Git Worktree Runner

Git Worktree Runner

Use isolated worktrees to execute multiple tickets in parallel from the same repository.

Input

• Ticket id and short slug.
• Repo root path.
• Base branch (typically main).
• Constraint: Branch names must use codex/<ticket-id>-<slug>.

Output

• Dedicated worktree folder per ticket.
• Isolated branch context for each agent.
• Clean teardown steps after merge.

Process

Phase 1: Provision Worktree

1. Fetch latest remote state.
2. Create branch: codex/<ticket-id>-<slug>.
3. Create worktree path (for example: ../wt-<ticket-id>).

Phase 2: Work in Isolation

1. Run implementation and tests inside the worktree path.
2. Before starting significant new work, fetch the latest state of the target branch and rebase your worktree branch. If the rebase introduces conflicts, resolve them before continuing.
3. Commit only files related to the ticket.
4. Push branch and open PR.

Phase 3: Teardown

1. After merge, remove worktree directory.
2. Delete local branch if no longer needed.
3. Prune stale worktree metadata.

Useful Commands

git fetch --all --prune
git worktree add ../wt-35 -b codex/35-b5-auto-unarchive main
cd ../wt-35
# ... work, commit, push, PR ...
cd -
git worktree remove ../wt-35
git branch -d codex/35-b5-auto-unarchive
git worktree prune

Completeness Checklist

• □ Worktree path maps one-to-one with one ticket.
• □ Branch naming matches convention.
• □ Worktree is removed after merge.

Rules

1. ALWAYS keep one ticket per worktree.
2. ALWAYS run commands from the intended worktree directory.
3. NEVER mix unrelated ticket commits in one worktree branch.
4. NEVER leave stale worktrees after ticket merge.