Desktop Code sessions: option to disable automatic worktree creation

Problem

Every time Claude Desktop starts a new Code session, it automatically creates a fresh git worktree under .claude/worktrees/<random-name>/. Each worktree gets its own project folder in ~/.claude/projects/, which means:

1. Conversation history is fragmented — no two Code sessions share a recents list, so previous sessions appear to "disappear" from the sidebar
2. Sessions cannot be resumed — starting a new Code session creates a new worktree, so there's no way to continue where you left off
3. Worktrees accumulate — orphaned worktrees pile up after sessions end
4. JSONL files get bucketed into worktree-specific project folders (e.g., C--Users-Andrew-myrepo--claude-worktrees-random-name-abc123/) instead of the main project folder (C--Users-Andrew-myrepo/)

This is especially confusing when a session is interrupted (crash, "inappropriate use" error, etc.) — the user starts a new session expecting to see their history, but it's in a completely different project folder that Desktop can't resolve.

Expected behavior

An option to run Code sessions directly in the repo's working directory (like the CLI does) without creating a worktree. This would keep all sessions in the same ~/.claude/projects/ folder and preserve conversation history continuity.

Suggested solution

One or more of:

• Setting in claude_desktop_config.json: e.g., "codeSessionWorktrees": false to disable worktree creation
• Per-project option: let users choose "use worktree" vs "use repo directly" when starting a Code session
• Session resume: even with worktrees, allow resuming a previous session in the same worktree instead of always creating a new one

Environment

• Claude Desktop on Windows 11
• Claude Code v2.1.119
• Repo with multiple prior Code sessions, all in separate worktree project folders

Workaround

Using Claude Code CLI (claude in terminal) instead of Desktop's Code tab — the CLI operates directly in the repo directory and maintains conversation continuity.