Claude Code v2.1.218: /code-review Now Runs as a Background Subagent, /deep-research Goes Manual

Anthropic released Claude Code v2.1.218 on July 22, 2026, the successor to version v2.1.217, with a set of changes affecting everyday work with this CLI coding agent — Anthropic’s command-line tool that carries out programming tasks with the help of models and tools.

Why does /code-review now run in the background?

As of this version, the /code-review command runs as a background subagent — a helper process that executes separately from the main conversation. This means the user can keep working or start another task while the code review runs in the background, instead of waiting for it to finish before the next step.

/deep-research moves to manual invocation

Unlike previous behavior, /deep-research in v2.1.218 no longer triggers automatically, but only when the user explicitly invokes it. In addition, claude mcp list and /mcp now show the HTTP status or error message when an MCP server (the protocol connecting tools and data sources to the model) fails to connect, and skill/plugin frontmatter booleans now accept yes/no, on/off, and 1/0 values. Server-managed settings for benign toggles no longer require approval, while agent markdown filenames may no longer contain a colon.

What are the key bug fixes?

Windows paths with the \u prefix no longer get corrupted into CJK characters, which resolves the issue of incorrectly displayed paths. Pressing the left arrow key no longer deletes an entire conversation without user confirmation, and pasting multi-line text no longer collapses into a single line. Usage measurement for Bedrock ARNs through the gateway was also fixed, and screen reader announcements were added for text deletion.

Conclusion

Version v2.1.218 builds on v2.1.217 with a shift toward background code-review execution, a stricter manual approach to deep research, and a range of fixes that directly affect the reliability of everyday work — from Windows paths to MCP diagnostics.