Gemini

Gemini plugin for Claude Code

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

What You Get

• /gemini:review — read-only code review against your local git state
• /gemini:adversarial-review — steerable review that challenges implementation and design choices
• /gemini:rescue — delegate investigation, bug fixes, or implementation tasks to Gemini
• /gemini:status — check active and recent Gemini jobs
• /gemini:result — retrieve the stored output of a finished job
• /gemini:cancel — cancel an active background job
• /gemini:setup — verify installation and configure options

Requirements

• Node.js 18.18 or later
• Gemini CLI (npm install -g @google/gemini-cli) and a Google account, or a GOOGLE_API_KEY for direct SDK access

Install

/plugin marketplace add gustavoBonze/gemini-plugin-cc
/plugin install gemini@gustavobonze-gemini
/reload-plugins
/gemini:setup

If Gemini CLI is missing and npm is available, /gemini:setup will offer to install it automatically.

To authenticate manually, run gemini once and complete the sign-in flow, or set:

export GOOGLE_API_KEY=your-key

Usage

/gemini:review

Runs a Gemini review on your current work. Review-only — no fixes applied.

/gemini:review
/gemini:review --base main
/gemini:review --scope branch --base main
/gemini:review --background
/gemini:review --wait --effort high
/gemini:review --thinking --thinking-budget 2048
/gemini:review --image screenshot.png

Flags:

Flag	Description
--wait	Run in foreground without asking
--background	Run as background job without asking
--base <ref>	Compare current branch against <ref> (e.g. main)
--scope auto|working-tree|branch	Review scope (default: auto)
--effort <level>	Effort level: none, minimal, low, medium, high, xhigh
--depth low|medium|high	Depth of analysis (default: medium)
--model <alias|name>	Model override (see Models)
--thinking	Enable reasoning tokens
--thinking-budget <n>	Reasoning token budget (0 = off, positive = limit)
--image <path>	Include an image in the review context (repeatable)
--cwd <dir>	Run from a different directory
--prompt-file <path>	Load review focus text from a file

---

/gemini:adversarial-review

Same as /gemini:review but frames the review adversarially — questions design decisions, tradeoffs, and assumptions. Accepts an optional focus text.

/gemini:adversarial-review
/gemini:adversarial-review --base main challenge the caching strategy
/gemini:adversarial-review --background look for race conditions
/gemini:adversarial-review --effort xhigh question the auth flow

Supports all the same flags as /gemini:review, plus free-form focus text at the end.

---

/gemini:rescue

Delegates a task to Gemini through the gemini:gemini-rescue subagent. Use this for investigation, bug fixes, implementation, or research.

/gemini:rescue investigate why the tests are 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 --fresh start a new thread and plan the refactor
/gemini:rescue --background investigate the flaky integration test
/gemini:rescue --write implement the feature described in the issue
/gemini:rescue --effort high --thinking diagnose the memory leak
/gemini:rescue --image diagram.png explain what this architecture diagram means

Flags:

Flag	Description
--wait	Run in foreground
--background	Run as background job
--resume	Continue the last Gemini thread from this session
--fresh	Force a new thread
--write	Allow Gemini to write files (default: read-only)
--effort <level>	Effort level (same levels as review)
--depth low|medium|high	Analysis depth
--model <alias|name>	Model override
--thinking	Enable reasoning tokens
--thinking-budget <n>	Reasoning token budget
--image <path>	Include an image in the context (repeatable)
--cwd <dir>	Run from a different directory
--prompt-file <path>	Load task text from a file

---

/gemini:status

Shows active and recent Gemini jobs for the current repository. Outputs a table with actionable follow-up commands.

/gemini:status
/gemini:status job-abc123
/gemini:status --all
/gemini:status --wait job-abc123
/gemini:status --json

Flags: [job-id], --wait, --timeout-ms <ms>, --poll-interval-ms <ms>, --all, --json

---

/gemini:result

Shows the stored final output for a finished job.

/gemini:result
/gemini:result job-abc123
/gemini:result --json

---

/gemini:cancel

Cancels an active background job.

/gemini:cancel
/gemini:cancel job-abc123

---

/gemini:setup

Checks installation and authentication. Offers to install Gemini CLI if missing.

/gemini:setup
/gemini:setup --enable-review-gate
/gemini:setup --disable-review-gate

Models

The default model is gemini-3.1-pro-preview. Override with --model:

/gemini:review --model flash
/gemini:rescue --model pro investigate the bug

Alias	Model	Best for
flash	gemini-2.5-flash	Fast, lightweight tasks
pro	gemini-3.1-pro-preview	Deep analysis (default)

Effort Levels

--effort selects a model automatically based on the required intensity:

Level	Model
none, minimal, low	gemini-2.5-flash
medium, high, xhigh	gemini-3.1-pro-preview

--model takes priority over --effort, and --effort takes priority over --depth.

Multimodal

Include images in reviews or rescue tasks with --image. Requires GOOGLE_API_KEY (SDK mode).

/gemini:review --image screenshot.png
/gemini:rescue --image before.png --image after.png explain the visual diff

Supported formats: PNG, JPEG, WEBP, GIF, BMP.

SDK Mode

When GOOGLE_API_KEY is set, the plugin calls the Gemini API directly via the @google/genai SDK instead of spawning a Gemini CLI process for every call. This eliminates startup overhead and keeps a persistent client across calls. The SDK is required for multimodal (--image) and thinking (--thinking-budget) features.

export GOOGLE_API_KEY=your-key

/gemini:setup reports whether the plugin is running in SDK mode or direct CLI mode.

Thinking Mode

Enable Gemini's reasoning tokens for deeper analysis:

/gemini:review --thinking
/gemini:rescue --thinking --thinking-budget 4096 debug this race condition

--thinking activates reasoning. --thinking-budget <n> sets the token limit (0 disables thinking, positive values set the budget). Requires SDK mode.

Review Gate

Automatically run a Gemini review before Claude finalizes each response:

/gemini:setup --enable-review-gate
/gemini:setup --disable-review-gate

When enabled, a Stop hook intercepts Claude's response and runs a focused Gemini review. If issues are found, Claude is asked to address them before finishing.

License

Apache-2.0 — © 2026 Gustavo Bonze