Skip to content

slides.md

Latest commit

 

History

History
1682 lines (1166 loc) · 31.8 KB

slides.md

File metadata and controls

1682 lines (1166 loc) · 31.8 KB
Raw
theme seriph
background https://images.unsplash.com/photo-1451187580459-43490279c0fa?ixlib=rb-4.0.3&auto=format&fit=crop&w=1920&q=80
addons
slidev-addon-qrcode
class text-center
highlighter shiki
lineNumbers true
info ## Agentic Coding with Gemini CLI By Kenneth Kousen Learn more at [KouseniT](https://kousenit.com)
drawings
persist
transition slide-left
title Agentic Coding with Gemini CLI
mdc true
slidev
slide-number controls progress
true
true
true
css unocss
<style> .slidev-page-num { display: block !important; opacity: 1 !important; visibility: visible !important; position: fixed !important; bottom: 1rem !important; right: 1rem !important; z-index: 100 !important; color: #666 !important; font-size: 0.875rem !important; } </style>

Agentic Coding with Gemini CLI

Google's AI Agent in Your Terminal

Press Space for next page

Contact Info

Ken Kousen Kousen IT, Inc.

Course Overview

  • Duration: 5 hours of hands-on learning
  • Format: Instructor-led with multiple labs
  • Hands-on Labs: Real codebases in Python, JavaScript, Java
  • Prerequisites: Command-line experience, development background

Topics Covered

  • Foundation: Installation, CLI basics, authentication
  • Core Skills: File operations, shell integration, context management
  • Customization: GEMINI.md, custom commands, settings.json
  • Safety: Sandbox mode, approval policies, checkpointing
  • Advanced: MCP integration, skills/extensions, session management

What's New in 0.30 – 0.37

  • Agent Skills stable and default-on · SDK + SessionContext (0.30)
  • Policy Engine replaces --allowed-tools / tools.exclude (0.30)
  • Gemini 3.1 Pro Preview available via /model (0.31)
  • Workspace model steering + parallel extension loading (0.32)
  • A2A remote agents over authenticated HTTP (0.33)
  • Plan Mode (opt-in via --approval-mode plan); gVisor/LXC sandboxing on Linux (0.34)
  • Multi-registry MCP/extensions, macOS Seatbelt, Windows sandboxing, Git worktrees (0.36)
  • Ctrl+G replaces Ctrl+X for external editor; persistent policy approvals (0.37)
  • Latest stable track: Gemini CLI 0.37.x

What is Gemini CLI?

  • Open-source AI agent from Google
  • Gemini 3.1 Pro Preview (newest) · Gemini 3 Pro · Gemini 2.5 Pro/Flash
  • Built-in tools: Google Search, file ops, shell, web fetch
  • Model Context Protocol (MCP) support
  • Designed for developers who live in the terminal

Key Differentiators

  • Massive Context: 1 million token context window
  • Google Search Grounding: Real-time web access
  • Free Tier Available: Generous API limits
  • Open Source: Fully open, community-driven
  • MCP Native: Built-in Model Context Protocol support

Models Available

  • Gemini 3.1 Pro Preview: Newest, added in Gemini CLI 0.31
  • Gemini 3 Pro: Most intelligent stable model, best for complex coding
  • Gemini 2.5 Pro: Strong performance, 1M token context
  • Gemini 2.5 Flash: Faster, lower cost option
  • Auto routing: CLI picks the best model for each task
  • Workspace model steering (0.32+): pin a model per workspace

API Access

  • Free Tier: Generous daily limits
  • Google AI Ultra: Full Gemini 3 Pro access
  • Paid API Key: Pay-as-you-go pricing
  • Enable via /settings → "Preview features"

📖 Get API Key: Google AI Studio

Installation

  • npm (recommended): npm install -g @google/gemini-cli
  • npx (no install): npx @google/gemini-cli
  • Verify: gemini --version
  • Current version: 0.37.x (latest stable as of April 2026)
# Install globally
npm install -g @google/gemini-cli

# Verify installation
gemini --version

Authentication

  • Environment Variable: export GEMINI_API_KEY="your-key"
  • Global .env file: ~/.gemini/.env
  • Project .env file: ./.gemini/.env
  • Google Cloud: GOOGLE_CLOUD_PROJECT for Vertex AI
# Option 1: Environment variable
export GEMINI_API_KEY="your-api-key"

# Option 2: Global .env file
echo 'GEMINI_API_KEY=your-api-key' >> ~/.gemini/.env

# Option 3: Project .env file
mkdir -p .gemini && echo 'GEMINI_API_KEY=your-api-key' >> .gemini/.env

Basic Usage Modes

  • Interactive REPL: gemini - Start a conversation
  • One-shot: gemini "prompt" - Single response
  • Piped input: echo "task" | gemini
  • Interactive with context: gemini -i "initial context"
# Interactive mode
gemini

