Theron S

How I run multiple Claude Code agents in parallel using git worktrees. Please comment if there are better ways :) ``` function wt() { git worktree add ../worktrees/$1 main cd ../worktrees/$1 && git checkout -b $1 } ``` What happens under the hood: • `git worktree add` creates a linked working directory • Same .git repository, completely separate working files • No need to stash or commit when switching tasks • Each worktree can have different code states running Example: `wt feature-auth` creates: ``` my-app/ ├── .git/ (shared repository) ├── src/ (main worktree) └── ../worktrees/ └── feature-auth/ (isolated worktree) └── src/ (independent working files) ```` The cleanup function removes all worktrees with surgical precision: ```bash function wtc() { git worktree list --porcelain | grep -B2 "branch refs/heads/" | grep "worktree" | cut -d' ' -f2 | xargs -I {} git worktree remove {} } ```` Pipeline breakdown: • `list --porcelain` → machine-readable format • `grep -B2` → captures worktree path above branch info\ • `cut -d' ' -f2` → extracts just the path • `xargs` → removes each worktree cleanly Why this changes everything: 1. Run Claude on issue #123 while another fixes PR #456 2. No context switching—each agent has its own filesystem 3. Test breaking changes without touching main workspace 4. Compile different versions simultaneously Git worktrees were built for this. Multiple working directories, one repository. Perfect for parallel AI development.