A field manual for Google's agentic terminal assistant

What it does

This repository is a curated guide—not the CLI itself—documenting around thirty advanced tactics for Google’s Gemini CLI, an open-source agentic terminal assistant. It gathers advice from the tool’s creators and the broader community into a single reference aimed at power users. The focus is on moving past basic chat and toward persistent project memory, custom tooling, and reliable automation.

The interesting bit

The guide treats the CLI less like a search engine and more like a junior dev who needs onboarding: it explains how to use GEMINI.md files as hierarchical project memory, how to let the agent build its own helpers on the fly, and how to checkpoint and restore state when an agentic plan goes sideways. That framing—documenting how to manage an agent rather than just query a model—is what elevates it above a standard cheat sheet.

Key highlights

• Hierarchical context loading via GEMINI.md files (global, project, and subdirectory levels) with /memory show and /memory refresh commands.
• Custom slash commands, MCP server integration, and on-the-fly tool creation to extend the CLI’s capabilities without leaving the terminal.
• Checkpointing and /restore for undoing agent actions, plus a “YOLO mode” for auto-approving tool executions when you’re feeling brave.
• Headless/scripting mode, session save/resume, and a GitHub Action for automating repository tasks in CI.
• Token spend tracking, telemetry insights, and a Corgi Mode Easter Egg (the essential feature).

Verdict

Grab this if you’re already using Gemini CLI and suspect you’re only scratching the surface of its agentic features. If you don’t yet have a gemini> prompt in your terminal, this is overkill—come back once you’ve crossed that threshold.