# One-shot mode
gemini "Explain what this codebase does"

# With initial context
gemini -i "You are a Python expert"

layout: image-right image: https://images.unsplash.com/photo-1555949963-ff9fe0c870eb?ixlib=rb-4.0.3&auto=format&fit=crop&w=1920&q=80 backgroundSize: cover

Core Features

Essential Capabilities

Master the fundamentals

File References with @

  • Reference files directly: @./src/main.js
  • Reference directories: @./src/ (recursive)
  • Reference images: @./screenshot.png
  • Multiple references in one prompt
# Reference a specific file
gemini "Explain @./src/app.py"

# Reference multiple files
gemini "Compare @./old.js and @./new.js"

# Reference a directory
gemini "Analyze the architecture in @./src/"

Shell Integration with !

  • Execute shell commands: !git status
  • Toggle persistent shell mode: !
  • Gemini can observe and analyze output
  • Combine with AI analysis
# In interactive mode:
> !npm test
# Gemini sees the test output

> !git diff
# Ask Gemini to analyze the changes

# Toggle persistent shell mode
> !

Slash Commands: Navigation

  • /help - Show available commands
  • /clear - Clear conversation history
  • /memory show - View loaded context
  • /memory refresh - Reload GEMINI.md files

Slash Commands: Workflow

  • /init - Generate project GEMINI.md
  • /prompt-suggest - Generate stronger prompts
  • /stats - Session and quota metrics

Slash Commands: Session Control

  • /resume - Open session browser
  • /rewind - Navigate and optionally revert history
  • /restore - Recover from checkpoint

Slash Commands in Action (Context)

# Show what context is loaded
/memory show

# Reload all GEMINI.md files
/memory refresh

# Generate a GEMINI.md for current project
/init

# Suggest stronger prompt wording
/prompt-suggest

Slash Commands in Action (Sessions)

# Open session browser
/resume

# Rewind through recent interactions
/rewind

# Restore from checkpoint list
/restore

Keyboard Shortcuts: Editing

  • Ctrl+L - Clear screen
  • Ctrl+V - Paste text/images
  • Ctrl+G - Open external editor (was Ctrl+X before 0.37)

Keyboard Shortcuts: Control

  • Ctrl+Y - Toggle auto-approval (YOLO mode)
  • Shift+Tab - Cycle approval mode (default/auto_edit/plan)
  • Ctrl+C - Cancel current operation
  • Ctrl+D - Exit Gemini CLI

Built-in Tools

  • File System: read_file(), write_file(), replace(), glob()
  • Shell: Execute terminal commands
  • Web: google_web_search(), web_fetch()
  • Memory: save_memory() for cross-session recall
# Gemini automatically uses appropriate tools
"Search the web for React 19 new features"
# Uses google_web_search()

"Read all Python files in src/"
# Uses glob() and read_file()

"Update the README with the changes we made"
# Uses write_file()

layout: image-right image: https://images.unsplash.com/photo-1488590528505-98d2b5aba04b?ixlib=rb-4.0.3&auto=format&fit=crop&w=1920&q=80 backgroundSize: cover

Safety & Control

Work Safely

Approval modes and sandboxing

Approval Modes

  • default (the default): Prompt for approval on each tool call
  • plan: Read-only — draft a plan, no execution until you approve
  • auto_edit: Auto-approve file edit tools only
  • yolo: Auto-approve ALL tool calls
# Default - interactive, prompts per tool call
gemini

# Draft a plan first (read-only until approved)
gemini --approval-mode plan

# Auto-approve edits only
gemini --approval-mode auto_edit

# Auto-approve everything (YOLO mode)
gemini --approval-mode yolo
# Or use Ctrl+Y in interactive mode

Plan Mode

  • Opt-in with --approval-mode plan — read-only until you approve
  • Drafts a structured plan before touching any file or tool
  • Open the plan in your external editor with Ctrl+G
  • Use Shift+Tab to cycle into default / auto_edit once you're ready
  • 0.33+ adds research subagents, annotations, and a copy subcommand
  • Great first step when exploring an unfamiliar codebase
# Launch in plan mode
gemini --approval-mode plan

Sandbox Mode

  • Isolate file operations away from your host
  • Multiple backends depending on your OS
  • Prevents accidental system changes
  • Perfect for exploring unfamiliar code
# Run in sandbox mode (picks the best backend for your OS)
gemini --sandbox

# Short form
gemini -s "Refactor this entire codebase"

Sandbox Backends (0.34 – 0.36)

