Claude Code: /code-review Now Runs as a Background Subagent
Claude CodeClaude Code changed /code-review to execute as a background subagent instead of running inline, so review work no longer fills up the active conversation. The review now targets whatever slash commands are currently stacked rather than blocking the main thread, letting developers keep working while a review runs behind the scenes. This continues a string of recent releases reworking how review commands behave and consume resources.
Key Takeaways
/code-reviewnow runs as a background subagent, freeing up the main conversation instead of filling it with inline review output.- The review's target is now the currently stacked slash commands, a change from reviewing the conversation state at the time of invocation.
- Developers can keep working while a review executes, checking results later in the agents view rather than waiting for it to finish inline.
- This follows two other 2026 changes to the same command: a 25% token-usage cut (late June) and the switch from automatic to manual invocation (mid-July).
- The same release fixed a data-loss risk: pressing the left arrow key after editing a message could previously discard the conversation with no undo, it now requires confirmation.
- The pattern across recent releases shows Anthropic deliberately trading automatic behavior for explicit, cheaper, less disruptive review workflows.
Sources & Mentions
1 external resource covering this update
Reviews That No Longer Interrupt
/code-review has, until now, run inline: invoking it filled the active conversation with the review's output and tool calls, forcing a developer to wait it out before continuing other work. Claude Code 2.1.218 changes this by running /code-review as a background subagent instead.
What Changes in Practice
Because the review now runs in the background, it no longer occupies the main conversation thread. The command's review target is now the set of currently stacked slash commands, rather than the conversation state at invocation time. Developers can keep working, issuing further commands, editing files, or starting another task, while the review executes concurrently, then return to the agents view to see its findings once ready.
Part of a Broader Pattern
This is the latest in a run of changes to how Claude Code's review tooling behaves. A late-June release cut /code-review's token usage by roughly 25% by consolidating five separate cleanup-finder passes into one. A mid-July release stopped /verify and /code-review from running automatically, requiring explicit invocation. Moving /code-review to a background subagent is the next step in that same direction: making review work cheaper, less intrusive, and more under the developer's explicit control.
Shipped Alongside
The same 2.1.218 release added screen-reader announcements for deleted text in --ax-screen-reader mode (covering word and line deletions via Option+Delete, Ctrl+W, Cmd+Backspace, Ctrl+U, and Ctrl+K), and fixed a bug where pressing the left arrow key right after editing a message could discard the conversation with no way to undo it, that action now asks for confirmation first.