Why I built a pipeline
I write, brainstorm and use Notion as my second brain. My website lived in a code repo.
Every essay meant copying text, fixing formatting, and remembering how the build worked.
I wanted to have my writing saved in one place and publish with one sentence.
How it works
Five stages: Write, Scan, Decide, Publish, Record.
- WriteI brainstorm and write in Notion, and prepare schematics if necessary. Then I set the entry to Ready.
- ScanClaude reads every Notion entry marked Ready, and every published entry I have changed since the last sync. It scans my Ideas Library for essays worth drafting. When I ask, it compares my LinkedIn profile with the résumé on this site.
- DecideClaude shows me one summary and asks about each real choice. Nothing changes without my review.
- PublishClaude writes the webpage and stamps it with the publish date, or the update date. It builds the site, which also makes a share image for every page. It previews the result and checks every changed line. Then it pushes to GitHub, and AWS Amplify puts the site live.
- RecordClaude marks the Notion entry Published and adds a dated entry to a log.
Guardrails
- Anything marked Confidential never publishes.
- I keep a list of sensitive terms, such as employer and internal project names. Claude flags each one in an essay and removes it, unless I say otherwise for that essay.
- Text that I wrote in Notion goes live as written. Claude fixes only clear typos and tells me what it fixed.
- Claude never signs in for me and never types a password.
- Claude can read and update my Notion pages without asking each time. Deleting or moving a page still needs my yes.
Memory between chats
- A new chat with Claude starts with no memory.
- So the pipeline keeps a log, with a copy in Notion. Each new chat reads it first, checks Notion for anything new or changed since the last sync, and tells me what is waiting and how my website goals are tracking.
- I can move to another computer and pick up where I stopped.
Counting visits
- The site counts visits with Umami. It uses no cookies and does not follow anyone across other sites, so there is no cookie banner.
- It counts only the live site, not my preview builds.
- It tells me which pages people actually read.
What I learnt
- The guardrails took more thought than the automation. Publishing is easy. Not publishing the wrong thing is the real challenge.
- Asking at each step builds trust. I let Claude work harder because it always asks first before publishing anything.
- Plain-English rules beat code. The pipeline is a few instruction files I can read and change, plus two small scripts that remind Claude of the rules at the right moment.
Try it yourself
The pipeline is open source. The repo includes a Notion template and step-by-step setup, including AWS Amplify hosting.