Headless servers have no DISPLAY or WAYLAND_DISPLAY, so xclip and xsel
cannot reach a clipboard even when installed. These shims use OSC 52 to
hand text to the terminal emulator on the other end of the SSH connection,
which puts it on the local machine's clipboard.
Inside tmux, pbcopy uses `tmux load-buffer -w -` so the text lands in the
tmux paste buffer and is forwarded outward (requires `set-clipboard on`).
Outside tmux it writes the escape sequence to /dev/tty directly. Both fall
back to xclip/wl-copy when a real display is present.
pbpaste reads the tmux buffer only; terminals refuse OSC 52 reads, since
that would let any remote program siphon the user's clipboard.
omo keys project memory by folder path, so each workspace started with an
empty store. owt new and owt open now link the workspace's memory id to the
main checkout's store, and owt done and owt rm remove the link.
owt new creates a randomly named workspace per task under ~/worktrees/<repo>,
copies .env files, installs dependencies and opens omo in a tmux window with
the workspace's Linear conventions. owt done checks merge state (git or a
merged Gitea PR via tea) before removing the worktree, branches and window.
The failure path advised --deep and a disk-space check regardless of cause.
On the first live run that was wrong on both counts: 101 GiB were free and
--deep was irrelevant. Docker Desktop had crashed on startup, and the actual
fix was launching it from the Dock.
Branch the advice on whether any Docker process survived the relaunch. No
survivors means the app did not stay up, which no programmatic relaunch can
work around; survivors mean it is up but not answering yet.
David Ibia
David Ibia