Scribe

Scribe is a publishing client I built for my own blog. Pages CMS lets you define a site with arbitrary collections and fields, which is flexible but makes editing through a generic CMS form a chore. Scribe reads that schema and hands each content type a first-class editing experience - a real block editor for posts, a tag manager - while preserving anything it doesn’t have a dedicated editor for as raw, editable fields.

How it works

• Schema-driven. It parses the site’s .pages.yml, auto-detects collections and fields, and a setup wizard maps each collection to an editing experience written into a config file.
• Lossless markdown. The riskiest code is a custom TipTap parser/serializer that round-trips markdown byte-stably, including verbatim raw HTML blocks and captioned images. It’s covered by round-trip fidelity tests.
• Image upload built in. Paste or drag an image and it enters a staging state - pick a name and destination - before it’s shipped to object storage or copied into the site’s local media dir.
• Pluggable auth. Runs open for local dev, or as a full OpenID Connect client (Authorization Code + PKCE + refresh, server-side sessions, optional group gate). In my homelab it’s the reference app secured against a self-hosted Hydra and locked to an admin group.

Notes

Go service over a Git repo checkout plus a private notes store, with a React + Vite + TipTap front end. Markdown stays the source of truth; the editor is just a view over it.