Git Worktrees for Parallel Agents

Using Git worktrees to run multiple Claude Code sessions against the same repository in separate working directories, so parallel agents can work without constantly colliding over the same files.

Key points

• Claude Code's --worktree flow creates isolated working directories tied to separate branches ^[src-011]
• This enables three to five parallel sessions to explore or implement different parts of a project without file conflicts ^[src-011]
• Worktrees are especially useful for comparison tasks: one agent can test an approach while another explores an alternative ^[src-011]
• The pattern works best when each agent has a clear ownership boundary and a narrow brief ^[src-011]
• GitHub's June 2026 explainer reframes worktrees for the AI coding era: humans and agents now work in parallel more often, so separate directories reduce stash conflicts and editor disruption ^[src-098].
• Worktrees carry operational costs: duplicate dependencies, folder cleanup, ignore-file hygiene, and the rule that one branch cannot be checked out in multiple worktrees at the same time ^[src-098].

Related entities

• Claude Code — runtime that launches worktree-backed sessions

Related concepts

• Agent Teams — related multi-agent coordination pattern
• Claude Code Decomposition Pattern — split the project into parallelisable parts
• Session Chaining — complementary approach for sequential decomposition
• Harness Engineering — wider operating layer that makes agent work safe to run in parallel

Source references

• ^[src-011] Nate Herk — Claude Code power features cluster (2026-04-20 to 2026-04-27)
• ^[src-098] Cassidy Williams – "What are git worktrees, and why should I use them?" (2026-06-16)