  • Docker / Podman: original cross-platform option
  • gVisor (runsc): Linux, added in 0.34 — strong kernel-level isolation
  • LXC: Linux, experimental in 0.34
  • macOS Seatbelt: native in 0.36, no Docker required
  • Windows sandboxing: native subagent sandboxing in 0.36
// ~/.gemini/settings.json
{
  "tools": {
    "sandbox": "docker"   // or "gvisor", "seatbelt", etc.
  }
}

📖 geminicli.com/docs/cli/settings/

Checkpointing

  • Automatic: Snapshots created before each file modification
  • Shadow Git: Stored in ~/.gemini/history/ (not your repo)
  • Includes: Files + conversation + tool call
  • Disabled by default: Must enable in settings
// ~/.gemini/settings.json
{ "general": { "checkpointing": { "enabled": true } } }

Restoring Checkpoints

# List and select a checkpoint to restore
/restore

# Shows timestamps + filename + tool name
# e.g., 2025-06-22T10-00-00_000Z-app.py-write_file

Restores files AND resets conversation to that point

layout: image-left image: https://images.unsplash.com/photo-1454165804606-c3d57bc86b40?ixlib=rb-4.0.3&auto=format&fit=crop&w=1920&q=80 backgroundSize: cover

Context Management

GEMINI.md Files

Project memory and instructions

What is GEMINI.md?

  • Project memory loaded automatically
  • Coding standards and conventions
  • Architecture context for the AI
  • Persistent instructions across sessions
  • Similar to CLAUDE.md or AGENTS.md

Hierarchical Loading

graph TD
    A[~/.gemini/GEMINI.md<br/>Global Rules] --> B[project/GEMINI.md<br/>Project Standards]
    B --> C[project/frontend/GEMINI.md<br/>Frontend Conventions]
    B --> D[project/backend/GEMINI.md<br/>Backend Patterns]

    style A fill:#FFE5CC,stroke:#333,stroke-width:2px,color:#000
    style B fill:#CCE5FF,stroke:#333,stroke-width:2px,color:#000
    style C fill:#E5CCFF,stroke:#333,stroke-width:2px,color:#000
    style D fill:#CCFFE5,stroke:#333,stroke-width:2px,color:#000
Loading

More specific files override general ones

Example GEMINI.md

# Project: Weather API
## Tech Stack
- Backend: Python Flask
- Database: PostgreSQL
- Testing: pytest
## Coding Standards
- Use type hints for all functions
- Follow PEP 8 style guide
- Write docstrings for public APIs
## Current Focus
Implementing caching layer for API responses

Memory Commands

  • /memory show - View combined context
  • /memory refresh - Reload all GEMINI.md files
  • /memory add <text> - Append to global GEMINI.md
  • /init - Generate starter GEMINI.md
# See what context is loaded
/memory show

# Add a quick note to global memory
/memory add "Always use async/await for database calls"

# Generate a project-specific GEMINI.md
/init

Modular Imports

  • Import other files with @file.md syntax
  • Break large context into components
  • Supports relative and absolute paths
# GEMINI.md

## Project Overview
@./docs/architecture.md

## Coding Standards
@./docs/style-guide.md

## API Documentation
@./docs/api-reference.md

layout: image-right image: https://images.unsplash.com/photo-1558494949-ef010cbdcc31?ixlib=rb-4.0.3&auto=format&fit=crop&w=1920&q=80 backgroundSize: cover

Configuration

Customize Your Setup

settings.json and environment

Configuration Layers

  1. Default values - Built-in defaults
  2. User settings - ~/.gemini/settings.json
  3. Project settings - .gemini/settings.json
  4. Environment variables - Including .env files
  5. Command-line arguments - Highest priority

settings.json Options

{
  "general": {
    "preferredEditor": "vscode",
    "vimMode": false,
    "checkpointing": { "enabled": true }
  }
}

settings.json Options (UI + Tools)

{
  "ui": {
    "hideTips": false,
    "hideBanner": false
  },
  "tools": {
    "sandbox": false
  }
}

Tool Configuration

  • tools.allowed: Whitelist available tools
  • Policy Engine (preferred): deny rules for blocking tools
  • tools.exclude still works but is deprecated since 0.30
{
  "tools": {
    "allowed": ["read_file", "write_file", "glob"]
  }
}

Pass a policy file to block tools:

gemini --policy ./policy.toml

Policy files use TOML with decision + priority fields per rule. See the current schema at the docs link below.

