Feature request for product/service
Cursor IDE
Describe the request
Earlier versions of Cursor had excellent first-class worktree support, and it was one of my favorite features. The new /worktree slash-command flow is a real regression. A few concrete pain points:
1. Latency before anything happens. The agent has to read the worktree prompt template, run git rev-parse, mkdir, git worktree add, check .cursor/worktrees.json, then run setup commands. Each of those is a separate tool call with its own round trip. Going from “I want a worktree” to “I can type a command in it” takes 30–60+ seconds instead of ~1 second.
2. Approval spam. Every shell step prompts me to click “Run”, even for commands I’ve allowed a hundred times (git worktree add, mkdir ~/.cursor/worktrees/..., just install, cp .env). The first mkdir reliably fails inside the agent sandbox and then has to retry with elevated permissions — another click. This breaks flow completely.
3. No tab/window indicator for worktrees. Once I have multiple chats and multiple windows, I have no way to tell which tab is operating in a worktree vs the main checkout, or which worktree. I used to be able to glance at the window and know. Now I’m squinting at paths in terminal output.
4. It’s implemented as an instruction to the LLM rather than a product feature. That’s why it’s slow and fragile — the agent has to do filesystem discovery every time, and any change to the template risks breaking the flow. This should live in the IDE, not in a prompt.
What I’d like back / want added:
- Native (non-agent) worktree creation, like the earlier versions had.
- Auto-approval of the known worktree setup commands (or a “trust this flow” toggle).
- A visible badge/label on the tab title, window title, or status bar showing: worktree name, branch, and whether it’s a worktree vs the main checkout.
/apply-worktreeand/delete-worktreeas real commands, not templates.
I still love Cursor — this one feature just got worse and I’d love to see it restored.