The session opened with a question I’d been ignoring: had any of the casehub-platform blog entries actually made it to mdproctor.github.io? I brought Claude in to check. We looked at the _notes/ directory on the Pages repo. All twelve entries were there — every one from 2026-05-19 through 2026-05-22, already published from previous sessions. I’d expected a gap. Instead, nothing to fix.

That left one loose end. If publish-blog were run as a standalone command — outside of work-end, which already knows to pass $WORKSPACE/blog/ — it would scan docs/_posts/ by default and find nothing. This workspace keeps blog entries in blog/. The auto-publish path at work-end was fine; standalone wasn’t.

We added a one-line blog-routing.yaml to the workspace root, extending the global config and leaving a comment about the source path. Small thing, but publish-blog picks it up and the routing is now explicit.

I also parked issue #8 — a preferences editor admin UI — I’m not doing UI work at the moment. The next real work is the GroupMembership OIDC provider, or confirming whether engine#329 was already fixed before I filed it.


<
Previous Post
The Four-Layer Agent
>
Next Post
Channel IDs make terrible audit keys