An agent-agnostic workflow kit for AI-powered Douyin (抖音) image-note auto-publishing. Turn the full "hotspot → note → images → publish → log" pipeline into a reusable, industry-configurable workflow that runs inside OpenClaw, Hermes, Codex, or any agent system that can browse the Douyin Creator Center and call local scripts.
It is not another one-off upload script. It is a workflow engineering kit: every stage is a prompt/contract/schema, so you can swap the model, the agent system, the image generator, or the publisher independently.
📄 中文摘要：一套不绑定特定智能体与模型的抖音图文自动发布工作流——读取实时热点、按行业画像生成图文、AI 直出带字主封面 + Pillow 渲染支撑卡片（统一 3:4 竖版 1080×1440）、发布时重搜热点校验、选不上即停发、BGM 验证歌名+时长、账号凭证零落盘。内置 AI 科普与本地生活餐饮两类行业示例，支持 OpenClaw / Hermes 等 Agent 系统一键迁移部署。
📖 完整中文版请见 README.zh-CN.md。
Most Douyin automation repos on GitHub are single-purpose upload scripts (PyAutoGUI / Playwright) that break when the page changes. This kit inverts the problem:
| Typical upload script | This kit |
|---|---|
| Hard-coded flow for one account | Industry profile JSON generalizes to any niche |
| Bound to one agent / model | Prompt + contract per stage, works with OpenClaw / Hermes / Codex |
| Publishes whatever you give it | Validates assets before publishing; aborts if the hotspot can't be selected |
| Stores cookies/tokens in config | Zero credential storage — manual login, publisher owns its session |
| No image QA | Enforces 3:4 vertical (1080×1440), bans blurred-border/stretched covers |
| Logs in chat | Structured per-post CSV logs, failure screenshots |
Five specialized agents cover eight pipeline stages:
flowchart TD
P[Industry Profile JSON<br/>audience / positioning / keywords] --> S1
S1["① collect hotspots<br/>(hotspot picker)"] --> S2["② select hotspot<br/>(hotspot picker)"]
S2 --> S3["③ write note<br/>(content writer)"]
S3 --> S4["④ build image prompts<br/>(image prompt builder)"]
S4 --> S5["⑤ generate images<br/>AI cover + Pillow support cards"]
S5 --> S6["⑥ validate assets<br/>(compliance checker)"]
S6 --> S7["⑦ publish<br/>(publisher agent, re-search hotspot)"]
S7 --> S8["⑧ record result<br/>CSV log + failure screenshot"]
Key reliability rules baked into the pipeline:
- Hotspot re-verification — the hotspot is re-searched on the live publish page before selecting; if it is gone, publishing stops instead of forcing it.
- BGM verification — a music track counts as selected only when the page shows a title and a duration.
- Credentials — no passwords, cookies, QR-login data, or browser profiles ever live in the kit.
- Image QA — 3:4 vertical enforced, cover text baked in by the image model, support cards rendered deterministically with Pillow.
douyin-agent-workflow-kit/
├── workflow-kit/ # Generic workflow package (docs in Chinese)
│ ├── agent_specs/ # Task specs for OpenClaw / Hermes / generic agents
│ ├── adapters/ # Publisher adapter contracts (social-auto-upload)
│ ├── configs/ # Config examples (account / publisher / workflow)
│ ├── prompts/ # Reusable per-stage prompts (5 agents)
│ ├── templates/project/ # Scaffolded project layout (assets, drafts, logs…)
│ ├── tools/ # init / validate / demo-images / publish scripts
│ └── workflow/ # Stage definitions & I/O contracts
│
└── industry-skill/ # Industry-configurable agent skill (docs in English)
├── SKILL.md # Main skill: safety rules, workflow, rules
├── agents/openai.yaml # Agent manifest for OpenAI-compatible setups
├── examples/ # 2 ready-made industry profiles (AI science, restaurant)
├── references/ # content / image / publish / industry-config rules
└── scripts/ # batch planning, support-card rendering, publish wrapper
# Windows
python workflow-kit/tools/init_project.py --target C:\douyin-ai-project --account-alias my-alias --display-name "My Douyin Name"
# macOS / Linux
python3 workflow-kit/tools/init_project.py --target ~/douyin-ai-project --account-alias my-alias --display-name "My Douyin Name"Then validate a draft and publish one row:
python workflow-kit/tools/check_note_ready.py --project . --id demo
python workflow-kit/tools/publish_with_sau.py --project . --id demo --social-root /path/to/social-auto-uploadSee workflow-kit/INSTALL.md for the full walkthrough.
- Copy
industry-skill/into your agent's skills directory (or package it however your system loads skills). - Provide an industry profile JSON (start from
industry-skill/examples/). - Ask the agent to run the Douyin publishing workflow for your niche.
The full spec is in industry-skill/SKILL.md.
Publishing is delegated to a publisher through a small contract (workflow-kit/adapters/social-auto-upload/), so you can use any automation tool. The reference implementation targets dreammis/social-auto-upload (a 9k+ star multi-platform uploader) via its CLI.
| Project | What it is |
|---|---|
| dreammis/social-auto-upload | The uploader itself — this kit is a workflow layer on top of it |
| withwz/douyin_upload | PyAutoGUI upload script |
Various douyin-auto-publish repos |
Single-purpose scripts, no workflow/industry abstraction |
This kit's differentiator: the publishing decision chain (hotspot re-search, BGM verification, abort conditions) and the industry-profile abstraction make it usable as a genuine multi-agent workflow, not just a form filler.
- You will comply with the platform's terms and applicable laws; use at your own risk.
- Any account restriction, ban, or penalty is solely your responsibility.
- Keep publishing frequency low; this kit intentionally aborts rather than forces publishing when conditions are not met.
The authors are not liable for any consequences of using this software.
MIT © 2026 Finn763 — see LICENSE.