Claude Code running in a terminal window, showing the agent interface Anthropic
by VibecodedThis

Claude Code Gets a Safe Mode and In-Session Directory Switching in v2.1.169

Anthropic shipped Claude Code v2.1.169 on June 8 with 30 changes. The headliners are a --safe-mode flag for troubleshooting without customizations and a /cd command that lets you move a session to a different directory without clearing the prompt cache.

Share

Anthropic pushed Claude Code v2.1.169 on June 8 with 30 changes. Two new features stand out: a --safe-mode startup flag and a /cd slash command for changing directories mid-session.

Safe Mode

The --safe-mode flag (or the CLAUDE_CODE_SAFE_MODE environment variable) launches Claude Code with all customizations stripped out. That means no CLAUDE.md files, no plugins, no skills, no hooks, and no MCP servers. It’s a clean-slate troubleshooting mode for when you suspect a plugin or custom config is causing unexpected behavior and you want to reproduce the problem without any project-level modifications in play.

There’s also a companion setting, disableBundledSkills, that hides bundled skills and built-in slash commands from the model entirely.

Directory Switching Without Cache Busting

The /cd command solves a workflow friction that gets annoying on larger projects: if you’re working in one directory and need to hand off to a different part of the repository, changing the working directory used to break the prompt cache. The new command moves the active session to a different directory while keeping the cache intact.

Windows Reliability

Five of the bug fixes target Windows specifically. The claude -p flag was hanging on Windows during slash-command scanning. Git Credential Manager popups were appearing at startup. The agents view had display issues when running under WSL inside Windows Terminal. Plugin cache invalidation wasn’t working correctly, and the auto-updater’s retry logic needed work.

Session and Security Fixes

Stale permission prompts were reappearing after session reconnection. Background agents were ignoring project environment variables. Enterprise MCP policies weren’t being enforced on reconnect and fresh installs. And a security fix prevents untrusted project settings from being used to configure OpenTelemetry certificate paths, which could theoretically expose MCP certificate information.

The full list is in the Claude Code changelog and on claudeupdates.dev.

Share