ClaudeAI Coding Agents

Code Review plugin for Claude Code: review, how it works and whether it is worth it (2026)

Before you start

  • 8 min read

Boxes marked Run in terminal are what you type; plain text explains why.

What you get in one minute#

The code-review plugin is one prompt file that makes Claude review a GitHub pull request with five parallel reviewers and post one comment; it is a niche tool now, because Claude Code has a built-in /code-review that does more.

Claude Code logo

Logo: the official Claude Code logo from Anthropic's press kit (anthropic.com/press-kit)

The marketplace page shows 438,525 installs (October 2026).

VerdictDetail
Useful?Niche. A readable PR-comment recipe; the built-in /code-review covers most needs.
Best forTeams that want a CLAUDE.md-aware review posted as one PR comment through gh.
Skip it ifYou review local branches, or need a reliable unattended CI gate.
Made byAnthropic (README names Boris Cherny), Apache 2.0 licence.
Version + dateREADME says 1.0.0, but plugin.json has no version, so it installs as "unknown". Folder last changed 20 February 2026.
Install command/plugin install code-review@claude-plugins-official
Token costNo published figure. Our count: about 1,900 tokens when invoked, then the sub-agents do the spending.

What it does#

You give it a pull request, and it decides whether to review it, checks it against your CLAUDE.md files and posts the findings as one comment. It adds one slash command and nothing else.

  • 1 command: /code-review:code-review <PR number or URL> (the namespaced name; a bare /code-review is the built-in skill).
  • 0 skills, 0 agents, 0 hooks, 0 MCP servers. The "agents" are sub-agents the prompt asks Claude to start.
  • Needs the GitHub CLI gh, logged in.
  • Skips closed, draft, trivial and already-reviewed pull requests.

How it works in the background#

One 92-line prompt is the whole program: Claude follows its steps and launches cheap and strong sub-agents in turn. The picture follows one review.

How the code-review plugin works: you run the command with a pull request, Claude Code loads the command file, a Haiku agent checks eligibility, two Haiku agents list CLAUDE.md files and summarise the change, five Sonnet agents review in parallel, one Haiku agent scores each issue from 0 to 100, issues scoring 80 or more lead to a final eligibility check and a pull request comment, otherwise nothing is posted

  • Step 5 reviewers: CLAUDE.md compliance, a shallow bug scan, git blame history, comments on earlier PRs, and code comments.
  • Nothing forces the steps. Claude may skip some; see the complaints below.

Under the hood: dissection#

The plugin is four files and no code that runs by itself; the only logic is a Markdown prompt.

The file tree#

Codetext
code-review/
├── .claude-plugin/
│   └── plugin.json            name, description, author (no version)
├── commands/
│   └── code-review.md         the prompt: steps, rubric, comment format
├── LICENSE                    Apache 2.0
└── README.md                  says 4 agents; the prompt has 5

What each file is#

FileLanguageWhat it doesWhen it runs
plugin.jsonJSONName, description, authorAt install and load
commands/code-review.mdMarkdown promptSteps 1 to 8, confidence rubric, comment templateWhen you type the command
README.mdMarkdownDescribes the pluginNever loaded by Claude

Key excerpts#

commands/code-review.md, the permission line:

File contenttext
allowed-tools: Bash(gh issue view:*), Bash(gh search:*), Bash(gh issue list:*),
Bash(gh pr comment:*), Bash(gh pr diff:*), Bash(gh pr view:*), Bash(gh pr list:*)

The same file, the filter step:

File contenttext
6. Filter out any issues with a score less than 80.
   If there are no issues that meet this criteria, do not proceed.

Disk and network#

  • Writes: nothing locally. The prompt posts one comment with gh pr comment.
  • Network: only GitHub through gh, plus the model calls.
  • Needs: gh and a GitHub remote. Two other copies exist (anthropics/claude-plugins-public is identical; anthropics/claude-code has a newer four-agent variant with --comment).

A bar chart of file sizes in the code-review plugin: LICENSE 11,358 bytes, the command file 7,422, the README 7,321 and plugin.json 234

Chart: file sizes from the GitHub tree at commit 41f619d, 27 October 2026. The prompt is the only part Claude reads.

What users say#

25 GitHub issues read: 0 praise, 3 mixed, 22 complaints, but the split is skewed. Issue trackers collect problems, and many reports come from unattended CI runs. Reddit was not read. The Hacker News comments we found were about the built-in command or Anthropic's managed service, so they are not counted.

A bar chart of how the 25 code-review plugin user reports lean: 0 praise, 3 mixed feature requests and 22 complaints, from GitHub issues between February and October 2026

Chart: our count of the 25 reports, October 2026. Not a satisfaction rate.

