Claude Code용 Gemini 플러그인

Use Google's Gemini CLI from inside Claude Code to review code or delegate tasks. Why bring Gemini into Claude Code? Gemini 2.5 Pro offers a 1M-token context window, a distinct reasoning style, and strong code analysis — making it a useful second opinion alongside Claude. Instead of switching tools, you get both models collaborating in the same session: Claude drives, Gemini reviews or investigates, results come back inline. Origin: This plugin is a port of codex-plugin-cc, adapted from OpenAI's Codex App Server Protocol to Google's ACP (Agent Client Protocol). See Differences from codex-plugin-cc for details. | Command | Purpose | |---|---| /gemini:review | Read-only Gemini code review | /gemini:adversarial-review | Steerable challenge review | /gemini:rescue | Delegate a task to Gemini | /gemini:status | List active and recent jobs | /gemini:result | Show full output for a finished job | /gemini:cancel | Cancel an active background job | /gemini:setup | Check install, auth, and toggle review gate | - Node.js 18.18 or later - Google account or Gemini API key - Sign in with Google (free tier: 60 req/min, 1,000 req/day) or set GEMINI_API_KEY from AI Studio. - Sign in with Google (free tier: 60 req/min, 1,000 req/day) or set Steps 1–2 run in your terminal. Steps 3–4 run inside a Claude Code session. claude plugin marketplace add sakibsadmanshajib/gemini-plugin-cc claude plugin install gemini@google-gemini /reload-plugins /gemini:setup If Gemini CLI is not installed, the plugin will offer to install it for you (npm install -g @google/gemini-cli ). If Gemini CLI is installed but not authenticated, run !gemini in Claude Code to authenticate interactively, or set GEMINI_API_KEY in your environment. Runs a Gemini review on your current work. Note: Code review especially for multi-file changes might take a while. It's generally recommended to run it in the background. Use it when you want: - a review of your current uncommitted changes - a review of your branch compared to a base branch like main Use --base for branch review. It also supports --wait and --background . It is not steerable and does not take custom focus text. Use /gemini:adversarial-review when you want to challenge a specific decision or risk area. Examples: /gemini:review /gemini:review --base main /gemini:review --background This command is read-only and will not perform any changes. When run in the background you can use /gemini:status to check on the progress and /gemini:cancel to cancel the ongoing task. Runs a steerable review that questions the chosen implementation and design. It can be used to pressure-test assumptions, tradeoffs, failure modes, and whether a different approach would have been safer or simpler. It uses the same review target selection as /gemini:review , including --base for branch review. It also supports --wait and --background . Unlike /gemini:review , it can take extra focus text after the flags. Use it when you want: - a review before shipping that challenges the direction, not just the code details - review focused on design choices, tradeoffs, hidden assumptions, and alternative approaches - pressure-testing around specific risk areas like auth, data loss, rollback, race conditions, or reliability Examples: /gemini:adversarial-review /gemini:adversarial-review --base main challenge whether this was the right caching and retry design /gemini:adversarial-review --background look for race conditions and question the chosen approach This command is read-only. It does not fix code. Hands a task to Gemini through the gemini:gemini-rescue subagent. Use it when you want Gemini to: - investigate a bug - try a fix - continue a previous Gemini task - take a faster or cheaper pass with a smaller model Note: Depending on the task and the model you choose these tasks might take a long time and it's generally recommended to force the task to be in the background or move the agent to the background. It supports --background , --wait , --resume , and --fresh . If you omit --resume and --fresh , the plugin can offer to continue the latest rescue thread for this repo. Examples: /gemini:rescue investigate why the tests started failing /gemini:rescue fix the failing test with the smallest safe patch /gemini:rescue --resume apply the top fix from the last run /gemini:rescue --model pro investigate the flaky integration test /gemini:rescue --model flash fix the issue quickly /gemini:rescue --background investigate the regression You can also just ask for a task to be delegated to Gemini: Ask Gemini to redesign the database connection to be more resilient. Notes: - if you do not pass --model , Gemini chooses its own defaults - model aliases: pro (gemini-2.5-pro),flash (gemini-2.5-flash),flash-lite (gemini-2.5-flash-lite) - you can also pass concrete model names like gemini-3-pro-preview - follow-up rescue requests can continue the latest Gemini task in the repo Lists active and recent Gemini jobs for this repository. /gemini:status /gemini:status /gemini:status --wait Shows the full stored output for a finished job. /gemini:result /gemini:result Cancels an active background job. /gemini:cancel /gemini:cancel Checks whether Gemini is installed and authenticated. If Gemini is missing and npm is available, it can offer to install Gemini for you. You can also use /gemini:setup to manage the optional review gate. /gemini:setup --enable-review-gate /gemini:setup --disable-review-gate When the review gate is enabled, the plugin uses a Stop hook to run a targeted Gemini review based on Claude's response. If that review finds issues, the