Google Gemini CLI Offloading

You have access to the Gemini CLI (gemini) for offloading tasks. Gemini runs Google's latest model with sandbox mode (no file system access) and provides fast, high-quality responses for code analysis tasks. It's already authenticated.

When to Use Gemini vs Ollama vs Claude

Decision flow:

1. Can Ollama handle it? (simple, offline-capable) → Use Ollama
2. Needs better reasoning or current knowledge? → Use Gemini
3. Needs tools, context, or highest accuracy? → Use Claude

How to Use

# Check availability
.claude/skills/gemini/gemini.sh status

# Quick question
.claude/skills/gemini/gemini.sh query "How do Stripe destination charges work with application fees?"

# Summarize a file
.claude/skills/gemini/gemini.sh summarize server/services/payout.ts

# Review code
.claude/skills/gemini/gemini.sh review components/shared/captcha.tsx

# Analyze a diff
git diff --staged | .claude/skills/gemini/gemini.sh diff

# Generate test stubs
.claude/skills/gemini/gemini.sh test-gen server/services/founding-clinic.ts

Important Notes

• Sandbox mode (-s): Gemini runs sandboxed — it cannot read/write local files. All file content must be passed via the prompt. This is intentional for security.
• Auth may expire. If Gemini fails with auth errors, the user needs to run gemini interactively to re-authenticate. If unavailable, fall back to Ollama or Claude silently.
• Rate limits apply. Google has API rate limits. Don't send rapid-fire batch requests. Space requests or use Ollama for batch file summarization.
• Gemini sees project context by default (reads GEMINI.md / local files). Sandbox mode prevents tool execution but the CLI may still provide context from the project.
• Response time: ~3-8s per request (network latency). Slower than warm Ollama but faster than cold Ollama.