The hub and the projects share standards and stay aware of each other through git alone, with no coupling between them. The canonical machine copy of this standard is in the repository at hub/standards/cross-project-sync.md.

The rule

Communication is git-only, one-directional per flow, and happens only on explicit request. There are no submodules, no package dependencies, no build-time coupling, and no automation that reaches across repositories. Each side reads a shallow, read-only clone of the other when asked to. Both directions track the dev branch.

This keeps the repositories modular and prevents recursion: because nothing on one side automatically triggers a pull on the other, they cannot set each other off in a loop.

The two flows

• Hub reads projects. The hub keeps read-only shallow clones of each project under assets/references/ (git-ignored, never committed) so their changes can be summarised on the blog.
• Projects read the hub. Each project keeps its own read-only shallow clone of the hub and copies what it needs out of hub/standards/ and hub/templates/ into its own tree, committing that copy. Adoption is a copy, not a live link.

Why not submodules

Submodules pin a commit and couple repositories at clone and build time — the opposite of the goal here. A throwaway shallow clone in a git-ignored folder provides the content to read with no coupling and no history weight.