The canonical home for CodeRabbit's agent-native skills and plugin packaging. Use it to install AI-powered code review into 35+ coding agents, Gemini CLI, Antigravity CLI, Claude Code, Cursor, and other supported agent environments.
CodeRabbit detects bugs, security issues, and quality risks before you merge.
MCP review pilot: this branch prepares a plugin with a pinned native
CodeRabbit CLI dependency and a hosted MCP connection. The intended flow is install the
plugin → connect CodeRabbit once → “Review my changes.” There is no separate
CLI login or persistent local daemon. MCP authorizes one review; the pinned CLI
runs its existing collection, submission and review stream. The installed package
fixes its launcher; an unavailable hook requires setup, not a shell fallback. Use complete CLI findings when visible; get_code_review
retrieves the same saved result for missing/incomplete output and later sessions.
The hook variant targets Claude Code through the plugin-owned MCP tool name. Codex hook assembly is blocked until the host can establish plugin server identity; the published standalone Codex plugin remains unchanged. Separately assembled direct packages require their own execution and stdin qualification. Hook trust, local execution, and network permissions still apply. A hosted connector alone cannot collect local files.
This is not published. The source tree alone is not an installable MCP package: choose bundled native files for a manual pilot or a small package that prepares the pinned CLI during setup, with a verified hosted endpoint, using the packaging instructions. Keep the released installation paths below separate from this pilot; they are not fallback instructions for the MCP review skill.
For account connection, optional setup, review recovery and the explicit existing CLI-only option, read the MCP setup guide. It describes the pilot's current availability separately from the released integrations below.
Released CLI integration: install the CodeRabbit CLI via the CLI docs, then tell your agent: “Review my code.” The review command starts the authentication flow when needed.
These are the released CLI integration paths. For the separate unpublished MCP pilot, use package assembly and installation, connect through MCP browser sign-in, and trust the post-tool hook when using that variant.
Use the CLI docs for the primary install path. They cover Homebrew, the install script, and CLI usage.
Choose the path that matches your coding agent.
For agents that support portable SKILL.md files, use the
skills docs.
npx skills add coderabbitai/skillsInstallation options for the skills installer:
| Flag | Purpose |
|---|---|
-g, --global |
Install to user directory instead of project |
-a, --agent |
Target specific agents (for example claude-code) |
-s, --skill |
Install particular skills by name |
--all |
Install all skills to all agents without prompts |
In Claude Code, install the plugin from the official marketplace:
/plugin marketplace update
/plugin install coderabbit@claude-plugins-official
CodeRabbit is also listed in the Claude directory. Add it from Customize > Plugins > Discover in claude.ai or the Claude desktop app, and it will sync to Claude Code sessions where you're signed in with the same claude.ai account. If you install from both places, Claude Code loads the marketplace copy.
For the full setup flow, see the Claude Code integration guide.
This repository now includes Cursor marketplace metadata in
/.cursor-plugin/plugin.json.
After publication, Cursor marketplace installs use:
/add-plugin coderabbit
For the current recommended setup, see the Cursor integration guide.
Until v1.2.0 is published as the Latest release, Gemini CLI users can install
the native extension from main:
gemini extensions install https://github.com/coderabbitai/skills --ref main
gemini extensions listAfter v1.2.0 is published, the --ref main option can be omitted.
The repository-root gemini-extension.json packages
the portable skills, /coderabbit:review command, and both subagents.
This native extension is separate from the generic skills-installer path above.
Antigravity CLI users can install the native plugin directly from this repository:
agy plugin install https://github.com/coderabbitai/skills
agy plugin listThe repository-root plugin.json manifest packages the skills,
review command, and both subagents for Antigravity CLI's plugin discovery.
Codex users can install the official CodeRabbit plugin by following the Codex app integration guide.
Native Codex metadata in .codex-plugin/plugin.json
references the existing skills/ directory. Marketplace publication remains a
separate step; this source change does not update the live listing.
For an at-a-glance inventory of active and repo-packaged distribution paths, see DISTRIBUTION_CHANNELS.md.
| Path | Purpose |
|---|---|
skills/ |
Portable CodeRabbit skills for agents that support SKILL.md. |
.claude-plugin/ |
Claude Code plugin marketplace metadata. |
commands/ |
Native review commands for Gemini CLI and Antigravity CLI. |
agents/ |
code-reviewer and autofix subagents shipped to Claude Code, Gemini CLI, and Antigravity CLI. |
.cursor-plugin/ |
Cursor marketplace metadata. |
.codex-plugin/ |
Codex metadata referencing the shared skills. |
gemini-extension.json |
Gemini CLI extension manifest. |
plugin.json |
Antigravity CLI plugin manifest. |
assets/ |
Shared marketplace and brand assets. |
DISTRIBUTION_CHANNELS.md |
Maintainer inventory of live, packaged, and in-development channels. |
Once installed, just ask your agent:
Review my code
Check for security issues
What's wrong with my changes?
