Getting Started with Claude Code /code-review | /simplify Deprecated in v2.1.148

From /simplify to /code-review — Background and Scope of the Rename

On May 21, 2026, at 20:39 UTC (05:39 JST on May 22, Japan time), Anthropic released Claude Code v2.1.147, renaming the /simplify command — which had long been used as a general-purpose cleanup tool — to /code-review. The primary sources are the v2.1.147 release notes on GitHub and the official Anthropic changelog, both of which explicitly note that the command rename was accompanied by the addition of effort level options and the --comment option. The intent behind the rename was to clarify that this is a dedicated code review command and to lay the groundwork for upcoming PR integration features. The "formatting for readability" aspect of what /simplify used to do has been removed from the new command's responsibilities, meaning anyone who regularly used /simplify for code formatting purposes must switch to another tool.

The Cleanup Behavior Bundled with the Old /simplify Has Been Removed

As its name implied, the old /simplify handled "simplification," and in addition to code review, it included light lint-style formatting tasks such as removing unused imports and tidying up conditional expressions. Starting with v2.1.147, /code-review is designed to focus exclusively on review, with formatting delegated to dedicated tools like prettier, eslint --fix, or ruff format. Teams that had been skipping a pre-review formatting step on the assumption that "running /simplify will also handle formatting" will need to introduce a separate format step in their CI pipeline.

Scope of Impact — Wholesale Replacement of Documentation, Training Materials, and Integration Workflows

Assets that referenced /simplify span a wide range: internal wikis, onboarding training materials, READMEs, CI scripts, Makefiles, and notes in Pull Request templates, among others. Even a day after the release, search engines still show a large volume of old /simplify documentation, creating a real risk that new employees who find those pages will get stuck on "command not found" errors. Internally, it is a good idea to distribute the grep examples from the final section of this article for bulk text replacement, and teams with public-facing articles about the command should consider annotating that published content at the same time.

Basic Usage of /code-review — Specifying Effort Levels and Targets

/code-review has the same interface in both interactive mode and headless mode. In interactive mode, simply typing /code-review in the Claude Code chat will run a review against the currently staged changes in the repository. In headless mode, you can call it directly from the shell as claude --code-review, and the exit code and standard output can be handled by CI. The effort level is a concept that did not exist in the old /simplify, allowing you to switch the strictness of the review and the amount of token consumption across three tiers. Applying the same level of strictness to both critical PRs and minor fixes is not cost-effective, and explicit level specification translates directly into standardizing your team's workflow.

When to Use Each Effort Level (high / medium / low)

/code-review high is intended for branches immediately before a release, changes with a wide blast radius such as payment processing or user authentication, and PRs that include breaking changes to publicly-facing APIs. Claude will perform a deep read including test case coverage, edge cases, and implicit side effects that are being depended upon. /code-review medium (the default when no level is specified) is suited for typical feature development, covering conformance to functional specifications and detection of common bug patterns in a balanced way. /code-review low is applied to changes with a localized scope — such as fixing string literals, documentation diffs, or renaming for refactoring purposes — to minimize review time and token consumption.

Specifying Targets — By File, Commit Range, or PR

When called without arguments, /code-review targets the "uncommitted changes on the current branch," but you can pass file paths directly to restrict the review to that scope. To specify a commit range, you can use Git's rev-list syntax directly, such as /code-review HEAD~3..HEAD. To run a review on a per-PR basis, the most practical operational pattern is to check out the branch locally with gh pr checkout <number> and then run claude --code-review --comment. The output includes severity markers (critical / major / minor / nit) and recommended actions; by treating critical and major findings as CI failures, you can structurally prevent review blind spots.

Posting Inline Comments to GitHub PRs with the --comment Option

The --comment option added in v2.1.147 posts the results of /code-review directly as inline comments on a GitHub PR. Previously, the only practical options for connecting Claude Code review results to a PR were receiving them via standard output and manually copy-pasting them, or pasting the whole result into the PR body. With --comment, CI can now generate a state where "Claude's comments remain on the relevant lines." Having review results tied to specific lines of code means you won't lose context when you come back to review the PR later.

Prerequisites — GitHub Authentication and Permission Scopes

As a prerequisite for --comment, the CI execution environment must either have gh auth login completed, or a token with the necessary permissions to post PR comments must be provided in the GH_TOKEN environment variable. For GitHub Actions, make sure to explicitly declare in the job definition that GITHUB_TOKEN has pull-requests: write scope for the repository. For GitLab CI, Jenkins, or CircleCI, the standard approach is to store a bot user's Personal Access Token as a CI secret variable and export it as GH_TOKEN when the job starts.

Minimum Configuration for Calling from CI

The minimum configuration can be broken down into the following steps:

1. Restrict the CI job trigger condition to "PR opened or synchronized" (running a review on every push will cause token consumption to balloon).
2. Inside the job, check out the target PR's branch and complete the Claude Code setup (e.g., npm install -g @anthropic-ai/claude-code).
3. Run claude --code-review --comment --pr <PR number> (use --pr to explicitly specify the target PR).
4. Evaluate the exit code and fail the CI step if there are any critical findings, blocking the merge.

