform>` | Platform-native rewrite |
+| `autocrew review <id>` | Content review (sensitivity + quality) |
+| `autocrew pre-publish <id>` | 6-point pre-publish checklist |
+| `autocrew profile` | Show creator profile |
+| `autocrew learn <id> --signal edit --feedback "too formal"` | Capture feedback |
+| `autocrew intel` | Manage inspiration sources |
+
+Run `autocrew --help` for the full command list.
+
+---
+
+## Free vs Pro
+
+| Feature | Free | Pro |
+|---------|------|-----|
+| Topic research | Public search (Zhihu/Weibo) | + Competitor crawling + platform hot lists |
+| Content writing | Full workflow | Full workflow |
+| De-AI-ification | Yes | Yes |
+| Sensitive word scan | Yes | Yes |
+| Platform rewrite | Xiaohongshu + Douyin | + WeChat + Bilibili |
+| Cover generation | 3:4 | + 16:9 + 4:3 |
+| Publishing | Xiaohongshu + Douyin | All platforms |
+| Style calibration | Yes | Yes |
+| Learning loop | Yes | Yes |
+| Competitor monitoring | No | Yes |
+| Video ASR extraction | No | Yes |
+| Analytics reports | No | Yes |
+
+---
+
+## Development
+
+```bash
+git clone https://github.com/alextangson/AutoCrew.git
+cd AutoCrew
+npm install
+npm test        # 341 tests
+```
+
+Start the web dashboard in dev mode:
+
+```bash
+cd web && npm run dev
+```
+
+See [CONTRIBUTING.md](CONTRIBUTING.md) for guidelines.
+
+---
+
+## License
+
+MIT &copy; [alextangson](https://github.com/alextangson)

