gemini-thinking

Gemini Thinking

Consult Google's Gemini model for a second opinion on any task. Run the script, pass your question, and incorporate the response into your own thinking.

Before You Start

Check if GEMINI_API_KEY is set in the environment. If not, ask the user to configure it:

1. Get a Gemini API key from https://aistudio.google.com/apikey
2. Add to ~/.zshrc: export GEMINI_API_KEY=your_key_here
3. Run source ~/.zshrc

Script

python3 scripts/gemini_think.py "Your question here" [--model MODEL] [--file FILE] [--system INSTRUCTION] [--thinking-budget N]

Options

Option	Values
--model	gemini-3-pro-preview (default, strongest reasoning), gemini-3-flash-preview (fast, balanced)
--file	Path to a file to include as context
--system	System instruction to guide the model's behavior
--thinking-budget	Max thinking tokens, 0-8192 (default: 8192 = maximum reasoning depth)

When and How to Use

Use this as a thinking tool, not a replacement for your own judgment. Good patterns:

• Debugging: Stuck on a bug? Describe the symptoms and code, ask Gemini what it thinks.
• Architecture: Weighing two approaches? Describe both and ask for trade-off analysis.
• Code review: Pass a file with --file and ask for a review.
• Sanity check: "Does this logic handle edge cases correctly?" with context.
• Brainstorming: "What are alternative approaches to solve X?"

Example Usage

# Quick question (uses Gemini 3 Pro with max thinking by default)
python3 scripts/gemini_think.py "What are the trade-offs between REST and GraphQL for a read-heavy internal API?"

# Code review with file context
python3 scripts/gemini_think.py --file src/auth.py "Review this auth module for security issues"

# Fast model for simple questions
python3 scripts/gemini_think.py --model gemini-3-flash-preview "Is it better to use a Set or Array for deduplication in JavaScript?"

# Lower thinking budget for quick answers
python3 scripts/gemini_think.py --thinking-budget 1024 "What's the difference between a mutex and a semaphore?"

# With system instruction for focused feedback
python3 scripts/gemini_think.py --system "You are a senior security engineer" --file config.yaml "Audit this configuration for security concerns"

Guidelines

• Default to gemini-3-pro-preview for complex reasoning. Use gemini-3-flash-preview for quick factual questions.
• Thinking budget is set to 8192 (max) by default for deepest reasoning. Lower with --thinking-budget 1024 for faster, cheaper responses.
• Always include relevant context — Gemini can't see the codebase, so pass code via --file or inline in the prompt.
• Present Gemini's response as a second perspective, not as authoritative. Synthesize it with your own analysis.
• Don't over-use — reserve for genuinely ambiguous or challenging decisions, not routine tasks.