https://blog.dengler.io/ Sam Dengler Blog 2026-10-10T16:24:58.000Z https://github.com/jpmonette/feed Sam Dengler Blog <![CDATA[Images and diagrams in a post]]> https://blog.dengler.io/images-and-diagrams/ 2026-10-10T16:24:58.000Z This post checks that a screenshot and a diagram survive the trip from a Claude Doc to the published site.

A screenshot​

This is the first post as the site rendered it. The image was uploaded to the Doc and placed in the Post tab.

The first post rendered on the blog, with the post list on the left and the section list on the right

A diagram​

This diagram was drawn in the Doc. It shows the path a post takes to the site.

Nothing reaches the site without a merged pull request: Claude Doc, Markdown export, converter, pull request, published site

The pull request is the one step that waits for me.

]]> Sam Dengler https://github.com/samdengler <![CDATA[Writing in Claude Docs, publishing with Docusaurus]]> https://blog.dengler.io/claude-docs-to-docusaurus/ 2026-10-10T15:58:17.000Z This blog is written in Claude Docs and published as a static Docusaurus site. I wanted the drafting and research to happen where I already work with Claude, and I wanted the published result to be plain files in a git repo.

How a post gets here​

  1. Each post is one Claude Doc with two tabs. The Post tab holds the writing. The Research tab holds notes and sources and is never published.
  2. An index Doc lists every post with its slug, date, tags, summary, and status.
  3. When I set a post to Ready, Claude exports the Post tab to Markdown and runs a converter script that adds front matter and writes the file into the repo.
  4. Claude opens a pull request. I read the diff and merge it.
  5. GitHub Actions builds the site and deploys it to GitHub Pages.

Docs as the source of truth​

Edits happen in the Doc and flow one way to the site. The repo is never edited by hand, so the two cannot drift. The pull request is the review step. It shows what changed since the last publish.

Readable by machines​

Every publish also writes a posts.json index, a plain Markdown copy of each post, and an llms.txt file. An MCP server can list, search, and fetch posts from those files without scraping HTML.

Open items​

Images and diagrams are next. The converter does not handle them yet, so this first post is text only.

]]> Sam Dengler https://github.com/samdengler