gemini-prompting

name	gemini-prompting
description	Guidelines for composing effective Gemini prompts from Claude Code
user-invocable	false

Gemini Prompting

Use this skill inside gemini:gemini-dispatch to shape prompts before forwarding.

When to use Gemini vs Claude

Signal	Use Gemini	Use Claude
Input > 100K tokens	✅ 1M context	❌ may lose details
Full codebase review	✅ sees everything	❌ needs chunking
Cross-model validation	✅ different training	—
Code generation	❌	✅ stronger coding
File editing	❌ no edit tools	✅ native tools
Quick tasks	❌ higher latency	✅ faster

Prompt guidelines

• Be explicit about the output format you want.
• For code review: include the full diff, not a summary.
• For analysis: include the full file content when possible (Gemini handles it).
• For research: ask for citations and evidence.
• Keep prompts structured with clear sections.

Model selection

• Default (no --model): Gemini picks its best available model.
• For maximum context: --model gemini-2.5-pro (1M tokens).
• For speed: --model gemini-2.5-flash.

Only add --model when the user explicitly asks or when the task clearly needs the Pro model's full context window.