package/README_CN.md

@@ -0,0 +1,190 @@
+# AutoCrew
+
+**一个人的 AI 内容工作室 -- 从选题到发布，全流程自动化。**
+
+[English](README.md) | 中文
+
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE)
+[![npm version](https://img.shields.io/npm/v/autocrew.svg)](https://www.npmjs.com/package/autocrew)
+[![Node.js](https://img.shields.io/badge/node-%3E%3D18-brightgreen.svg)](https://nodejs.org)
+
+---
+
+AutoCrew 是为中文新媒体创作者打造的 AI 内容运营工具。覆盖选题调研、文案撰写、敏感词审核、去 AI 味、多平台改写、一键发布的完整流程。
+
+支持独立 CLI 使用，也可作为 [Claude Code](https://claude.ai/code) 或 [OpenClaw](https://openclaw.dev) 的插件。
+
+```
+你：帮我找这周小红书的选题
+AutoCrew：（搜索热榜 + 结合你的风格 → 推荐 5 个带评分的选题）
+
+你：把第 2 个写成笔记
+AutoCrew：（按你的写作风格生成 → 自动检查敏感词 → 输出标题备选 + hashtag）
+
+你：太 AI 味了，改口语一点
+AutoCrew：（去 AI 味处理 → 记录你的偏好 → 下次自动应用）
+
+你：发到小红书
+AutoCrew：（发布前 6 项检查 → 自动发布）
+```
+
+---
+
+## 快速开始
+
+```bash
+npm install -g autocrew
+autocrew init
+autocrew research
+```
+
+`init` 创建本地数据目录（`~/.autocrew/`），`research` 从公开来源发现热门选题。
+
+---
+
+## 核心功能
+
+### 选题调研
+- 从公开来源发现热门选题（知乎热榜、微博热搜、行业关键词）
+- 基于你的领域和受众自动评分
+- 灵感源监控管道
+
+### 内容写作
+- 风格校准 -- 4 阶段深度校准，学习你的表达方式
+- Hook-Body-CTA-Title 结构，针对中文社媒优化
+- 多平台改写 -- 同一选题生成小红书版、抖音版，各自独立
+
+### 内容审核
+- 敏感词扫描（内置词库 + 平台特定限流词）
+- 去 AI 味（去除 AI 生成痕迹）
+- 质量评分：信息密度、hook 强度、CTA 清晰度
+
+### 发布
+- 支持平台：小红书、抖音、微信视频号、微信公众号、B站
+- 发布前 6 项自动检查
+- 封面生成 + A/B/C 候选审核
+
+### 学习循环
+- 自动记录你的每次修改
+- 同类修改累积 5 次后，自动提炼为写作规则
+- 下次写作时自动应用
+
+### Web 控制台
+- 内置 Web UI，可视化管理内容
+- 文案编辑器 + 时间轴式视频制作
+- 支持导出剪映工程文件
+
+---
+
+## 运行原理
+
+AutoCrew 完全运行在你的本地机器上。所有内容、草稿、创作者档案都存储在 `~/.autocrew/`，不会上传任何数据。
+
+核心是独立的 Node.js CLI，包含 20+ 工具和 15+ 技能。薄薄的插件层让它可以作为 Claude Code MCP 服务器或 OpenClaw 扩展使用 -- 但不依赖任何一个。
+
+```
+~/.autocrew/
+├── creator-profile.json    # 创作者档案（领域、受众、写作规则）
+├── STYLE.md                # 写作风格档案
+├── topics/                 # 选题库
+├── contents/               # 内容项目
+│   └── content-xxx/
+│       ├── meta.json       # 元数据 + 状态 + 素材索引
+│       ├── draft.md        # 当前草稿
+│       ├── timeline.json   # 视频时间轴
+│       └── versions/       # 版本历史
+└── learnings/              # 修改记录 + 提炼的写作规则
+```
+
+---
+
+## 搭配 Claude Code 使用
+
+在 Claude Code 设置中添加 MCP 服务器：
+
+```json
+{
+  "mcpServers": {
+    "autocrew": {
+      "command": "npx",
+      "args": ["tsx", "/path/to/AutoCrew/mcp/server.ts"]
+    }
+  }
+}
+```
+
+然后直接用自然语言对话 -- "帮我找选题"、"写成小红书笔记"、"去 AI 味"。
+
+## 搭配 OpenClaw 使用
+
+```bash
+openclaw plugins install autocrew
+```
+
+安装后直接在 OpenClaw 对话里使用，无需额外配置。
+
+---
+
+## CLI 命令
+
+| 命令 | 说明 |
+|------|------|
+| `autocrew init` | 初始化数据目录和创作者档案 |
+| `autocrew status` | 流水线概览 |
+| `autocrew research` | 发现热门选题 |
+| `autocrew topics` | 列出选题 |
+| `autocrew start <topic>` | 从选题开始创作 |
+| `autocrew contents` | 列出内容项目 |
+| `autocrew humanize <id>` | 去 AI 味 |
+| `autocrew adapt <id> <platform>` | 平台改写 |
+| `autocrew review <id>` | 内容审核（敏感词 + 质量） |
+| `autocrew pre-publish <id>` | 发布前 6 项检查 |
+| `autocrew profile` | 查看创作者档案 |
+| `autocrew learn <id> --signal edit --feedback "太正式了"` | 记录反馈 |
+| `autocrew intel` | 灵感源管理 |
+
+运行 `autocrew --help` 查看完整命令列表。
+
+---
+
+## Free vs Pro
+
+| 功能 | Free | Pro |
+|------|------|-----|
+| 选题调研 | 公开搜索（知乎/微博热榜） | + 对标账号爬取 + 平台热榜 |
+| 内容写作 | 完整工作流 | 完整工作流 |
+| 去 AI 味 | 支持 | 支持 |
+| 敏感词检测 | 支持 | 支持 |
+| 平台改写 | 小红书 + 抖音 | + 公众号 + B站 |
+| 封面生成 | 3:4 | + 16:9 + 4:3 |
+| 发布 | 小红书 + 抖音 | 全平台 |
+| 风格校准 | 支持 | 支持 |
+| 学习循环 | 支持 | 支持 |
+| 对标账号监控 | 不支持 | 支持 |
+| 视频 ASR 提取 | 不支持 | 支持 |
+| 数据分析报告 | 不支持 | 支持 |
+
+---
+
+## 开发
+
+```bash
+git clone https://github.com/alextangson/AutoCrew.git
+cd AutoCrew
+npm install
+npm test        # 341 个测试
+```
+
+启动 Web 控制台开发模式：
+
+```bash
+cd web && npm run dev
+```
+
+参见 [CONTRIBUTING.md](CONTRIBUTING.md) 了解贡献指南。
+
+---
+
+## License
+
+MIT &copy; [alextangson](https://github.com/alextangson)

