Gemini CLI Configuration & Skills

My Understanding

The Gemini CLI is a powerful interface for interacting with Google’s models directly from the terminal. It uses specific context layers like GEMINI.md and “Skills” to adapt its behavior to different projects and tasks.

Key Settings & Concepts

1. GEMINI.md (The Most Important Setting)

This is a file you place in the root of your project. It acts as the “Brain” or “Permanent Instructions” for the AI agent.

• Precedence: Instructions here take absolute precedence over the AI’s default behavior.
• Usage: Define naming conventions, coding styles, or folder structures.

2. Skills (Specialized Expertise)

Skills are specialized, on-demand modules that provide the AI with expert workflows without cluttering the initial context.

• How they work: Gemini CLI scans for skills in your workspace (.gemini/skills/) or global user directory.
• Activation: They are not loaded by default. If you ask for a task that matches a skill’s description (like “help me create a new skill”), the AI uses the activate_skill tool to “pull in” those expert instructions.
• Advantage: They save tokens and keep the AI focused on the specific rules of a complex task (like PR reviews or security audits).

3. Difference: Gemini Skills vs. Others

• Gemini Skills: These are procedural instructions (usually in a SKILL.md file) that tell the AI how to think and which local scripts to run. They are deeply integrated into the CLI’s tool-calling logic.
• Standard Plugins (Obsidian): Most plugins (like “Templater”) are UI-based features that add buttons or menus. Gemini Skills are “under-the-hood” expert modes for the CLI agent itself.

Model Selection

• Gemini 2.0 Flash: (Default) Fast and great at instruction following.
• Gemini 1.5 Pro: Best for complex architectural reasoning.

Related

• Google AI Overview
• Introduction
• Obsidian Usage