A Claude Code skill that turns a one-line idea into a published, evidence-backed page. Recon, real lab experiments, verification loops, two human checkpoints. Free and MIT.
概览
A Claude Code skill that turns a one-line idea into a published, evidence-backed page. Not a prompt. A pipeline: it researches what already ranks, rents a real server, deploys and breaks the actual software, takes its own screenshots, writes in your voice, ties every factual claim to an evidence file, and stops at two fixed points for you to approve. This is the generic version of the system behind — same contract, same tools, same hard-won lessons, with the site-specific parts stripped out so you can point it at your own niche. Most "AI content" workflows are a prompt, a generate button, and a publish button. The model writes from its memory of the internet, which means it writes the average of what everyone else already said. That average is the slop. This bets the other way. So the system spends most of its time generating evidence and only then writes.
README
以下为原文 README 节选，完整内容请查看原文。
Content Factory — Starter Kit
A Claude Code skill that turns a one-line idea into a published, evidence-backed page.
Not a prompt. A pipeline: it researches what already ranks, rents a real server, deploys and breaks the actual software, takes its own screenshots, writes in your voice, ties every factual claim to an evidence file, and stops at two fixed points for you to approve.
This is the generic version of the system behind learnwithhasan.com/guide/ai-content-factory/ — same contract, same tools, same hard-won lessons, with the site-specific parts stripped out so you can point it at your own niche.
Free, MIT-licensed, no email required.
Why this exists
Most “AI content” workflows are a prompt, a generate button, and a publish button. The model writes from its memory of the internet, which means it writes the average of what everyone else already said. That average is the slop.
This bets the other way. The scarce thing is not text generation, it is first-hand evidence. So the system spends most of its time generating evidence and only then writes.
The result of one real run: a server rented for about 21 hours, 11 containers deployed, a
connection pooler deliberately broken and diagnosed, 49 mapped claims — and a total lab bill
of $0.76. That run is included in example/, unedited.
What is in the box
content-factory/ ← the skill. Copy this whole folder into .claude/skills/
├── SKILL.md the contract: stages, rules, and where it must stop for you
├── SETUP.md read this once before the first run
├── modules/ lab · recon · assets · verify · knowledge base
├── profiles/ how a verified draft becomes a page on YOUR site
├── templates/ PIECE.md · claims map · evidence manifest · recon report
├── memory/ 17 lessons from real runs — the part that compounds
└── lab/ the tools (Python, mostly stdlib)
example/ a real piece, start to finish, lightly scrubbed
docs/ QUICKSTART.md · ADAPT.md
The tools
| Tool | What it does | Needs |
|---|---|---|
lock.py |
Per-piece locks and ports so several pieces run in parallel without collisions | — |
ledger.py |
Renders LEDGER.md from every piece’s frontmatter, so nothing is ever hand-maintained |
— |
lint_draft.py |
Mechanical voice lint: the tells that survive a careful read | — |
capture.py |
Playwright screenshots to a fixed spec, including authenticated pages | Playwright |
auth.py |
Log in once by hand in a real browser, reuse the session for every later capture | Playwright |
do.py |
Create, drive, and destroy lab VMs — guarded so it cannot delete a sibling’s box | DO token |
推荐工具
换一个关键词，或者移除筛选条件。
安装
npx skillfish add hassancs91/content-factory-starter