本地优先的跨平台内容发布工作台 —— 一次创作，全网分发。
面向内容创作者、自媒体运营和工具链验证场景。输入一份 Markdown 源内容，自动生成适配微信公众号、知乎、B站、小红书四个平台的版本，支持编辑、校验、模拟/草稿/辅助发布，全程不依赖云端服务。
- 界面与文档默认使用中文；英文内容仅保留在独立英文文档区、代码标识、文件格式、接口路径和第三方产品名中。
- 用户可见的提示、按钮、表单、状态、错误和导出说明使用中文。
Markdown 输入 → CanonicalPost 统一模型 → 4 平台适配生成 → 预览/编辑/确认 → 发布前校验 → 发布（simulate / draft / assist）
| 模块 | 能力 |
|---|---|
| 输入 | Markdown / 富文本 / 纯文本三种模式，图片拖拽上传 |
| 统一模型 | CanonicalPost 承载 7 种 Block 类型（段落、标题、图片、引用、代码、列表、分割线） |
| 适配生成 | 插件化 PlatformAdapter + Registry 注册机制，新增平台不入侵核心流程 |
| 平台预览 | 四平台独立预览标签页，实时切换查看 |
| 编辑确认 | 分平台编辑标题、正文、标签和元数据，AI 生成内容须人工确认 |
| 发布校验 | 通用校验 + 平台专项 MVP 检查，含错误码和字段级提示 |
| 发布 | simulate（模拟）、draft（草稿）、assist（辅助，manual-only）、publish（真实发布，预留） |
| 日志 | SQLite 持久化，Posts · Drafts · Jobs · Logs · Assets 可追溯 |
| 验收 | npm run test:acceptance 一键跑通 10+ 步闭环验证 |
| 平台 | 适配器 | 默认模式 | 适配产物 |
|---|---|---|---|
| 公众号 | wechat.ts |
draft |
长文草稿 + 封面提示 |
| 知乎 | zhihuAssist.ts |
assist |
问答风格 + 话题标签 + 逻辑提示 |
| B站 | bilibili.ts |
simulate |
专栏/视频标题 + 分区建议 + 置顶评论 |
| 小红书 | xhsAssist.ts |
assist |
笔记正文 + 话题标签 + 封面文字 + 卡片文案 |
| Mock | mock.ts |
simulate |
完整模拟链路（开发测试用） |
- 默认不做真实发布
- AI 生成内容须人工确认后才能执行非 simulate 模式
- 辅助发布（assist）最终发布操作标记为
manual-only - 不绕过登录、验证码或平台风控机制
publish模式须二次确认
- Node.js >= 24
- npm >= 11
npm install # 安装依赖
npm run dev # 一键启动 API (:3333) + 桌面工作台 (:5173)打开页面后点击「一键跑通本地闭环」或「载入示例内容」即可开始体验。
| 命令 | 说明 |
|---|---|
npm run dev |
一键启动 API + 桌面工作台 |
npm run dev:api |
单独启动 API 服务 |
npm run dev:desktop |
单独启动桌面工作台 |
npm run build |
构建全部 workspace |
npm run typecheck |
类型检查 |
npm run test:acceptance |
运行自动化验收测试 |
flash-promoter/
├── apps/
│ ├── desktop/ # React 桌面工作台 (Vite)
│ └── local-api/ # Fastify API 服务
├── packages/
│ ├── core/ # 核心模型 / 适配器 / 校验 / 生成
│ └── storage/ # SQLite 持久化层
├── docs/ # 文档 / 动态横幅 SVG
├── scripts/ # 自动化验收脚本
├── data/ # 本地 SQLite 数据库
└── flash-promoter_prd.md # 完整 PRD
会真的发布到公众号/知乎/B站吗？
当前 MVP 阶段不会执行真实发布。所有操作在本地闭环完成。publish 模式预留了接口但需二次确认才会开通。
和其他发布工具有什么区别？
本地优先 + 插件化适配架构。你完全控制内容所在的设备，不依赖云服务。平台适配器以插件形式注册，新增平台不入侵核心流程。
需要平台账号吗？
模拟发布（simulate）不需要任何账号。草稿（draft）和辅助（assist）模式在 MVP 阶段也是全模拟的，不调用真实 API。
- 公众号真实草稿 API 接入
- B站真实投稿参数 API
- 小红书封面图/卡片图导出
- Playwright 浏览器自动化辅助发布
- Electron / Tauri 桌面打包
- 本地加密凭据存储
- 团队协作 & 排期发布
MIT
Local-first multi-platform content publishing workbench — write once, publish everywhere.
