Why not use a worktree?Does this work on Linux?What happens on ext4?Can I run it from inside a clone?How do I keep the canonical current?How do I undo one?What if the canonical moves?Does it need GitHub, or a remote at all?How is this different from a fresh git clone?

Does this work on Linux?

The script is POSIX and uses no BSD-only spelling, so the reflink path calls `cp --reflink=always` and reads free space through `df -k`. It has not been run on Linux. The probe measures what a copy actually consumed, so a filesystem it guesses wrong about refuses the clone rather than making a full copy of 13 GB.

The script is POSIX and uses no BSD-only spelling, so the reflink path calls cp --reflink=always and reads free space through df -k. It has not been run on Linux. The probe measures what a copy actually consumed, so a filesystem it guesses wrong about refuses the clone rather than making a full copy of 13 GB.

Why not use a worktree?

Use one when the tracked files are all you need. A worktree shares one repository, which is why it costs almost nothing and why it carries almost nothing: no dependencies, no build output, no initialised submodules, and a branch that no second worktree may check out.

What happens on ext4?

It refuses and says why. ext4 has no way to share blocks between two files, so every copy would cost the full size.