v2.1.169

Source: Anthropic Claude Code Date: 2026-06-08 URL: https://github.com/anthropics/claude-code/releases/tag/v2.1.169

Summary

Claude Code v2.1.169 (released June 8, 2026) is a substantial maintenance and hardening release — 30+ changes covering new flags, bug fixes, and UX improvements. Major additions: --safe-mode / CLAUDE_CODE_SAFE_MODE to disable all customizations (CLAUDE.md, plugins, skills, hooks, MCP) for troubleshooting; /cd command to move working directory without breaking prompt cache; disableBundledSkills setting to strip bundled commands from model context. The bulk of the release addresses enterprise-reliability gaps: MCP policy enforcement on reconnect, stale permission prompts reappearing in remote sessions, background agents ignoring project-level env settings, and plugin lock file accumulation.

Implications

• Coding agents / dev tooling thread: --safe-mode is a diagnostic primitive that signals the extension surface (plugins, hooks, MCP) is complex enough to need a clean-room escape hatch — useful signal about how heavily customized production deployments are becoming.
• Enterprise hardening arc: The MCP policy enforcement fixes (allowedMcpServers/deniedMcpServers not enforcing on reconnect or first-session) and the untrusted-settings security fix (OTEL cert path injection) are meaningful trust/supply-chain signals; enterprise orgs are deploying Claude Code at scale and the attack surface is expanding accordingly.
• Background agent maturity: Multiple fixes to background agent behavior (env settings, flag preservation across retire→wake, worktree awareness) show the agent-orchestration layer is under active stress from real production use.
• Prompt cache discipline: The /cd command preserving cache across directory changes is a small but telling optimization — it shows teams are actively managing cache hit rates, not just treating the context window as free.