Built for content creators, social media operators, and toolchain validators. Input one Markdown source and auto-generate platform-adapted versions for WeChat Official Account, Zhihu, Bilibili, and Xiaohongshu — with editing, validation, and simulated/draft/assist publishing — all local, zero cloud dependency.
- UI and documentation default to Chinese; English content is kept only in the standalone English section, code identifiers, file formats, API paths, and third-party product names.
- User-facing prompts, buttons, forms, status messages, and errors use Chinese.
Markdown Input → CanonicalPost (Unified Model) → 4-Platform Adaptation → Preview/Edit/Confirm → Validation → Publish (simulate / draft / assist)
| Module | Capabilities |
|---|---|
| Input | Markdown, rich text, and plain text modes; drag-and-drop image upload |
| Unified Model | CanonicalPost with 7 block types (paragraph, heading, image, quote, code, list, divider) |
| Adaptation | Plugin-based PlatformAdapter + Registry — add platforms without touching core logic |
| Preview | Per-platform preview tabs with real-time switching |
| Edit & Confirm | Per-platform editing of title, body, tags, and metadata; AI-generated content requires human confirmation |
| Validation | Common + platform-specific checks, error codes, and field-level hints |
| Publishing | simulate · draft · assist (manual-only) · publish (reserved, gated) |
| Logging | SQLite-backed: Posts · Drafts · Jobs · Logs · Assets |
| Acceptance | npm run test:acceptance — 10+ E2E verification steps in one command |
| Platform | Adapter | Default Mode | Output |
|---|---|---|---|
wechat.ts |
draft |
Long-form draft + cover prompt | |
| Zhihu | zhihuAssist.ts |
assist |
Q&A style + topics + logic hints |
| Bilibili | bilibili.ts |
simulate |
Article/video title + partition + pinned comment |
| Xiaohongshu | xhsAssist.ts |
assist |
Note body + hashtags + cover text + card copy |
| Mock | mock.ts |
simulate |
Full simulation pipeline (dev & test) |
- No real publishing by default
- AI-generated content blocked from non-simulate modes until confirmed
- Assist packages marked
manual-only - No bypass of login, captcha, or platform risk controls
publishmode requires second confirmation
- Node.js >= 24
- npm >= 11
npm install # Install dependencies
npm run dev # Start API (:3333) + desktop (:5173) in one commandClick 「一键跑通本地闭环」 or 「载入示例内容」 to get started.
| Command | Description |
|---|---|
npm run dev |
Start API + desktop together |
npm run dev:api |
Start API server only |
npm run dev:desktop |
Start desktop workbench only |
npm run build |
Build all workspaces |
npm run typecheck |
Type check |
npm run test:acceptance |
Run acceptance tests |
flash-promoter/
├── apps/
│ ├── desktop/ # React desktop workbench (Vite)
│ └── local-api/ # Fastify API server
├── packages/
│ ├── core/ # Core models / adapters / validation / generation
│ └── storage/ # SQLite persistence layer
├── docs/ # Documentation / animated hero SVG
├── scripts/ # Acceptance test script
├── data/ # Local SQLite database
└── flash-promoter_prd.md # Full PRD
Does it really publish to WeChat / Zhihu / Bilibili?
No. The current MVP does not perform real publishing. All operations complete locally. The publish mode is reserved and requires a second confirmation to activate.
How is this different from other publishing tools?
Local-first + plugin architecture. You control where your content lives — no cloud dependency. Platform adapters are registered as plugins; adding a platform never touches core logic.
Do I need platform accounts?
Simulate mode requires no accounts. Draft and assist modes in MVP are fully simulated — no real API calls.
- Real WeChat draft API
- Real Bilibili submission API
- Xiaohongshu image/card export
- Playwright browser automation
- Electron / Tauri packaging
- Encrypted credential vault
- Team workflow & scheduling
MIT