wt: exec shell vs shell integration

Previously. Problem statement: my wt tool’s cd subcommand does not survive a terminal split — the new pane opens at the original directory, not the worktree. worktrunk’s wt does(!) A subprocess cannot change its parent’s working directory (PWD). wt.sh works around it by replacing the script process with a fresh shell: cd "$target_path" || exit 1 exec "$SHELL" The exec "$SHELL" is the culprit. The interactive shell process is gone, replaced by a new one rooted in the worktree. The terminal emulator’s record of “what shell is (currently) running in this pane” still points to the original PID’s cwd at split time — not the new shell’s. Shell history, env vars, pushd/popd stack, prompt state: all lost too (not that I particularly care about all these). worktrunk ships a shell integration step: % wt config shell install That installs a function in ~/.bashrc / ~/.zshrc (or equivalent for other shells) which evals the binary’s output in the current shell. The cd happens in the shell we are…