Claude Code v2.1.169: --safe-mode for clean-room debugging, /cd without cache loss

Claude Code v2.1.169 (June 8) is a quality-of-life release with three features worth adopting.

--safe-mode (or CLAUDE_CODE_SAFE_MODE) launches with all customizations disabled — CLAUDE.md, plugins, skills, hooks, MCP servers. This is the bisect tool the ecosystem needed: when behavior goes weird, you can now answer "is it my config or the tool?" in one restart instead of manually unwinding settings.

/cd changes the session's working directory mid-session without breaking the prompt cache — useful for monorepo hopping where restarting previously meant paying the full cache-warm cost again. And disableBundledSkills hides bundled skills, workflows, and built-in slash commands from the model entirely — relevant if you run a tightly curated context budget and don't want Anthropic's defaults competing with your own skills.

Also notable in the 30-change list: enterprise managed MCP policies (allowedMcpServers/deniedMcpServers) were not being enforced through several code paths (reconnection, --mcp-config, first session after install) — fixed, so if you rely on those policies for governance, this is a security-relevant upgrade, not optional. A Windows regression that made claude -p hang is fixed, the "CLAUDE.md is too long" warning now scales with the model's context window, and CPU usage during streaming is down.

Sources: DevelopersIO v2.1.169 breakdown, Claude Code changelog