Run a code review
Review my PR
Review the directory at ../my-service
The agent will automatically:
- Check if CodeRabbit CLI is installed
- Run the review on your changes
- Present findings grouped by severity
- Optionally fix issues and re-review
When you ask for a specific review directory, the agent can pass CodeRabbit CLI
--dir <path> after confirming that path is inside an initialized Git working
tree.
CodeRabbit supports 35+ coding agents.
| Agent | Project Path | Global Path |
|---|---|---|
| Amp, Kimi Code CLI | .agents/skills/ |
~/.config/agents/skills/ |
| Antigravity CLI | .agents/skills/ |
~/.gemini/config/skills/ |
| Claude Code | .claude/skills/ |
~/.claude/skills/ |
| Cline | .cline/skills/ |
~/.cline/skills/ |
| CodeBuddy | .codebuddy/skills/ |
~/.codebuddy/skills/ |
| Codex | .codex/skills/ |
~/.codex/skills/ |
| Command Code | .commandcode/skills/ |
~/.commandcode/skills/ |
| Continue | .continue/skills/ |
~/.continue/skills/ |
| Crush | .crush/skills/ |
~/.config/crush/skills/ |
| Cursor | .cursor/skills/ |
~/.cursor/skills/ |
| Droid | .factory/skills/ |
~/.factory/skills/ |
| Gemini CLI | .gemini/skills/ |
~/.gemini/skills/ |
| GitHub Copilot | .github/skills/ |
~/.copilot/skills/ |
| Goose | .goose/skills/ |
~/.config/goose/skills/ |
| Junie | .junie/skills/ |
~/.junie/skills/ |
| Kilo Code | .kilocode/skills/ |
~/.kilocode/skills/ |
| Kiro CLI | .kiro/skills/ |
~/.kiro/skills/ |
| Kode | .kode/skills/ |
~/.kode/skills/ |
| MCPJam | .mcpjam/skills/ |
~/.mcpjam/skills/ |
| Moltbot | skills/ |
~/.moltbot/skills/ |
| Mux | .mux/skills/ |
~/.mux/skills/ |
| Neovate | .neovate/skills/ |
~/.neovate/skills/ |
| OpenClaude IDE | .openclaude/skills/ |
~/.openclaude/skills/ |
| OpenCode | .opencode/skills/ |
~/.config/opencode/skills/ |
| OpenHands | .openhands/skills/ |
~/.openhands/skills/ |
| Pi | .pi/skills/ |
~/.pi/agent/skills/ |
| Pochi | .pochi/skills/ |
~/.pochi/skills/ |
| Qoder | .qoder/skills/ |
~/.qoder/skills/ |
| Qwen Code | .qwen/skills/ |
~/.qwen/skills/ |
| Replit | .agent/skills/ |
N/A (project-only) |
| Roo Code | .roo/skills/ |
~/.roo/skills/ |
| Trae | .trae/skills/ |
~/.trae/skills/ |
| Trae CN | .trae/skills/ |
~/.trae-cn/skills/ |
| Windsurf | .windsurf/skills/ |
~/.codeium/windsurf/skills/ |
| Zencoder | .zencoder/skills/ |
~/.zencoder/skills/ |
AI-powered code review that finds bugs, security issues, and suggests improvements using CodeRabbit.
Use when:
- You want to review code changes before committing or merging
- Checking for bugs, security vulnerabilities, or anti-patterns
- Getting PR feedback or suggestions for improvements
- Running automated code quality checks
Categories covered: Bug detection, security analysis, code quality, performance issues, best practices
Triggers: "review my code", "check for bugs", "security review", "PR feedback", "run coderabbit"
Capabilities:
- Analyzes code changes for bugs, security issues, and anti-patterns
- Preserves finding severities (critical, major, minor, trivial, info, none)
- Supports autonomous fix-review cycles
- Reviews tracked changes by default, with committed and uncommitted scopes
- Supports directory-scoped reviews through CodeRabbit CLI
--dir <path>
Safe fix workflow for unresolved CodeRabbit GitHub PR review threads, with per-issue review and approval.
Use when:
- You already have an open GitHub PR reviewed by CodeRabbit
- You want to apply suggested fixes from unresolved current CodeRabbit review threads
- You want guided fixes with explicit approval for each change
Categories covered: Review-thread extraction, issue prioritization, guarded fixes, consolidated commit and PR summary
Triggers: "coderabbit autofix", "fix coderabbit", "cr fix"
Capabilities:
- Fetches unresolved current CodeRabbit review threads for the current PR
- Parses and prioritizes issues by severity
- Applies fixes only after validating the issue and getting approval
- Produces a single consolidated commit and posts a PR summary comment
Create, update, or validate repository configuration through the official CLI.