stop is blocked so Claude can address them first. Warning: The review gate can create a long-running Claude/Gemini loop and may drain usage limits quickly. Only enable it when you plan to actively monitor the session. /gemini:review /gemini:rescue investigate why the build is failing in CI /gemini:rescue --background redesign the error handling across the API layer /gemini:status This plugin is a port of codex-plugin-cc, which wraps OpenAI's Codex CLI. The two plugins share the same command interface and plugin structure, but differ in how they communicate with their respective AI backends. | Aspect | codex-plugin-cc | gemini-plugin-cc | |---|---|---| | Backend CLI | codex (OpenAI Codex CLI) | gemini (Google Gemini CLI) | | Protocol | App Server Protocol (ASP) — HTTP REST with SSE streaming | Agent Client Protocol (ACP) — JSON-RPC 2.0 over stdio | | Connection | HTTP server (codex --app-server ) | Persistent broker over Unix socket (gemini --acp ) | | Session management | Thread-based (thread/start , thread/cancel ) | Session-based (session/new , session/set_mode ) | | Write control | sandbox: "workspace-write" vs "read-only" | approvalMode: "auto_edit" vs "default" | | Model effort | --effort parameter (none → xhigh) | Not available via ACP (use --model instead) | | Streaming | SSE events from HTTP endpoint | JSON-RPC notifications over stdio | - Same commands: Both plugins expose identical slash commands ( review ,adversarial-review ,rescue ,status ,result ,cancel ,setup ). - Same review logic: Diff collection, untracked file reading, branch comparison, and prompt construction are shared. - Different transport: Codex uses an HTTP app server with SSE streaming. Gemini uses a JSON-RPC broker over Unix sockets. The broker keeps a persistent gemini --acp child process alive for the session. - No effort parameter: Codex supports --effort to control thinking budget. Gemini CLI does not expose an equivalent via ACP, so this plugin uses--model selection instead. - Authentication: Codex uses ChatGPT accounts or OpenAI API keys. Gemini uses Google accounts or Gemini API keys from AI Studio. The codex plugin's architecture (command definitions, job tracking, state persistence, background workers, review prompt construction) is protocol-agnostic. Porting it to Gemini required replacing only the transport layer (acp-client.mjs , acp-broker.mjs ) and the prompt execution functions in gemini.mjs , while keeping everything else intact. The plugin communicates with Gemini CLI via ACP (Agent Client Protocol) — a JSON-RPC 2.0 interface over stdio. A persistent broker process keeps the connection alive across multiple commands within a Claude Code session. If you want to change the default model or settings, configure them in your Gemini settings file: User-level: ~/.gemini/settings.json Project-level: .gemini/settings.json (overrides user settings) Your configuration will be picked up based on: - user-level config in ~/.gemini/settings.json - project-level overrides in .gemini/settings.json Check out the Gemini CLI docs for more configuration options. | Method | Setup | Best For | Tested | |---|---|---|---| | Sign in with Google | gemini (interactive) | Desktop use | Yes | | Gemini API Key | export GEMINI_API_KEY=... | CI/headless | No | | Vertex AI | export GOOGLE_CLOUD_PROJECT=... | Enterprise | No | Delegated tasks and any review gate runs can be directly resumed inside Gemini by running gemini --resume with the session ID from /gemini:result or /gemini:status . Claude Code ──[Bash]──> gemini-companion.mjs ──[Unix socket]──> ACP Broker | gemini --acp (persistent) - gemini-companion.mjs — Main CLI handling all subcommands - acp-broker.mjs — Persistent daemon multiplexing JSON-RPC requests via Unix socket - acp-client.mjs — Client with broker-first, direct-spawn fallback - lib modules — Git context, state persistence, job tracking, rendering If you are already signed into Gemini on this machine, that account should work immediately. This plugin uses your local Gemini CLI authentication. If you only use Claude Code today and have not used Gemini yet, you will also need to authenticate. The free tier (Sign in with Google) gives you 60 requests per minute and 1,000 per day. Set GEMINI_API_KEY for headless use, or run !gemini inside Claude Code to authenticate interactively. The plugin starts Gemini in ACP mode (gemini --acp ) and communicates via JSON-RPC. A broker process keeps the connection alive for the duration of your Claude Code session and is automatically cleaned up when the session ends. Yes. The plugin inherits your ~/.gemini/settings.json and any project-level .gemini/settings.json overrides. Yes. Because the plugin uses your local Gemini CLI, your existing authentication method and config still apply. If you use Vertex AI, ensure GOOGLE_CLOUD_PROJECT and GOOGLE_CLOUD_LOCATION are set. Current version: v1.0.0 — tested on Linux/macOS with Google OAuth only. Windows is untested. | Area | Status | |---|---| Core commands (review , rescue , status , result , cancel ) | Working | | Background jobs + broker persistence | Working | Review gate (/gemini:setup --enable-review-gate ) | Working — see warning in docs | Scope validation (--scope flag) | Known bug #1 — falls through to default silently | | Protocol method mismatch edge cases | Known bug #2 | | Windows (Native) and macOS | Untested | | WSL and Linux | Tested | GEMINI_API_KEY auth | Untested | | Vertex AI auth | Untested | Requirements reminder: Gemini CLI (@google/gemini-cli ) must be installed and authenticated separately — this plugin is a bridge, not a bundled runtime.