Enable the Smarter /init Experiment to Get Interviewed, Not Overwritten

By default, running /init rewrites your CLAUDE.md and AGENTS.md from scratch — which is fine for new projects but destructive for ones you've been refining for months. There's an experimental flag that changes this behaviour entirely.

Add this to your ~/.claude/settings.json:

{
  "CLAUDE_CODE_NEW_INIT": "1"
}

With this flag enabled, /init no longer bulldozes your existing files. Instead, it reads what you already have and interviews you — asking about your project's current goals, gaps, and pain points. It then suggests targeted improvements: new skills to add, hooks to wire up, or context sections that are missing.

This is especially useful on mature projects where your CLAUDE.md has evolved over time. Rather than starting from a blank template, Claude builds on what's already working and proposes specific additions.

To try it: enable the flag, run /init in your project, answer the questions, and review the suggestions before accepting any changes. Run /context afterward to verify the final state of what's loaded.

Let Claude improve your setup instead of replacing it.

via @vinhnx