The interface is localized. Skill descriptions and documents are machine-translated by AI into your language; code, commands, paths and technical terms stay in the original, and the author's document governs.
Explore skills
Development · SKILL.md

worktree

worktree

Use for non-trivial or parallel work that should happen in a separate git worktree.

Independent GitHub snapshot · Automated classification · Original documentation governs

Skill documentation

GitHub

name: worktree description: Use for non-trivial or parallel work that should happen in a separate git worktree.

Worktree

If topology=multi-module (triage announcement or coordinator spec/plan declaration), read ../references/multi-module.md. Treat each repository as an independent Git and worktree boundary; never run a workspace-wide Git mutation from their common parent.

Before creating worktrees, resolve each registered repository and verify status and baseline. Map paths in the coordinator plan and use the change-set ID in names. A missing repo, unrelated dirty change, or red baseline blocks the change. Detached HEAD is allowed for read-only integration, not implementation. Do not clone or clean automatically.

git worktree add ../<repo>-<topic> -b <type>/<topic>   # <type>: feat | fix | chore | ...
cd ../<repo>-<topic> && <setup> && <run tests>

Baseline must be green in every participating repository. If red, STOP - don't build on broken ground. Done → ship.

Tool scope

Worktrees are separate directories from the main repo. Semantic indexing (code graph, symbol search) covers the original repo, not the worktree. File tools (read, grep, glob) work normally. For short-lived worktrees this is acceptable; for long sessions, note this limitation to the agent at dispatch.

Saved skills