wt

name	wt
description	Create a fresh git worktree from a short description and drop the session straight into it. Use when the user types `/wt <description>` or asks to "start a worktree for X", "spin up a worktree", "work on X in a worktree", or "branch off into a worktree". Derives a kebab-case name from the description and calls the native EnterWorktree tool (fresh from origin/main).

/wt <description> is a thin ergonomic wrapper over the harness's native EnterWorktree tool. It turns a free-text description into a tidy worktree name so the user can type /wt improve cluster coverage and land in an isolated workspace without hand-naming anything.

The skill creates + enters the worktree, then offers to link real catalog data. It never starts the dev server (that stays opt-in via /dev). The link-data offer exists because a fresh worktree has no public/data/ at all — every catalog file there is a gitignored build output or a data/-staged copy — so without linking, the renderer falls back to the synthetic procedural cloud. The offer is a yes/no prompt, not an auto-run: doc/planning worktrees don't need data, and a pipeline-rebuild worktree wants its own public/data/.

Steps

1. Read the description from the slash-command arguments (everything after /wt). If it's empty, ask the user what the worktree is for rather than inventing a name.

2. Derive a kebab-case name from the description:

   • lowercase everything
   • keep only letters, digits, and dashes; replace any run of other characters (spaces, punctuation) with a single -
   • trim leading/trailing dashes
   • keep it short — roughly the first 3-4 meaningful words, max ~40 chars. Drop filler words ("the", "a", "for", "to") when trimming for length.

   Examples:

   • /wt improve cluster coverage → improve-cluster-coverage
   • /wt fix the InfoCard hover bug → fix-infocard-hover-bug
   • /wt experiment with MSDF kerning → experiment-msdf-kerning

3. Call EnterWorktree with name set to the derived kebab string. The harness branches fresh from origin/main (the default worktree.baseRef), creates the worktree under .claude/worktrees/, and switches the session into it. The harness prefixes the branch with worktree- automatically — pass only the clean kebab name, not a worktree- prefix.

4. Report the new worktree path and branch that EnterWorktree returns, in one line, so the user knows where they are.

5. Offer to link real data. Ask the user (yes/no) whether to run /link-data now, noting the worktree starts with no public/data/ so the renderer would otherwise fall back to the synthetic cloud. If they say yes, invoke /link-data (it symlinks this worktree's public/data/ → main's). If they're here for doc/planning work or a deliberate pipeline rebuild, a "no" is the right answer — don't push. Don't link without an explicit yes.

Guardrails

• Already in a worktree? EnterWorktree refuses to nest. If the session is already inside a worktree, don't silently fail — tell the user they're in <current worktree> and ask whether to ExitWorktree (keep or remove) first, then re-run /wt.
• Name collision. If a worktree with the derived name already exists, EnterWorktree will error. Append a short disambiguating suffix (e.g. -2) and retry, or ask the user for a different description.
• Don't commit unrelated work into the new branch. The worktree is for the task the description names; if the user later wants something off-topic committed, that's a separate /wt.