wochap/wt/skill

wt

Use in projects laid out as a bare .git with worktrees as sibling dirs:

project/
├── .git/       # bare repository
├── main/       # default-branch worktree
└── feature-x/  # branch worktree (slashes in branch names become '-')

wt clone <url> [dir]              # new wt project: bare .git + default worktree
wt switch                         # stdout: default-branch worktree path
wt switch <ref> [name]            # stdout: worktree path (created if missing)
wt switch -b <branch> [from]      # new branch + worktree; stdout: its path
wt list                           # all worktrees with status
wt rename <new-name>              # rename current worktree dir; stdout: new path
wt pull <source>                  # squash source worktree's HEAD into current, staged
wt pull <source> --staged         # apply only source's staged diff
echo y | wt rm <name>             # remove worktree + local branch
wt doctor                         # repair broken worktree links

• wt changes no directory by itself. The cd wrapper in wt.plugin.sh is

for interactive shells only. Use the printed path: cd "$(wt switch <ref>)" or git -C "$(wt switch <ref>)" ....

• Stdout of switch, clone, rename is only the absolute path; status,

hook output and errors go to stderr. Nonzero exit means no path.

• <name> for rm/pull is the worktree dir name (e.g. feature-login),

not the branch name. pull also accepts a path to another clone.

• rm and pull (into a dirty target) prompt [y/N] on stdin; pipe echo y

only when removal or the merge is intended. rm deletes the local branch; --force drops uncommitted changes, --remote also deletes origin/<branch>. Never pass either without the user asking.

• wt switch runs post_create hooks from the worktree's wt.json (copy,

symlink, command) only for newly created worktrees. A hook failure exits nonzero but leaves the worktree and branch in place; fix and rerun the step by hand.

• pull stages changes in the current worktree; it does not commit. On

conflicts, resolve them manually.