Two-site publishing system

A typed content graph compiled deterministically into two fully prerendered sites, with an agent-operable editorial workflow.

One typed content graph, compiled into two independent publications. Records are validated Markdown with typed frontmatter rather than pages. Each build selects the records eligible for the site being built, projects every one of them into that site's own presentation, and emits fully prerendered routes with their own metadata, feeds, sitemap and search index. Nothing is decided at runtime from the hostname, so the two publications cannot drift into each other by accident.

What it does

Citations resolve and number themselves across a document set. Media passes through one optimization pipeline that generates responsive widths, posters and inline placeholders. An editor writes the same records the public build consumes, so publishing is a commit rather than a database write. An output gate runs on every build and fails it when one publication's identity appears in the other's output. It checks text, canonical metadata, and the images themselves, which are pinned by hash so an unreviewed picture cannot reach the wrong site.

AI-native workflow

Coding agents worked across the content model, the compiler, the prerender, the media pipeline, search, the editor and the tests. I own the schema, the site boundaries, the verification rules and every release.