gemini

gemini-cli-plugin-cc

Use Gemini CLI from Claude Code to review code or delegate tasks.

Inspired by codex-plugin-cc — this plugin brings the same cross-model workflow to Gemini.

Features

• Code review — Get Gemini's perspective on your code with /gemini:review
• Adversarial review — Stress-test specific assumptions with /gemini:adversarial-review
• Task delegation — Hand off work to Gemini in the background with /gemini:delegate
• Job management — Track and retrieve results with /gemini:status, /gemini:result, /gemini:cancel

Prerequisites

• Claude Code installed
• Gemini CLI installed and authenticated:

  npm install -g @google/gemini-cli
  gemini auth login
  

• Node.js 18.18 or later

Installation

From the marketplace

/plugin marketplace add Szetty/gemini-cli-plugin-cc
/plugin install gemini@gemini-cli-plugin-cc

For development

git clone https://github.com/Szetty/gemini-cli-plugin-cc.git
cd gemini-cli-plugin-cc
npm install
claude --plugin-dir .

Usage

Verify setup

/gemini:setup

Code review

/gemini:review
/gemini:review security
/gemini:review performance src/server.js

Adversarial review

/gemini:adversarial-review find race conditions
/gemini:adversarial-review break the auth flow src/auth/

Delegate a task

/gemini:delegate investigate why the tests are failing
/gemini:status
/gemini:result 1

Cancel a background job

/gemini:cancel 1

MCP Tools

The plugin exposes 7 tools via MCP:

Tool	Description
gemini_setup	Verify Gemini CLI installation and authentication
gemini_review	Run a code review (model: pro/flash/flash-lite)
gemini_adversarial_review	Steerable adversarial review
gemini_delegate	Delegate a task to run in the background
gemini_status	Check background job status
gemini_result	Retrieve completed job output
gemini_cancel	Cancel a running background job

How it works

The plugin wraps the gemini CLI in an MCP server. Synchronous tools (review, adversarial review) run gemini -p <prompt> --output-format json and return the result. The delegation tools spawn gemini as a detached background process and track jobs via a local registry in the system temp directory.

Configuration

The plugin respects standard Gemini CLI configuration:

• Model selection: pro, flash, flash-lite — omit to use Gemini's default
• Approval mode: default, auto_edit, or yolo (for delegation)

Vertex AI users: the model aliases pro, flash, and flash-lite are resolved by the Gemini CLI and may not be available in all Vertex AI projects. If you see a ModelNotFoundError, pass an explicit model ID instead, e.g. /gemini:review with model gemini-2.5-flash.

License

Apache-2.0