Gemini CLI Tips and Tricks for Agentic Coding (github.com)

Google’s developer guide bundles ~30 pro tips for Gemini CLI, an open‑source, agentic terminal assistant that brings Gemini model capabilities into your shell. The guide covers installation (npm/npx), two auth routes (Google account free tier with generous limits or API key for higher quotas and enterprise data‑use protections), and the CLI’s safe-by-default REPL where proposed file or shell changes are shown as diffs and require approval. It documents features for everyday developer workflows—multimodal inputs, passthrough shell (!) commands, headless/scripting mode, checkpoint/restore undo, session save/resume, token caching/stats, and a YOLO auto‑approve mode (use cautiously). Technically notable: persistent context via GEMINI.md (hierarchical loading from project and global .gemini files) lets you encode style, architecture, and project memory so prompts don’t repeat; custom slash commands are simple TOML templates (~/.gemini/commands or .gemini/commands) that turn prompt patterns into reusable one‑liners; and MCP (Model Context Protocol) servers let you add bespoke tools—local scripts or microservices—to extend Gemini (examples: Figma, Google Docs, clipboard). Together these enable reproducible, team‑shareable agentic workflows while keeping sensitive integrations local. The guide also highlights observability (telemetry), IDE integration (VS Code), and tradeoffs around security and token spend—practical considerations for teams adopting agentic coding in production.