new-worktree

name	new-worktree
description	Create an isolated git worktree under .claude/worktrees/ for a sub-agent to work in. Use this before dispatching any coder agent so that parallel work cannot collide.

Skill: new-worktree

Labeling (Sleipnir): in user-facing prose, call this "sending Sleipnir to that branch" — the worktree-traversal labeling convention (see ravenclaude-core/CLAUDE.md → "Sleipnir"). Labeling only; the mechanics below (git worktree) are unchanged.

Inputs

• role — one of: architect, backend-coder, frontend-coder, fullstack-coder, tester.
• slug — short kebab-case identifier of the task (e.g. auth-refresh, nav-redesign).
• base — branch to fork from. Default: main.

Procedure

1. Validate inputs. Slug must match ^[a-z0-9][a-z0-9-]{1,40}$. Reject otherwise.
2. Resolve paths.
   • Path: .claude/worktrees/<role>-<slug>/
   • Branch: agent/<role>/<slug>
3. Refuse if either exists. If the path or branch already exists, surface it — do not silently reuse. The Team Lead must decide whether to clean up the old one or pick a new slug.
4. Sync base. git fetch origin <base> so the worktree starts from current upstream.
5. Create the worktree.

   git worktree add -b agent/<role>/<slug> .claude/worktrees/<role>-<slug> origin/<base>
   

6. Install dependencies if needed. If the project requires a per-tree install (e.g. pnpm install), run it once in the new worktree before handing off.
7. Report the path and branch back to the Team Lead. The Team Lead then briefs the agent and points it at this directory.

Peer-process / IDE

After create, write the lane stamp (plugins/ravenclaude-core/scripts/write-lane-stamp.sh <dest> <task> <branch> <created_by>) and open a dedicated VS Code window (rcwt new / code -n <path>). Never add the new worktree as a second folder in an existing window — multi-root pools Chat context. Do not restyle the sub-agent procedure above.

Cleanup

Worktrees are NOT auto-removed. Use cleanup-worktrees when the task is integrated.

Why this exists

Two parallel coder agents on the same working tree will silently corrupt each other's diffs. Worktrees are cheap; recovery from a stomped diff is not.

Why a skill, when the Agent tool has built-in isolation: "worktree"?

The Agent tool's native worktree isolation is a one-off, ephemeral worktree that gets cleaned up automatically when the sub-agent makes no changes. This skill is for the Team Lead's managed worktrees that persist across sub-agent runs: predictable paths under .claude/worktrees/<role>-<slug>/, predictable branch names agent/<role>/<slug>, and tracked by cleanup-worktrees so they don't accumulate silently. Use the native isolation for fire-and-forget exploration; use this skill when the worktree is part of a multi-step Team Lead workflow that needs to be visible to subsequent agents.