Claude Code v2.1.218: /code-review Moves to Background, /deep-research Goes Manual

TL;DR: v2.1.218 (released today) runs /code-review as a background subagent — your conversation stays clean while the review runs — and deep-research no longer auto-launches without explicit invocation.

Two behavioral changes that affect daily workflows:

1. /code-review → background subagent. Running /code-review now delegates to a background subagent. The review no longer fills your conversation window; Claude's main context stays clean; and you can continue your next task while the review runs. The result surfaces when the subagent completes. The review target is the stacked slash command's changes.

2. /deep-research → manual-only. Claude previously launched /deep-research automatically when it judged a question warranted deep investigation — triggering a long, expensive session you didn't explicitly request. That auto-launch is gone; you must invoke it explicitly.

Other highlights: Plan mode auto no longer prompts you to approve Bash commands the static analyzer can prove are read-only. claude mcp list and /mcp now show HTTP status and error text when a server fails to connect. Fixed: Windows paths with \u-prefixed segments were being corrupted into CJK characters in tool inputs. Fixed: left arrow key was silently discarding the conversation with no undo.

Why it matters: If you were piping /code-review output anywhere, verify your pipeline still works — the result is now delivered via the background subagent model, not as conversation text.

Sources: Claude Code v2.1.218 release notes · Claude Code changelog · Claude Updates — claudeupdates.dev