package/adapters/openclaw/index.ts

@@ -0,0 +1,68 @@
+/**
+ * AutoCrew — thin OpenClaw plugin adapter.
+ *
+ * Bridges the standalone AutoCrew runtime to OpenClaw's plugin API.
+ * All logic lives in src/ — this file only does registration + system prompt injection.
+ */
+import { bootstrap } from "../../src/cli/bootstrap.js";
+import type { PluginConfig } from "../../src/runtime/context.js";
+
+const autocrewPlugin = {
+  id: "autocrew",
+  name: "AutoCrew",
+  description:
+    "AI content operations crew — automated research, writing, and publishing pipeline for Chinese social media.",
+  configSchema: {
+    type: "object" as const,
+    additionalProperties: false,
+    properties: {
+      data_dir: { type: "string" as const },
+      pro_api_key: { type: "string" as const },
+      pro_api_url: { type: "string" as const },
+      gateway_url: { type: "string" as const },
+      gemini_api_key: { type: "string" as const },
+      gemini_model: { type: "string" as const },
+    },
+  },
+
+  register(api: any, config?: PluginConfig) {
+    const { runner } = bootstrap(config);
+
+    // Bridge tools to OpenClaw API
+    for (const def of runner.getTools()) {
+      api.registerTool(
+        () => ({
+          name: def.name,
+          label: def.label,
+          description: def.description,
+          parameters: def.parameters,
+          async execute(_id: string, params: Record<string, unknown>) {
+            return runner.execute(def.name, params);
+          },
+        }),
+        { names: [def.name] },
+      );
+    }
+
+    // System prompt injection
+    if (typeof api.on === "function") {
+      api.on("before_prompt_build", (event: any) => {
+        event.appendSystemContext = `
+<autocrew_instructions>
+你现在已加载 AutoCrew 插件。在帮助用户创作内容时，遵循以下规则：
+
+1. **先拆解再执行**：收到复杂请求时（如"帮我写一篇小红书"），先调用 autocrew_pipeline templates 展示完整步骤，用户确认后再逐步执行。
+2. **文件优先**：所有内容产出必须通过 autocrew_content save 保存到 ~/.autocrew/，不要只输出到聊天里。用户需要在下次会话中找到之前的产出。
+3. **完成后汇报**：每步完成后说明做了什么、结果是什么、下一步是什么。不要只说"完成了"。
+4. **使用已有工具**：优先使用 autocrew_* 系列工具完成任务，而不是手动操作文件。工具链：研究(autocrew_research) → 创建选题(autocrew_topic) → 写稿(autocrew_content save) → 去AI化(autocrew_humanize) → 审核(autocrew_review) → 封面(autocrew_cover_review) → 预发布检查(autocrew_pre_publish) → 发布(autocrew_publish)。
+5. **风格校准**：写内容前检查 ~/.autocrew/STYLE.md，确保产出符合用户的写作风格。
+6. **标题硬限制**：所有选题标题必须 ≤20 个中文字符（含标点符号和 emoji）。这是小红书等平台的硬性限制。超过 20 字的标题必须重写，不能截断。生成选题时，加载 skills/title-craft 方法论来指导标题创作。
+7. **去AI味+审核自动执行**：写完内容后，自动运行 autocrew_humanize（去AI味）和 autocrew_review（审核），不要问用户"要不要去AI味"或"要不要审核"——直接做。这是质量底线，不是可选功能。
+8. **告知文件位置**：保存内容后，必须告诉用户文件保存在哪里以及如何打开。格式：「📄 内容已保存到：~/.autocrew/contents/{id}/draft.md」+ 打开方式。
+</autocrew_instructions>`;
+      });
+    }
+  },
+};
+
+export default autocrewPlugin;

