thurbox: Skill for Claude Code

Thurbox remote SSH and WSL hosts

Working reference indexed by AGENTS.md. The rationale behind these decisions is owned by the docs under docs/; a change that invalidates what this says updates it in the same PR.

Remote SSH & WSL Sessions

Sessions can run on an off-local host while the TUI runs locally: a remote machine over SSH, or a local WSL distro (wsl.exe). A WSL distro is modeled as "SSH without the ssh" — the only difference is the launch prefix (wsl.exe -d <distro> vs ssh <dest>); tmux, git, the agent, and the worktrees all run inside the distro at native Linux paths, so everything downstream of the launcher (control-mode protocol, POSIX quoting, worktree layout) is identical to the SSH path — no wslpath translation. Hosts are declared as data in ~/.config/thurbox/hosts.toml (seeded commented-out; fresh install = zero SSH hosts, behaves as before), plus WSL distros are auto-discovered (wsl.exe -l -q) with no config — on Windows and, via interop, from inside a distro as well. What discovery never offers is the distro thurbox is running in: a loopback (HostDef::is_wsl_loopback, keyed on $WSL_DISTRO_NAME) is this machine, so sessions on it are local. It is dropped from both halves of the registry — discovered and configured (the configured one with a warning). A configured host that reaches a sibling but is named after the current distro (shadows_current_wsl_distro) registers as wsl:<us> and is therefore indistinguishable from a loopback row. It is kept exactly as written — it works — and what it costs instead is the repair, for that one name: the rows under it are the bug's local rows and the host's own sibling rows at once, so both rewrites are wrong for half of them and neither is attempted (a notice says so, counted with rows_recorded_on and skipped when the claimed spelling holds no row). That is one case of a general rule — the repair never rewrites a spelling a host it still serves registers under, including a dropped loopback's own free-label name, which auto-discovery may hand straight to a real sibling. Rows a released build already relabelled wsl:<us> (being shareable by default, the loopback was mirrored, and its "host" database was this database, so the pass rewrote our own local rows as remote) are put back by a one-time repair rather than by the migration: schema v47 only marks it owed, and session_ops::repair_wsl_loopback_rows runs it from every startup that opens the database — the TUI boot and the thurbox-cli entrypoint. Siblings stay ordinary hosts. The seeded file documents every field inline; the schema: