Claude Code Worktrees, Limits Draw Fire

Claude Code is powerful, but every extra layer of autonomy increases the blast radius when repo state handling goes wrong. If the tool is going to own worktrees, it needs to be boringly safe.

• –Worktree automation is useful only if it preserves uncommitted changes by default and makes destructive actions explicit
• –This is the classic agentic-tool tradeoff: fewer prompts for the user, more hidden side effects for the system
• –Rate-limit complaints matter because coding agents are workflow utilities, not chat toys; predictable quotas are part of trust
• –Anthropic’s own docs push worktrees and parallel sessions hard, so failures around repository isolation are especially visible
• –If Claude Code wants to be the default terminal copilot, it needs stronger guardrails and clearer recovery paths