Do I need to install anything to use /code-review?

No. It ships bundled with the Claude Code CLI — update Claude Code and type /code-review in a session. There is no marketplace install and no claude.ai Settings > Skills upload for it.

Is /code-review the same as the Claude Code Review GitHub App?

No. The skill reviews a diff locally in your terminal on any plan. The managed Code Review product is a separate research-preview GitHub App for Team and Enterprise orgs that posts inline PR comments from Anthropic infrastructure and averages $15–25 per review.

Can it fix what it finds?

Yes — /code-review --fix applies the findings to your working tree after the review. Edits from a background review sit outside checkpoints, so /rewind will not revert them; use git if you want them gone.

What does /code-review ultra do?

It escalates to ultrareview, a deeper multi-agent review run in the cloud against your repo's default branch. It requires a claude.ai account, may bill usage credits, can post findings to a github.com PR with --post, and is not available on Bedrock, Vertex, Foundry, or ZDR orgs — where it falls back to a local review.

Skill review · Anthropic official

Claude Code Review Skill

The claude code review skill this search points to is /code-review, the review skill Anthropic bundles into the Claude Code CLI itself. Type the command and it reviews your branch's commits ahead of upstream plus any uncommitted changes, or whatever target you pass — a file path, a PR number, a branch name, or a ref range like main...my-feature. The one thing to know before you go hunting for an install step: there is none. It ships with the CLI, which separates it from two similarly named Anthropic products — an older `code-review` plugin in the anthropics/claude-code plugin marketplace, and the paid "Code Review" GitHub App, a research preview for Team and Enterprise plans that averages $15–25 per review. The bundled skill runs as a background subagent, takes effort levels from low through max, escalates to a cloud "ultrareview" with `ultra`, and can apply its own findings with --fix. Community alternatives exist; most people need none of them.

View the sourcelicense unclear · 143,657★ · facts checked September 1, 2026

By Robert MacKelfresh

Founder, Automation Squad · September 1, 2026

Install

1Install or update Claude Code — npm install -g @anthropic-ai/claude-code (or the native installer). The skill ships bundled with the CLI; there is no marketplace install and no Settings > Skills upload.
2In a session inside a repo with changes, run /code-review — optionally with an effort level (/code-review high) or a target (/code-review 123, /code-review main...my-feature).
3Add flags as needed: --fix to apply findings to your working tree, --comment to post them as inline PR comments on a PR target.
4For the deeper cloud review, run /code-review ultra — requires claude.ai authentication and may bill usage credits after a confirmation dialog.
5Optional: to stop Claude and scheduled tasks from launching it on their own, add {"skillOverrides": {"code-review": "user-invocable-only"}} to ~/.claude/settings.json.

What it does

Local diff review: Reviews your branch's commits ahead of upstream plus uncommitted changes for correctness bugs and reuse/simplification/efficiency cleanups; pass a file path, PR number, branch name, or ref range (main...my-feature) to review something else. /review has been an alias since v2.1.223.
Effort levels: low and medium report only high-confidence findings; high through max broaden coverage and may include less-certain ones. With no level typed, it reuses the last level you typed — even from an earlier session.
Flags: --fix applies the findings to your working tree after the review; --comment posts findings as inline PR comments; --post (v2.1.227+) preselects posting an ultra cloud review's findings to a github.com PR as a comment from your account.
Background execution: Runs as a background subagent with its own context window so it doesn't fill your conversation; findings arrive when it completes. Runs in the foreground in -p/non-interactive mode or when a review is already in progress.
Cloud escalation: /code-review ultra launches the deeper multi-agent ultrareview on Anthropic infrastructure, scoped to your branch against the repo's default branch plus uncommitted changes; requires claude.ai authentication.

Common questions

Do I need to install anything to use /code-review?: No. It ships bundled with the Claude Code CLI — update Claude Code and type /code-review in a session. There is no marketplace install and no claude.ai Settings > Skills upload for it.
Is /code-review the same as the Claude Code Review GitHub App?: No. The skill reviews a diff locally in your terminal on any plan. The managed Code Review product is a separate research-preview GitHub App for Team and Enterprise orgs that posts inline PR comments from Anthropic infrastructure and averages $15–25 per review.
Can it fix what it finds?: Yes — /code-review --fix applies the findings to your working tree after the review. Edits from a background review sit outside checkpoints, so /rewind will not revert them; use git if you want them gone.
What does /code-review ultra do?: It escalates to ultrareview, a deeper multi-agent review run in the cloud against your repo's default branch. It requires a claude.ai account, may bill usage credits, can post findings to a github.com PR with --post, and is not available on Bedrock, Vertex, Foundry, or ZDR orgs — where it falls back to a local review.

Before you install

Three Anthropic things share the name: the bundled /code-review skill, an older code-review plugin (v1.0.0, by Boris Cherny) in the anthropics/claude-code plugin marketplace, and the paid managed "Code Review" GitHub App. It is easy to read docs for one while running another.
The bundled skill's prompt/source is not published. The anthropics/claude-code repo (143k stars) is the issue tracker plus example plugins, and it carries no open-source license (GitHub reports none).
The managed Code Review product is a research preview, Team/Enterprise only, averages $15–25 per review billed as usage credits, and is unavailable to Zero Data Retention orgs.
Local /code-review reads CLAUDE.md but not REVIEW.md — REVIEW.md customization only applies to the managed GitHub App product.
--fix edits from a background review land outside session checkpoints, so /rewind cannot undo them; you have to revert with git.
ultra is unavailable on Amazon Bedrock, Google Cloud's Agent Platform, Microsoft Foundry, and ZDR orgs — it silently falls back to a local review.
Behavior has churned across versions: the command was /simplify before v2.1.147, /review only became an alias in v2.1.223, and the remembered-effort-level behavior is recent — most third-party writeups are stale on at least one of these.
At high through max effort the review may include findings it is less sure about, i.e. more false positives; reviews also consume normal token usage.

Facts come from the skill's own repository and docs on the date above, via the GitHub API — not from memory, and never from a sponsor.

Worth comparing

code-review plugin (official marketplace) — Anthropic's older 4-parallel-agent PR reviewer with an 80-point confidence filter; install via /plugin marketplace add anthropics/claude-code, then /plugin install code-review@claude-code-plugins.
awesome-skills/code-review-skill — Community skill (MIT, ~1.9k stars, pushed Aug 2026): 20+ language/framework review-guideline library loaded progressively; install with npx skills add awesome-skills/code-review-skill.
pr-review-toolkit plugin — Also in Anthropic's official plugin marketplace; a set of specialized PR-review agents for teams wanting more granular review commands.

← All MCP servers & skills