AI Website Workbench
Your AI tool is ready (any tool on the ZCode main path works) → this guide hands the B2B Export AI Workbench to it and walks the site-building workflow from customer files to a live website. For SEO and GEO methodology, see the other two pages in this section — pointers at the end.
🌟 Goal: hand the AI website workbench — a repository packed with site-building methods — to your AI, give it customer files and an AllinCMS account, and run the workflow from files to a live site. The workbench's own words: the AI works by the method, you nod, you verify afterwards — it is not one-click generation.
Meet the Workbench
What's inside is not software but methods and tools: how to build a website, write B2B articles, send cold outreach, do SEO/GEO, handle inquiries — all organized into steps both humans and AI can read. Core modules:
| What you want | Where in the workbench | Maturity (per the repo) |
|---|---|---|
| Build a B2B site from files (AllinCMS, API-driven end to end) | website-content-ops toolkit | Most mature, with real build evidence |
| Update products / articles on an existing site | Same → content update flow | Fairly mature, written after review |
| Turn company material into English B2B articles | B2B article standard | Method mature, results pending real data |
| Channel playbooks (cold email / LinkedIn / SEO / GEO / trade fairs…) | Playbooks index | Mostly usable methods, partly pending validation |
💡 Version caliber: mother repo
v0.3.4(Stable); the site-building sub-library is0.4.0-preview(not Stable — production actions need step-by-step approval). The repository's current README wins.
Quick Start: One Prompt
You need only two things: an AI that can read and write local files (ZCode, Claude Code, Codex all work) and an AllinCMS account. Send the whole block below to the AI and fill in what you want at the end:
【B2B 外贸工作台 · 一键启动】把本块整体复制发给你的 AI。
■ 仓库：https://github.com/tony-apan/b2b-export-ai-workbench-source.git
■ 第一步：clone 后安装建站工具包并自测：
python3 sub-libraries/website-content-ops/SKILL-INSTALL/install.py
（Windows 用 install.cmd）缺 Node/Python 给我可复制命令，不要跳过。
■ 第二步：我要做的事是：______（例：建网站 / 写 B2B 文章 / 外贸增长建议）
■ 规则：建站类读 sub-libraries/website-content-ops/，方法类先读 wiki/index.md；
删除/覆盖/发布先列清单等我逐条批准；凭据只进环境变量；情况不明先问不要猜。
The AI will clone the repository, install the site-building toolkit, and self-check. If Node or Python is missing, it will give you copy-paste install commands — follow them; you do not need to know how to program.
Building: What You Give, What You Approve
After installation, just describe what you need in plain language. Using "build a new B2B site" as the example:
You give the AI three things:
- Customer files: PDF, DOCX, spreadsheets, an existing website, product photos all work;
- An AllinCMS account: site building goes through AllinCMS's API pipeline, so you need an account;
- One sentence of intent: e.g. "build a website for North American wholesalers using these files".
The AI does the work; you nod at three checkpoints:
- ✅ Before delete / overwrite / publish: the AI must list the operations and wait for your item-by-item approval;
- ✅ Credentials go only into environment variables: your AllinCMS account details never enter chat history or the repository;
- ✅ Verify afterwards: going live is not the end — check pages, content, and forms against its checklist.
Three Safety Rules
The workbench repository is public, but three lines must not be crossed:
- Real customer files, accounts, and passwords never enter the repository — files and run data stay in a private folder on your computer, handled by you and the AI;
- Credentials go only into environment variables — never into chat history or config files;
- Delete, overwrite, and publish require a listed plan and your approval — this rule ships inside the start prompt; do not tell the AI to "skip confirmations".
SEO / GEO: Where and How
A live site is only the start. The workbench ships channel playbooks for SEO / GEO (Playbooks index → Channels) — just tell the AI "optimize this site using the workbench's SEO playbook".
This site's companion methodology lives in this same section: B2B Site SEO from Zero is the full execution roadmap, and Out of Ideas? Ask AI solves "what to write". The workbench's methods plus those two pages form one complete kit.
📋 Common Pitfalls
| Pitfall | Consequence | How to avoid |
|---|---|---|
| 🔴 Putting customer files or account passwords into the repository | Public leak of private data; irreversible | Files stay in a local private folder; credentials only in environment variables |
| 🔴 Telling the AI to "skip confirmations and publish" | Wrong content goes straight live | Keep item-by-item approval; check the operation list before publishing |
| 🟡 Treating the workbench as "one-click generation" | Misaligned expectations — you still have to supervise | It is "AI works by the method + you nod + you verify" |
| 🟡 Treating the preview sub-library as stable | Production actions break with no safety net | Sub-library 0.4.0-preview needs step-by-step approval; the repository README wins |
| 🟢 Assuming "a method exists" means "results are proven" | Overestimated effect, disappointing rollout | The repo labels maturity per module — set expectations by those labels |
✅ Completion Check
- The toolkit installed and self-checked without errors (install.py ran clean);
- You handed over customer files and the AllinCMS account (credentials only in environment variables);
- The site was generated, and you approved each upload / publish step by step;
- You verified pages, content, and the inquiry form against the checklist;
- You know the next step: start content and optimization with the two SEO pages.
📚 Further Reading
- 📚 B2B Site SEO from Zero: the full SEO execution roadmap after launch
- 📚 Out of Ideas? Ask AI: topic research methods for your content
- 📚 Laifaxin AI outreach overview: with the site ready, start proactive outreach
With the site live, use 📄 B2B Site SEO from Zero to start content and optimization; to go straight to proactive outreach, see 📄 Laifaxin AI outreach overview.