working-in-parallel_skill

Overview

This skill guides working in parallel with other agents by using git worktrees as the recommended approach. It explains how to create isolated working directories that share Git history, how to avoid interfering with another agent’s session, and when to prefer full repo clones.

How this skill works

The skill instructs creating a separate worktree for each branch so multiple agents can edit the same repository without file conflicts. Each worktree has its own working files and branch context but shares the main repo’s .git metadata and remotes. It also covers safe setup checks, dependency installation, push/PR workflow, and cleanup commands.

When to use it

• Another agent is actively working in the same directory
• You need to work on a quick fix while a long-running task is in progress
• You want to develop multiple features in parallel without file conflicts
• You prefer minimal disk usage compared to multiple full clones
• When you need to push separate branches and create independent pull requests

Best practices

• Check project setup docs and project memory before running install commands
• Verify the correct package manager (npm, pnpm, bun) before installing dependencies
• Name worktree directories clearly (e.g., ../repo-feature-auth)
• Commit and push changes before removing a worktree
• Use git worktree list to track active worktrees and avoid stale directories
• If unsure or the user prefers full isolation, use separate repo clones instead

Example use cases

• Create a worktree to fix a critical bug while another agent runs tests in the main directory
• Spin up one worktree per feature branch when developing multiple features in parallel
• Use a worktree to prepare a PR for review while the main agent continues refactoring
• Start a temporary worktree for an experimental change and remove it after merging