Anthropic shipped Claude Code 2.1.147 on May 21, 2026 with a renamed and reworked code review command. /simplify is gone. In its place, /code-review reports correctness bugs at a chosen effort level and can post findings as inline GitHub PR comments. A regression in the same release broke the Bash tool, and a hotfix 2.1.148 landed at 01:16 UTC on May 22.

How to Integrate Into Your PR Workflow

Update to 2.1.148 first. The 2.1.147 bash regression made every shell command return exit code 127 on some setups, so skip straight to the patch. Then run claude, open the repo you want reviewed, and type one of the following:

  1. /code-review for a default-effort scan of the current diff. Claude reads the working tree, identifies correctness bugs, and prints them inline in the terminal.
  2. /code-review high for deeper analysis. Claude spends more tokens, follows more call paths, and surfaces edge cases the default effort can miss.
  3. /code-review --comment to push findings as inline GitHub PR comments via the GitHub REST API, scoped to the lines Claude flagged.

The old /simplify cleanup-and-fix behavior is removed, not folded into the new command. If you want auto-fixes, ask Claude in a chat turn after reviewing the findings.

Why It Matters

The configurable-effort knob is the interesting part. Until now, code review inside Claude Code was a fixed-cost pass. Teams that wanted thoroughness on a critical PR had no way to ask Claude to spend more compute on the diff. Now /code-review high gives reviewers a single-keystroke way to scale token spend with review stakes, the same model already used for /effort. Pairing that with --comment means correctness bugs land where the rest of your code review lives, in the PR thread, rather than as terminal output a reviewer has to copy by hand. Anthropic's Claude Code docs show the slash command appears in the new effort-tier UI alongside /plan and /explain.

Key Details

The release also pinned background sessions so Ctrl+T agents stay alive when idle, are restarted in place to apply Claude Code updates, and are shed only after non-pinned sessions under memory pressure. That fixes the long-running pain of background agents getting reaped mid-task. Other notable fixes from the 35-item changelog include a corrected CLAUDE_CODE_SUBAGENT_MODEL propagation to teammate processes, hook conditions like PowerShell(git push*) now matching correctly, and pasted text reaching agents as content instead of a [Pasted text #N] placeholder.

What to Do Next

Run claude --version, confirm you are on 2.1.148, and rebind any muscle memory from /simplify to /code-review. If you maintain a hook config that referenced /simplify, update it. And if you have not yet wired Claude Code into your PR review process, this release is the cleanest entry point yet, especially when combined with self-hosted sandboxes for repos you cannot send to Anthropic's cloud.