Roadmap

Next: automatic site migration#

Right now the entry point for a new project is /agntcms-init-from-artifact: you bring a Claude artifact app (React, Vue, or HTML+CSS), the skill reads the source, maps every block to a section type, migrates design tokens, and populates the content JSON. That gets you from a working app to a working CMS in one pass.

The next step is removing the artifact step entirely. Two things are planned:

Direct import from Claude Design. Instead of exporting to an artifact first, you hand agntcms a Claude Design bundle directly. The skill reads the bundle (tokens, mockups, copy), creates section types, fills the pages — same result as today, without the intermediate app generation step.

Migration of any existing site. Point the skill at any live URL or local project — not just a Claude artifact. The agent reads the HTML, extracts the design tokens and visual structure, maps blocks to section types, and wraps the whole thing in the agntcms content model. No prior Claude Design or artifact required.

Both of these build on the same core pattern as /agntcms-init-from-artifact — the infrastructure is already there. The work is in making the source reader robust enough to handle arbitrary markup and styles reliably.

After that — a proper Postgres adapter#

The filesystem adapter is the default in v0.5 and works well for the size of site agntcms targets — a few hundred pages max, one repo, content versioned through git. It is not the right shape for every deployment.

The plan is to add a Postgres adapter — properly meaning: drafts, published pages, version history, and the admin endpoints all behave the same as on the FS adapter, with no changes to section code or skills. The interesting design questions are around how snapshots and history map onto rows, how previews work without writing to the live database, and how multi-writer editing degrades — not whether to add a database driver. That part is easy.

The goal is to land it without compromising the single-stack, no-config feel of v0.5.

And after that — TBD#

Everything past the Postgres adapter is TBD by community and release feedback. v0.5 covers a tight slice on purpose. Before we extend the surface area, we want to see which edges people actually hit in real projects:

• Where the filesystem adapter starts to hurt (and whether Postgres is enough, or other backends matter).
• Which section patterns repeat across projects often enough to be worth pulling into the framework.
• Which skills feel wrong in practice — too narrow, too broad, missing a step.
• How agent-driven editing holds up at the scale of a real editorial team, not just a small site.

If you have run into something concrete, open an issue or start a discussion. That is how the next items on this list get picked.