package/bin/autocrew.mjs

@@ -0,0 +1,23 @@
+#!/usr/bin/env node
+
+/**
+ * AutoCrew CLI entry point.
+ * Uses tsx to run TypeScript directly — no build step needed.
+ */
+import { execFileSync } from "node:child_process";
+import { fileURLToPath } from "node:url";
+import path from "node:path";
+
+const __dirname = path.dirname(fileURLToPath(import.meta.url));
+const entry = path.join(__dirname, "autocrew.ts");
+
+try {
+  execFileSync(
+    path.join(__dirname, "..", "node_modules", ".bin", "tsx"),
+    [entry, ...process.argv.slice(2)],
+    { stdio: "inherit", cwd: path.join(__dirname, "..") }
+  );
+} catch (err) {
+  // tsx exits with the child's exit code — just forward it
+  process.exitCode = err.status ?? 1;
+}

package/bin/autocrew.ts

@@ -0,0 +1,13 @@
+#!/usr/bin/env node
+import { bootstrap } from "../src/cli/bootstrap.js";
+import { run } from "../src/cli/index.js";
+import { showBanner } from "../src/cli/banner.js";
+
+const { runner, ctx, eventBus } = bootstrap();
+
+// Show banner for interactive use (not for subcommands that pipe output)
+if (process.argv.length <= 2 && process.stdout.isTTY) {
+  showBanner("0.1.0");
+}
+
+await run(process.argv.slice(2), runner, ctx, eventBus);

package/openclaw.plugin.json

@@ -0,0 +1,36 @@
+{
+  "id": "autocrew",
+  "name": "AutoCrew",
+  "description": "AI content operations crew — research, write, and publish on autopilot. Free: full workflow for XHS + Douyin. Pro: deep crawling, competitor monitoring, analytics, TTS, digital human.",
+  "skills": ["./skills"],
+  "configSchema": {
+    "type": "object",
+    "additionalProperties": false,
+    "properties": {
+      "data_dir": {
+        "type": "string",
+        "description": "Local data directory (default: ~/.autocrew)"
+      },
+      "pro_api_key": {
+        "type": "string",
+        "description": "Pro API key for advanced features (deep crawling, competitor monitoring, analytics, TTS). Get one at autocrew.dev/activate"
+      },
+      "pro_api_url": {
+        "type": "string",
+        "description": "Pro API base URL (default: https://api.autocrew.dev)"
+      },
+      "gateway_url": {
+        "type": "string",
+        "description": "OpenClaw Gateway URL for Chrome Relay browser control (default: http://127.0.0.1:18789)"
+      },
+      "gemini_api_key": {
+        "type": "string",
+        "description": "Gemini API key for cover image generation. Free at aistudio.google.com/apikey"
+      },
+      "gemini_model": {
+        "type": "string",
+        "description": "Image generation model: 'auto' (default, tries gemini-native then imagen-4), 'gemini-native', or 'imagen-4'"
+      }
+    }
+  }
+}

