Gemini

---

Why This Plugin?

Claude Code is great. Gemini is great. This plugin lets you use both without leaving your terminal.

• Get a second opinion on your code from Gemini before shipping
• Challenge your design with adversarial review — pressure-test assumptions, tradeoffs, and failure modes
• Delegate tasks to Gemini and keep working in Claude Code while it runs in the background

Quick Start

Prerequisites

• Gemini CLI — npm install -g @google/gemini-cli
• Node.js 18.18+

Install

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

First Run

/gemini:review                    # review uncommitted changes
/gemini:review --base main        # review branch diff against main
/gemini:adversarial-review        # challenge your design decisions
/gemini:rescue investigate the bug  # hand a task to Gemini

Commands

Command	Description
/gemini:review	Structured code review (read-only)
/gemini:adversarial-review	Steerable challenge review targeting design and tradeoffs
/gemini:rescue	Delegate a task to Gemini via subagent
/gemini:setup	Check readiness, manage review gate
/gemini:status	Show running and recent jobs
/gemini:result	Show output of a finished job
/gemini:cancel	Cancel an active background job

All commands support --background, --wait, and -m <model> (aliases: auto, pro, flash, flash-lite).

Code Review

/gemini:review                          # working tree changes
/gemini:review --base main              # branch diff
/gemini:review --background             # run in background

Adversarial Review

Goes beyond code correctness — challenges design choices, hidden assumptions, and alternative approaches.

/gemini:adversarial-review
/gemini:adversarial-review --base main challenge the caching design
/gemini:adversarial-review --background look for race conditions

Task Delegation

/gemini:rescue investigate why tests are failing
/gemini:rescue -m flash fix the flaky integration test
/gemini:rescue --background redesign the connection pool

Check progress with /gemini:status, get results with /gemini:result, cancel with /gemini:cancel.

Review Gate (Optional)

When enabled, a Stop hook triggers a targeted Gemini review on each Claude response. If issues are found, the stop is blocked so Claude can address them first.

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

Warning: This can create a long-running Claude/Gemini loop and may drain usage limits quickly. Only enable when actively monitoring.

Architecture

Thin-forwarder pattern — no broker, no daemon:

Slash command → gemini-companion.mjs → lib/gemini.mjs → gemini -o stream-json (stdin)

• File-lock concurrency instead of a broker process
• Three-layer JSON extraction: prompt engineering → JSON block → plain-text fallback
• Activity-based timeout: resets on each stream event, 30min hard ceiling
• Full Windows support: shell spawning, UNC paths, taskkill, EAGAIN-safe reads

FAQ

Do I need a separate account? No. The plugin uses your local Gemini CLI authentication. Run gemini interactively to sign in, or set GEMINI_API_KEY.

Can I choose a different model? Yes. Pass -m <model> to any command. If a model is unavailable, the plugin suggests alternatives.

Community

This project is shared with the LINUX DO community.

Acknowledgments

Built on the architecture of openai/codex-plugin-cc. Key adaptations: direct Gemini CLI invocation via file-lock concurrency, three-layer JSON extraction for free-form output, and comprehensive Windows compatibility fixes.

Contributing

See CONTRIBUTING.md for development setup, code style, and pull request guidelines.

License

Apache License 2.0