On 13 November I took part in Umbraco Builders Day, a community build day where everyone brings a project and works on it in company. Mine was this blog. It was a Hugo site, and I wanted it in Umbraco 17 for one reason above the others: the front page.
A front page with an editor
The home page is laid out like a broadsheet, with a lead story, three second stories and a Start here band. Each slot has a dek written for the front page, and some have a shorter headline that only appears there. What goes where is an editorial decision, made by importance and not by date.
On the Hugo site that decision was a TOML file. Changing the lead meant editing data/frontpage.toml, running a Python script that checked the edition, committing, pushing and waiting for the build. It worked, and it meant the home page could only change with a commit.
In Umbraco the Front Page is a node with three block lists. I open it in the backoffice, pick the stories, write the deks and press publish. The story picker is a small property editor of my own, a flat list of the published articles, newest first, with a filter box, so I am not opening seven lanes in a tree to find one post among a hundred.
The rules moved from the script into the site, where they run on every publish of that node:
- one lead, and it has a cover image;
- three second stories and three in Start here;
- every story is a published article, and none appears twice;
- the lead and the second stories each come from a different desk;
- every slot has a dek, inside a word limit, that is not a copy of the post's standfirst;
- a front page headline fits its slot and has no colon or exclamation mark.
A publish that breaks one is cancelled, and the page the public sees is left as it was. That holds for the backoffice, for the Management API and for the job that refreshes the Trending now rail, so I can no longer publish a broken edition by forgetting to run a check.
Posts and editions that go out on their date
I write ahead, and eleven posts are queued as I write this. Hugo's scheduling was to skip any post dated in the future, with a cron job rebuilding the site at 07:15 UTC every morning so that a post appeared on its day. Everything I Automated on This Blog has the story of what went wrong when “new to the repo” and “published today” were treated as the same thing.
Umbraco has scheduling built in. Each queued post is saved with a publish time of 07:15 UTC on its date and stays unpublished until then, so a post that is not out yet cannot leak into a listing, the feed or the search index.
The front page schedules too, which is the part I wanted most. It is a document like any other, so I can set up Monday's edition on Friday, give it a time, and it goes out by itself through the same rules. An edition that leads with Monday's post has to be timed after that post, because the rules refuse a story that is not published yet.
What moved, and how
The site is Umbraco 17 on .NET 10 with SQLite. I did the work with Claude Code, which talks to Umbraco through the Umbraco Developer MCP server: it read the Hugo repo, proposed a content model, built it once I had agreed it, and migrated the content with a manifest so that a rerun never creates anything twice.
- 104 posts became Article nodes under seven lanes, which are the six Hugo series plus Specials. The lanes share one document type and one template, and get their colours and layout from settings on the lane.
- 47 poems became Poem nodes holding their book-page images and their text.
- 42 art pieces stayed as the programs they are, served unchanged, each with an Artwork node for its card.
- Post bodies stayed Markdown. The
figureandchartshortcodes are expanded by the site's own renderer, so nothing in a body was rewritten. - Old addresses redirect. Every
/posts/<name>/link answers with the page's new address.
The public site is still static files on Azure Static Web Apps. Umbraco runs on my laptop, a tool saves every page it renders, and a daily job uploads whatever changed. I get a CMS to edit in and nothing to patch or keep awake on the internet. The cost is that the laptop is the publisher, so a morning it spends in a bag is a morning nothing goes out until I run the deploy by hand.
The same skills, pointed at a CMS
I run this blog through a set of Claude Code skills, and they came across with their names unchanged. /frontpage proposes an edition, saves it to the Front Page node as a draft, checks it, shows me a screenshot and publishes when I agree. /trending asks the site to refresh the rail from the last seven days of page views. /blog-image generates a cover and uploads it to the Media library. /social-card builds the share images from the article's own fields. /deploy-now is the daily publish run by hand.
On Hugo each of those ended in a commit. Here each ends in a save to the database, which is written out as SQL text and committed once a day.
A GoatCounter package
Page views here are counted by GoatCounter, and I wanted the numbers beside the content. That became a package, Umbraco.Community.GoatCounter, with its source on GitHub. It was scaffolded with Lotte Pitcher's Opinionated Package Starter, which did the dull parts of making a package for me.
It adds an Analytics dashboard to the Content section and a Views tab to every document, so each page shows its own traffic. It also has a first-party beacon: the page posts to an address on your own site, and a queue forwards the views to GoatCounter in batches, which gets round the blockers that stop GoatCounter's own script. The Trending now rail reads its numbers through the same package.
The package is at version 0.0.3. GoatCounter's API sometimes answers a good request with a 404, so the client now retries a few times before giving up. That fix is in the repository and not yet in a release.
Tile walls as content
The Patterns posts are about a tile engine I wrote, which paints a wall from a handful of choices and a seed. On the Hugo site each wall was a page of its own. In Umbraco a wall is a property. The band under the header of the About page is stored on that page as this:
{ "engine": "azulejeria@2.5.0", "world": "wall", "pack": "azulejeria", "seed": 216,
"repeat": "module", "pal": "stepintodev", "motif": "laceria", "density": 4,
"hand": 0.4, "brush": 1, "scale": 110, "bands": true, "grout": true, "age": false }
The node holds the choices and the browser paints the wall from them. The engine's random numbers are seeded, so the editor's preview and the public page show the same tiles. The property editor has a live preview, the pack's presets as cards, and the sliders the original art page had. Each lane now has its name lettered in tiles as its band, and the 404 page breaks a wall into pieces seeded from the address you asked for.
Thank you
Thank you to the people who organised Umbraco Builders Day, and to everyone in the community and at Umbraco HQ who was around that day and helped. It was a rather good way to spend a day.