Isolate config bugs in Claude Code with --safe-mode, which strips CLAUDE.md, hooks, skills, and MCP

Digital Applied·medium signal

The new --safe-mode flag (June 2026) runs Claude Code with every customization removed — CLAUDE.md, hooks, skills, and MCP servers — to give you a clean diagnostic baseline when something misbehaves, so you can quickly tell whether your own config is the culprit. Admin-managed policy settings still apply, so enterprise security rules aren't bypassed during troubleshooting. It's the fastest first step when an agent starts acting strangely after a config change.