Granularity of Inline Comments and Boundaries with Existing Bots

Comments posted via --comment are tied to individual lines or diff hunks in principle, with an overall summary of the entire file consolidated into the PR body summary. When running alongside automated reviews from reviewdog or Renovate, a practical division of labor is to have Claude handle "design and logic review" while reviewdog handles "mechanical detection of lint/format issues." Having both leave comments on the same line degrades visibility in the PR, so a manageable operational configuration is to filter Claude's --comment output to only surface critical findings by effort level. Once CI integration is up and running, the new normal will be "PR review comments from Claude waiting for you when you wake up," which directly reduces the lead time spent waiting for reviews.

v2.1.148 Hotfix — Immediately Resolving the Bash Exit Code 127 Regression

Less than half a day after the v2.1.147 release, at 01:16 UTC on May 22, 2026 (10:16 JST), Anthropic published Claude Code v2.1.148 as an emergency hotfix. The primary source is the v2.1.148 release notes on GitHub, which explicitly states that this addresses a regression where "all commands run via the Bash tool return exit code 127." Exit code 127 is the shell's code for "command not found," and it occurs when PATH resolution is broken. In v2.1.147, there was a condition under which the Bash tool could not properly inherit the user's environment variables, resulting in a state where even basic commands like ls or git were being rejected as "not found." The roughly five-hour response time demonstrates how quickly Anthropic identified the problem, conveying a sense of urgency about the breadth of the impact.

How to Determine If You Are Affected

The check is two-step. First, check the output of claude --version — if it shows 2.1.147, you are affected; if it shows 2.1.148 or later, the fix has been applied. Next, in interactive mode, send a prompt that calls the Bash tool (e.g., "Run echo hello in Bash") and check whether the result contains exit code 127 or command not found. If both conditions are true, there is a high probability that a large number of integration workflows and CI jobs are currently failing, so the decision to immediately update and re-run past jobs in parallel is warranted.

Safe Upgrade Procedure

The update method differs depending on how Claude Code was installed:

1. Installed via npm: Run npm install -g @anthropic-ai/claude-code@latest and confirm you are on 2.1.148 or later with claude --version.
2. Environments with auto-update enabled: The latest version should be automatically applied on startup, but confirm the version with claude --version after launching to be safe.
3. Environments with auto-update disabled (e.g., enterprise audit requirements): Manually run the same command after reviewing the release notes. If you have an internal mirror with a pinned version, update the mirror at the same time.
4. Installed in a CI image: You will need to rebuild the base image and update the pinned version tag. If your Dockerfile has a version pinned like claude-code@2.1.147, update it to v2.1.148.

Following Up on CI and Integration Runs During the Affected Period

If you ran /code-review from CI at any point between installing v2.1.147 and upgrading to v2.1.148, all review findings that depended on the Bash tool (e.g., findings based on the results of running test commands) will have been treated as failures. The safe approach is to grep your CI job history to list all review results from the affected period and add them to the re-run queue. Reviews that do not call the Bash tool (static analysis only) are not affected by this regression, so searching the job logs by keyword to identify the scope of the impact will let you minimize the number of jobs you need to re-run.

Migration Checklist — Updating Internal Documentation, Integration Workflows, and Training Materials

The work of replacing /simplify with /code-review involves a mix of parts that can be handled mechanically with grep and parts that require reading the context and rewriting by hand. The most efficient approach is to tackle the mechanical replacements first, then fill in the rest manually. The table below organizes the typical replacement points by asset type.

grep Examples for a Full Replacement

The following are typical grep examples that work in local repositories and internal knowledge management tools. For broader searches, using git grep or ripgrep (rg) will be faster.

1. Full-text search within the repository: Run git grep -n "/simplify" to list matching lines with line numbers.
2. Count occurrences by file with ripgrep: Run rg -c "/simplify" to identify files with the highest revision cost.
3. Limit to shell scripts: Use rg -t sh "claude\\s+/simplify" to extract command calls in CI scripts.
4. Limit to documentation formats: Use rg -t md "/simplify" to bulk-extract internal Markdown files.
5. Run the replacement (with caution): Run git ls-files | xargs sed -i 's:/simplify:/code-review:g', but be sure to review the diff afterward, as this may accidentally match code comments or intentional uses of the word in English text.

Decision Criteria for Staying on a Pinned Old Version

In organizations where audit requirements mandate an approval process for updating external tools, an immediate update to v2.1.148 may not be physically feasible. In such cases, the decision criteria are: first, "are we running Bash-tool-dependent reviews from CI?"; second, "what is the availability SLA for the affected systems?"; and third, "can the team absorb the risk of being on v2.1.147 while waiting for approval?" If Bash-tool-dependent reviews are running, prioritize getting the update approved. If they are not, the blast radius is limited even with a standard approval process. Also check whether your internal guidelines include a clause that permits an expedited approval track for security updates.

Related Trends — New Commands and Concurrent Major Feature Expansion

Around the same time as the /code-review rename, two videos were published in the Claude Code space: a deep-dive video on the /goal command (published 2026-05-22, 2,127 views / YouTube) and a video discussing Claude Code's expansion into a major Agentic OS (published the same day, 1,770 views / YouTube). The direction in which Claude Code is evolving has clearly shifted from "one-off code generation" toward "an agentic foundation that encompasses goal setting and review," and reading the /code-review rename in that context makes its positioning easier to understand. For a broad overview of the continuous release cycle, uravation.com's five-consecutive-weeks release guide is a helpful resource, and for a Japanese-language summary of Claude Code updates, AI Souken's Claude Code Update Roundup is worth referencing.

Sources

• Claude Code v2.1.148 Release Notes (GitHub) — Hotfix details for the Bash exit code 127 regression
• Claude Code v2.1.147 Release Notes (GitHub) — /simplify → /code-review rename, effort level options, --comment addition
• Official Anthropic Changelog — Full update history for Claude Code
• AI Souken — Claude Code Update Explainer — Japanese-language summary (up to v2.1.147)
• uravation — Claude Code v2.1.101 to 30 Releases: A 5-Week Guide — Overview of the continuous release cycle
• YouTube — Claude Code /goal Deep Dive — Evidence of demand for new commands
• YouTube — Claude Code as an Agentic OS — Context of the concurrent major feature expansion