Ask “Set up CodeRabbit configuration”, “Make CodeRabbit reviews more thorough”,
or “Validate .coderabbit.yaml”. Where supported, invoke $coderabbit-config or
/coderabbit-config; the Claude plugin uses /coderabbit:coderabbit-config.
The distinct name leaves the host's /config command available.
Requires an official CLI supporting config protocol v2 (verified with 0.9.3). The CLI owns defaults, schema checks, previews, and writes. The skill supports review-style changes and the CLI's guided path settings; unsupported settings remain a CLI capability gap. Missing commands require an upgrade, never an agent-side YAML editor. Validation alone does not change files; saves require approval of the preview. Configuration does not require login or run a review.
This skill is included in the source package. Tagged archives and marketplace caches require their normal release or publication update to deliver it.
The components below describe the released CLI integration. The unpublished MCP pilot instead overlays its review/setup skills and review agent during assembly, bundles a verified native CLI and hosted MCP connection, and includes a post-tool hook when that launcher variant is selected. Hook execution requires host trust; the direct variant requires permitted local execution with separate stdin.
- Skills:
code-review,autofix, andcoderabbit-config. Run them as/coderabbit:code-review,/coderabbit:autofix, and/coderabbit:coderabbit-config, or ask in plain language. - Subagents:
code-reviewerruns a CodeRabbit review in its own context, andautofixcollects and checks unresolved CodeRabbit review threads and returns a fix plan without editing anything. Mention them as@agent-coderabbit:code-reviewerand@agent-coderabbit:autofix. - Manifest:
.claude-plugin/plugin.json
The Claude plugin has no slash commands of its own. In Claude, use
/coderabbit:code-review instead of the earlier
/coderabbit:coderabbit-review.
The plugin runs reviews through the locally installed CodeRabbit CLI, so it needs a shell on a machine where the CLI is installed and logged in. Claude Code has full support. Cowork and Claude chat on the web, desktop, and mobile can load the skills, but a review runs only where the CodeRabbit CLI is available, and chat doesn't run subagents.
- Marketplace manifest:
.cursor-plugin/plugin.json - Skills source:
skills/
For sandboxed CLI sessions that cannot find or execute an installed CodeRabbit CLI, follow the scoped executable-access recovery. It keeps diagnostics sandboxed and retains per-command host approval for auth and review.
- Native plugin manifest:
plugin.json - Skills source:
skills/ - Review command:
/coderabbit:reviewfromcommands/coderabbit/review.toml(converted to a skill during installation) - Subagents:
agents/code-reviewer.md,agents/autofix.md
- Native extension manifest:
gemini-extension.json - Skills source:
skills/ - Slash command:
/coderabbit:review(commands/coderabbit/review.toml) - Subagents:
agents/code-reviewer.md,agents/autofix.md
The source Claude plugin contains three skills and two subagents. It bundles no MCP servers, hooks, or executables.
- Local commands: the
code-reviewskill and thecode-reviewersubagent run the installedcoderabbitCLI and read-only Git commands. Theautofixskill uses the installed GitHub CLI (gh) to read CodeRabbit review threads on the current pull request and, when you approve, to create a pull request or post a summary comment. Theautofixsubagent only reads: it runs read-onlygitandghcommands and never edits, commits, pushes, or posts. - Data sent: when you run a review, the CodeRabbit CLI sends the code
changes under review and the repository context needed to review them to
CodeRabbit's service.
autofixexchanges pull request data with GitHub throughgh. The plugin sends nothing on its own. - Credentials: the CodeRabbit CLI and
ghhandle their own login and store their own credentials. The plugin doesn't read or send credentials. - Policy: CodeRabbit's handling, retention, and sharing of review data is described in the CodeRabbit Privacy Policy. Contact support@coderabbit.ai with questions.
The unpublished MCP pilot uses connector browser sign-in instead of a separate CLI login. Connector OAuth credentials remain with the host; the bundled CLI receives one scoped review handoff through stdin and sends the selected changes and review context to CodeRabbit. MCP can retrieve the same saved findings. Provider credentials remain server-side. A trusted hook or permitted direct launcher starts the bundled executable; neither requires a persistent daemon.
The coderabbit-config skill runs local CLI configuration commands. The CLI
fetches the public configuration schema over HTTPS and validates YAML locally;
it does not submit repository code for review. Approved saves change the local
configuration file, with commit and push left to the requested workflow.
- Documentation
- Support
- Privacy Policy
- Terms of Service
- CodeRabbit Documentation
- Vercel Skills CLI
- Agent Skills Specification
- Gemini CLI Extension Documentation
- Antigravity CLI Plugin Documentation
MIT