A horizontal bar chart of the 22 complaints by theme: 7 about reviews silently skipped or empty in CI, 4 about version, metadata or README mismatch, 4 about the prompt and docs contradicting themselves, 3 about permissions and unsafe git actions, 2 about confusion with the built-in command and 2 about runtime errors

Chart: themes of the 22 complaints, labelled by DIY Lab Hub, October 2026.

VoiceWhat they say
Silent CI runsA reporter (October 2026): after a Claude Code update the reviewers were skipped, yet the "No issues found" comment still posted.
CostThe same report: $1 to $24 and several minutes per full run, $0.08 to $0.37 when agents were skipped.
Two commandsA user (June 2026): the plugin and the built-in command share a name and confuse.
Stale docsThe README (July 2026) was confirmed out of date by a maintainer for the claude-code copy.
PermissionsA security note (August 2026): the gh pr comment grant is broader than needed.
WishesGitLab support (February 2026) and per-finding threads (May 2026).

Pros: why it is useful#

It is a small, readable recipe for a multi-agent PR review, and it respects your CLAUDE.md.

  • Cheap gatekeeping: Haiku does the eligibility and scoring steps; Sonnet does the review.
  • Fewer nitpicks: the prompt lists false positives to drop (linter issues, old code, style).
  • Traceable output: every finding links to code with a full SHA.
  • Easy to inspect and fork: 92 lines, Apache 2.0.

Cons: why it is not useful#

It overlaps with a stronger built-in command, and its own prompt has known holes.

  • Built-in wins on features: /code-review runs on any plan, reviews local branches, takes effort levels, --fix and --comment.
  • Threshold maths: the rubric scores 0, 25, 50, 75, 100 but the filter drops below 80, so a 75 never shows (issue, May 2026).
  • Re-reviews never happen: "already reviewed" has no commit check (issue, September 2026).
  • Silent failures in CI and a "version unknown" update notice.

Should you consider it? When it is useful#

Consider it for one case only: you want a CLAUDE.md-aware review posted to a GitHub PR from the terminal or a workflow. Otherwise use the built-in /code-review.

A decision flowchart for the code-review plugin: if you review your own branch locally use the built-in /code-review; otherwise if you do not need a GitHub PR comment from gh, skip it; if you do, install it when you run it by hand, and install it only with a check of the logs when you run it unattended in CI

SituationVerdict
Review your own branchBuilt-in /code-review
One PR comment via gh, run by handInstall
Unattended CI gateOnly with log checks
Every PR in a Team or Enterprise organisationLook at Anthropic's managed Code Review service

Install and check it#

You install it from Anthropic's official marketplace with one command, and gh must be logged in.

Linux#

Install gh from cli.github.com and run gh auth login. Then, inside Claude Code:

Codetext
/plugin install code-review@claude-plugins-official

Or from your shell:

Run in terminalbash
claude plugin install code-review@claude-plugins-official

Windows (PowerShell)#

Run in terminalpowershell
winget install --id GitHub.cli
gh auth login
claude plugin install code-review@claude-plugins-official

Mac (Terminal)#

Run in terminalbash
brew install gh
gh auth login
claude plugin install code-review@claude-plugins-official

Check it worked#

Run in terminalbash
claude plugin list
claude plugin details code-review

Type / and look for /code-review:code-review. Run it on a small test PR first.

Things to keep in mind#

  • Security. It runs no hooks or scripts, but the command pre-approves gh pr comment with any flags, and it posts publicly on your PR.
  • Cost. Each run starts at least 9 sub-agents plus one scorer per issue; watch /usage.
  • Updates. Auto-update is on for the official marketplace; the "unknown" version can repeat "plugin updated" notices.
  • Your repository: it only reads and comments, but one reporter saw Claude stash work and switch branch (issue, September 2026); commit first.
  • Organisations can restrict marketplaces.

Troubleshooting#

ProblemSolution
Error: Agent type 'haiku' not foundOlder Claude Code builds (March 2026 report); update Claude Code.
/plugin isn't available in this environment.You are in claude -p; install from your shell.
Green CI run, no comment postedEligibility step skipped it; read the full output, run with a fresh PR.
"No issues found" but no agents ranKnown report (October 2026); check the transcript.
Repeated "code-review plugin updated"Missing version field (issue); harmless.
gh errorsInstall gh and run gh auth login.

FAQ#

Is this the same as /code-review? No. The bare /code-review is a built-in skill; the plugin is /code-review:code-review.

Can I change the 80 threshold? Only by editing the command file in the plugin cache, and updates overwrite it.

Does it work with GitLab? Not in this copy; the built-in --comment supports GitLab merge requests.

Sources#

Not run by DIY Lab Hub: we read the files and docs but did not install the plugin on a test machine.

Checked against the official docs on 2026-10-27