package/package.json

@@ -0,0 +1,74 @@
+{
+  "name": "autocrew",
+  "version": "0.1.0",
+  "description": "One-person content studio powered by AI — from trending topics to published posts",
+  "type": "module",
+  "license": "MIT",
+  "author": "alextangson",
+  "repository": {
+    "type": "git",
+    "url": "https://github.com/alextangson/AutoCrew.git"
+  },
+  "homepage": "https://github.com/alextangson/AutoCrew#readme",
+  "bugs": {
+    "url": "https://github.com/alextangson/AutoCrew/issues"
+  },
+  "engines": {
+    "node": ">=18.0.0"
+  },
+  "workspaces": [
+    "packages/*"
+  ],
+  "bin": {
+    "autocrew": "./bin/autocrew.mjs"
+  },
+  "files": [
+    "bin/",
+    "src/",
+    "adapters/",
+    "skills/",
+    "templates/",
+    "LICENSE",
+    "README.md",
+    "README_CN.md",
+    "HAMLETDEER.md",
+    "openclaw.plugin.json"
+  ],
+  "keywords": [
+    "cli",
+    "ai-agent",
+    "content-creation",
+    "content-ops",
+    "automation",
+    "social-media",
+    "xiaohongshu",
+    "douyin",
+    "chinese-social-media",
+    "ai-writing",
+    "claude-code",
+    "mcp"
+  ],
+  "dependencies": {
+    "@hono/node-server": "^1.19.12",
+    "@sinclair/typebox": "0.34.48",
+    "gray-matter": "^4.0.3",
+    "hono": "^4.12.10",
+    "js-yaml": "^4.1.1",
+    "rss-parser": "^3.13.0",
+    "tsx": "^4.19.0"
+  },
+  "scripts": {
+    "test": "vitest run",
+    "test:watch": "vitest",
+    "prepublishOnly": "npm test"
+  },
+  "openclaw": {
+    "extensions": [
+      "./adapters/openclaw/index.ts"
+    ]
+  },
+  "devDependencies": {
+    "@types/js-yaml": "^4.0.9",
+    "vitest": "^4.1.2"
+  }
+}

package/skills/_writing-style/SKILL.md

@@ -0,0 +1,68 @@
+---
+name: _writing-style
+description: |
+  共享写作规范。被写作类 skill 引用，不直接由用户触发。
+---
+
+# 写作风格规范
+
+被 write-script、spawn-writer、remix-content 等写作 skill 加载为共享规范。
+
+## Hook（只写一个 — 选最强的角度）
+
+| 类型 | 什么时候用 |
+|------|-----------|
+| 痛点 | 受众有明显未解决的挫败感 |
+| 悬念 | 话题存在反直觉的事实或惊人数据 |
+| 理想态 | 话题卖一个令人向往的结果 |
+| 情感共鸣 | 触及身份认同、归属感或理想自我 |
+| 反差 | 普遍认知与现实之间有明显落差 |
+
+选命中最狠的那一个类型。写 1-3 句。
+
+**禁止**以下列方式开头：
+- "哈喽大家好"
+- "你有没有想过"
+- 任何泛泛的问候或提问
+
+## 正文结构
+
+正文由 **信息点链** 组成（5-8 个）。
+
+每个信息点遵循：**论点 → 为什么成立 → 具体例子或场景**。
+
+规则：
+- 每个信息点：80-150 字。正文总量：800-1500 字。
+- 信息点渐进排列 — 不要把最好的放在最前面。
+- 包含 1-2 个打破预期的转折。
+- 包含 1-2 个互动钩子（提问、"你猜怎么着"、评论区引导）。
+- **不写散文段落**。短句。一句一个意思。
+
+## 结尾 CTA
+
+- 1-2 句引导具体行动（收藏/评论/关注）。
+- 必须与内容价值挂钩 — "收藏这条，下次用得上" 好过 "觉得有用就点赞"。
+- 使用品牌 STYLE.md 中的语气。
+
+## 标题（只写一个 — 最好的那个）
+
+写标题前用 `web_search` 查目标平台 2-3 个热搜关键词。
+
+标题规则：
+- 自然嵌入 1 个热搜关键词。
+- 制造好奇缺口或承诺一个具体结果。
+- 15-25 字。可以加 emoji（如果确实加分的话）。
+- 不做标题党 — 内容必须兑现承诺。
+
+## 语气规则
+
+- **去 AI 味**：短句、口语化、像人说话。
+- **温暖非正式**：不端着，但有分寸。
+- **专家但接地气**：有干货，但不掉书袋。
+- **情绪优先于逻辑**：先打动，再说服。
+
+## 输出格式
+
+保存时：
+- `title`：唯一最优标题（title 字段不带 emoji）。
+- `body`：完整脚本纯文本。段落之间空行分隔。不使用 Markdown 标题。