ekino/gemini

Gemini Skill Guide

When to use this skill

• When the user explicitly asks to run Gemini CLI (e.g., "use gemini to...", "run gemini...", "gemini -p...")
• When the user references Google Gemini or the Gemini CLI tool by name
• When the user wants to leverage Gemini for code analysis, generation, or debugging tasks
• When the user mentions interactive Gemini sessions or Gemini chat mode
• When the user requests non-interactive Gemini workflows with structured output

Running a Task

1. Ask the user (via AskUserQuestion) which model to run (gemini-3-pro-preview,gemini-2.5-pro, gemini-2.5-flash, or other available models) in a single prompt.
2. Determine if the task requires interactive mode (default) or non-interactive mode (for scripting/automation).
3. Assemble the command with the appropriate options:
   • -m, --model <MODEL> - Specify the model to use
   • -p, --prompt <PROMPT> - For non-interactive mode
   • --include-directories <DIRS> - Include additional directories in context
   • --output-format <FORMAT> - Choose output format: text, json, or stream-json
   • --config <KEY=VALUE> - Set configuration options
4. For interactive sessions, use gemini to start an interactive chat session
5. For automated workflows or scripts, use gemini -p "your prompt here" with optional --output-format json for structured output
6. Run the command and capture the output, summarizing the outcome for the user
7. After Gemini completes, inform the user about available next steps, including continuing the conversation or switching modes

Quick Reference

| Use case | Command format | Key flags | | --- | --- | --- | | Interactive chat in current directory | gemini | Default interactive mode | | Interactive with specific model | gemini -m gemini-2.5-flash | -m <model> | | Include multiple directories | gemini --include-directories ../lib,../docs | --include-directories <dirs> | | Non-interactive with text output | gemini -p "prompt" | -p <prompt> | | Scripting with JSON output | gemini -p "prompt" --output-format json | --output-format json | | Streaming events for monitoring | gemini -p "prompt" --output-format stream-json | --output-format stream-json |

Following Up

• After every gemini command, immediately use AskUserQuestion to confirm next steps, collect clarifications, or decide whether to continue with follow-up prompts
• For interactive sessions, inform the user they can continue the conversation with additional prompts
• Restate the chosen model and mode when proposing follow-up actions

Built-in Tools

Gemini CLI comes with powerful built-in tools that can be used during sessions:

• File System Operations: Read, write, and manage files
• Shell Commands: Execute terminal commands
• Web Fetch & Search: Fetch web content and perform Google Search grounding
• MCP Server Integration: Extend capabilities with Model Context Protocol servers

Context Files (GEMINI.md)

Inform users that they can create a GEMINI.md file in their project root to provide persistent context and instructions that will be automatically loaded in every session.

Authentication

Gemini CLI supports multiple authentication methods:

• Login with Google (OAuth): 60 requests/min, 1,000 requests/day (free tier)
• Gemini API Key: 100 requests/day with Gemini 2.5 Pro (free tier)
• Vertex AI: Enterprise features with Google Cloud integration

Before running Gemini commands, ensure the user is authenticated. Common environment variables:

• GEMINI_API_KEY - For API key authentication
• GOOGLE_API_KEY - For Vertex AI authentication
• GOOGLE_CLOUD_PROJECT - For Google Cloud project
• GOOGLE_GENAI_USE_VERTEXAI - Set to true for Vertex AI

Error Handling

• Stop and report failures whenever a gemini command exits non-zero; request direction before retrying
• If authentication fails, guide the user through the authentication process
• When output includes warnings or errors, summarize them and ask how to proceed using AskUserQuestion
• For MCP server errors, check the MCP configuration in ~/.gemini/settings.json