Gemini Plugin for Claude Code
A Claude Code plugin that lets you run Google Gemini from slash commands, with a Codex-style companion runtime.
What it does
- Multimodal analysis: send text plus local image files to Gemini.
- Bulk content processing: run large prompt payloads or long-form analysis tasks through the companion runtime.
- Second-opinion review: run code/strategy review from your git working tree or branch diff.
- Rescue delegation: route implementation/investigation work to a thin Gemini rescue subagent.
- Background jobs: start long Gemini runs and check them later with status/result commands.
- Screenshot analysis: inspect one or more screenshots with Gemini vision input.
Available commands
/gemini:setup- Verify runtime and
GEMINI_API_KEY, optionally set default model. - Example:
/gemini:setup --model gemini-2.5-pro /gemini:analyze- Run general multimodal tasks.
- Example:
/gemini:analyze --image ./tmp/screen.png "Find UX issues in this flow" /gemini:review- Send git diff context to Gemini for a second-opinion review.
- Example:
/gemini:review --base main --scope src "Focus on regressions" /gemini:rescue- Delegate a larger task to the
gemini-rescuesubagent. - Example:
/gemini:rescue --model gemini-2.5-pro "Refactor this module and explain tradeoffs" /gemini:screenshot- Analyze screenshots/images directly.
- Example:
/gemini:screenshot --image ./tmp/ui.png "Identify layout and accessibility issues" /gemini:status- Check active/recent jobs and inspect a specific job.
- Example:
/gemini:statusor/gemini:status <job-id> --wait
Installation
From Claude Code:
1. Add local marketplace/repo source:
/plugin marketplace add /Users/gabrielc/CND-OS/gemini-plugin-cc
2. Install plugin:
/plugin install gemini
Configuration
Set GEMINI_API_KEY in your project .env.local (or shell environment used by Claude Code):
GEMINI_API_KEY=your_key_here
Then run:
/gemini:setup
Models
gemini-2.5-flash(default, free-tier friendly)gemini-2.5-progemini-3.1-pro-preview
You can override per command:
/gemini:analyze --model gemini-2.5-pro "Deep analysis prompt"




