<div align="center">

skill-gemini

Gemini inside Claude Code. Second opinions, web-grounded facts, diagrams, brainstorms — one skill.

Different model family = different blind spots. Use both.

</div>

---

Why

Claude is great. But a single model has blind spots. Gemini sees differently — different training, different reasoning patterns, different knowledge cutoffs. This skill gives Claude direct access to Gemini for everything: quick questions, critical reviews, real-time web facts, image generation, code visualization, and structured multi-round brainstorms.

One install. Five capabilities.

<div align="center"> <img src="assets/gemini-capabilities.png" alt="5 capabilities: Ask, Ground, Image, Visualize, Brainstorm" width="800" /> </div>

---

Quick Install

Claude Code (recommended):

/plugin marketplace add awrshift/skill-gemini

Manual:

mkdir -p .claude/skills/gemini/scripts .claude/skills/gemini/references
curl -sL https://raw.githubusercontent.com/awrshift/skill-gemini/main/SKILL.md \
  -o .claude/skills/gemini/SKILL.md
curl -sL https://raw.githubusercontent.com/awrshift/skill-gemini/main/scripts/gemini.py \
  -o .claude/skills/gemini/scripts/gemini.py

Setup (2 minutes)

1. Get API key at aistudio.google.com → Create API Key
2. Add to .env: GOOGLE_API_KEY=your_key_here
3. Install: pip install google-genai

That's it. The skill auto-discovers gemini.py from its bundle.

5 Capabilities

1. Ask — Text, Reasoning, Review

| You say | What happens | |---------|-------------| | "Ask Gemini about X" | Quick answer via Flash (fast, cheap) | | "Get a second opinion" | Critical review via Pro (deep reasoning) | | "Think deeply about X" | Maximum reasoning via Pro (high thinking) | | "Review this code" | Code review with focus areas |

python3 gemini.py ask "Quick question"
python3 gemini.py second-opinion "Is this right?" --context "details"
python3 gemini.py think "Design a caching strategy"
python3 gemini.py review "def process(data): ..." --focus "error handling"

2. Ground — Web-Verified Facts

Gemini searches Google before responding. Real-time facts, not training data.

python3 gemini.py ask "Latest Next.js version?" --grounded
python3 gemini.py second-opinion "Is Clerk still free?" --grounded

3. Image — Generate Visuals

Banners, diagrams, illustrations — via Gemini Nano Banana models.

"Generate a dark banner for my GitHub repo"
"Create an architecture diagram"

4. Visualize — Code to Diagrams

Three-phase pipeline: read your codebase → synthesize nodes/edges → render via Gemini Image.

"Visualize the architecture of this project"
"Draw the data flow"
"Show me how these services connect"

Supports: system-map, flow, connections, data-model. Full guide in references/visualize.md.

5. Brainstorm — 3-Round Adversarial Dialogue

Claude orchestrates. Flash researches facts. Pro reasons on them. 7 ideas → 3 survivors → 1 winner.

"Brainstorm how to launch this"
"Let's think through options with Gemini"

Full protocol with prompt templates in references/brainstorm.md. Cost: ~$0.35 per brainstorm.

Models (March 2026)

| Model | Price (in/out) | Best for | |-------|---------------|----------| | Pro (gemini-3.1-pro-preview) | $2 / $12 per 1M | Reasoning, brainstorm, second opinions | | Flash (gemini-3-flash-preview) | $0.50 / $3 per 1M | General tasks, code review (default) | | Flash-Lite (gemini-3.1-flash-lite-preview) | $0.25 / $1.50 per 1M | Research, grounding, fact-check |

Gotchas

• google-genai not google-generativeai — the new SDK package name (2026)
• Temperature 1.0 is mandatory — lower values cause looping in Gemini 3
• --grounded is a tool, not a parameter — adds Google Search capability
• Image generation requires Gemini MCP server — mcp__gemini-sdk__gemini_image
• Second opinion ≠ final answer — Claude should always evaluate Gemini's output critically

Works With

| Platform | Install | |----------|---------| | Claude Code | /plugin marketplace add awrshift/skill-gemini | | Codex CLI | Copy skills/gemini/ to .openai/skills/ | | Gemini CLI | Copy skills/gemini/ to .gemini/skills/ | | Any Agent Skills tool | Follow spec |

Part of the AWRSHIFT Ecosystem

• skill-awrshift — adaptive decision-making framework
• skill-brainstorm — standalone 3-round brainstorm (also bundled here)
• skill-telegram — Telegram user API integration
• ClawClaw Soul — persistent identity protocol for AI agents

Contributing

1. Fork the repository
2. Create a feature branch
3. Submit a pull request

License

MIT — see LICENSE for details.

---

<div align="center"> <em>Different model. Different blind spots. Better decisions.</em> </div>