📖 geminicli.com/docs/reference/policy-engine

File Filtering

  • respectGitIgnore: Honor .gitignore patterns
  • enableRecursiveFileSearch: Recursive completion
  • .geminiignore: Custom ignore patterns
{
  "context": {
    "fileFiltering": {
      "respectGitIgnore": true,
      "respectGeminiIgnore": true,
      "enableRecursiveFileSearch": true
    }
  }
}

Environment Variables

Variable Purpose
GEMINI_API_KEY Google AI Studio authentication
GOOGLE_API_KEY Vertex API key authentication
GEMINI_MODEL Override default model
GOOGLE_CLOUD_PROJECT GCP project for Vertex AI
GOOGLE_CLOUD_LOCATION GCP region
GEMINI_CLI_HOME Override Gemini CLI home directory
HTTP_PROXY Network proxy

Custom Context Filename

  • Change from GEMINI.md to a custom name
  • AGENTS.md is officially supported — friendly to multi-agent repos
  • Support multiple filenames in priority order
  • Include additional directories
{
  "context": {
    "fileName": ["AGENTS.md", "GEMINI.md", "CONTEXT.md"],
    "includeDirectories": ["~/shared-context"],
    "loadMemoryFromIncludeDirectories": true
  }
}

layout: image-right image: https://images.unsplash.com/photo-1516321318423-f06f85e504b3?ixlib=rb-4.0.3&auto=format&fit=crop&w=1920&q=80 backgroundSize: cover

Advanced Features

Power User Tools

MCP, Extensions, Sessions

Model Context Protocol (MCP)

  • Standard protocol for AI-to-system connections
  • Connect to external tools and services
  • Supports local commands, HTTP, and SSE
  • OAuth 2.0 for remote authentication

MCP Configuration: Firecrawl

{
  "mcpServers": {
    "firecrawl": {
      "command": "npx",
      "args": ["@modelcontextprotocol/server-firecrawl"],
      "env": {
        "FIRECRAWL_API_KEY": "${FIRECRAWL_API_KEY}"
      }
    }
  }
}

MCP Configuration: Database

{
  "mcpServers": {
    "postgres": {
      "command": "npx",
      "args": ["@modelcontextprotocol/server-postgres"],
      "env": {
        "CONNECTION_STRING": "${DATABASE_URL}"
      }
    }
  }
}

Managing MCP Servers

# List configured MCP servers
gemini mcp list

# Add a new MCP server
gemini mcp add github npx -y @modelcontextprotocol/server-github

# Remove an MCP server
gemini mcp remove github

Managing MCP Auth/Refresh

# List MCP tools/resources from inside session
/mcp list

# Authenticate with OAuth-enabled MCP server
/mcp auth github

# Refresh MCP tools/resources
/mcp refresh

MCP Server Options

  • command: Shell command to start server
  • args: Command arguments
  • env: Environment variables
  • cwd: Working directory

MCP Server Options (Advanced)

  • timeout: Startup timeout in ms
  • includeTools/excludeTools: Filter available tools
  • trust: Bypass tool confirmation for this server

Popular MCP Servers

  • Firecrawl - Web scraping, search, content extraction
  • PostgreSQL - Database queries and schema
  • Filesystem - Extended file operations
  • Slack - Team communication
  • Playwright - Browser automation

📖 Registry: modelcontextprotocol.io/registry

Multi-Registry Architecture (0.36+)

  • Gemini CLI now supports multiple extension/MCP registries side by side
  • Mix the public MCP registry with internal/private registries
  • Useful for teams that host vetted MCP servers internally
  • Extensions load in parallel (added in 0.32)
# List registries and what each provides
gemini extensions list
gemini mcp list

📖 geminicli.com/docs/changelogs/latest/

A2A Remote Agents (0.33+)

  • Gemini CLI can call remote agents over HTTP via the A2A protocol
  • Authenticated agent-card discovery for private catalogs
  • Same permissioning story as MCP: allow/deny per server
  • Good fit for internal "shared agents" a team maintains centrally

Extensions System

  • Extend Gemini CLI capabilities
  • Place in ~/.gemini/extensions/
  • Configure with gemini-extension.json
  • List with gemini --list-extensions
# List available extensions
gemini --list-extensions

# Use specific extensions only
gemini -e extension1 -e extension2

# Manage extensions
gemini extensions list
gemini extensions enable my-extension

Custom Commands

  • Create reusable prompt templates
  • Place TOML files in ~/.gemini/commands/
  • Access via slash command syntax
# ~/.gemini/commands/review.toml
description = "Review code for issues"

prompt = """
Review the following code for:
- Security vulnerabilities
- Performance issues
- Best practices violations

{{args}}
"""

Session Management

  • Resume sessions: Continue previous conversations
  • List sessions: View available sessions
  • Delete sessions: Clean up old conversations
# Resume the latest session
gemini --resume

# Resume by index
gemini --resume 3

# List available sessions
gemini --list-sessions

# Delete a session
gemini --delete-session 5

Output Formats

  • text: Default human-readable output
  • json: Structured JSON for scripting
  • stream-json: Streaming JSON for real-time
# JSON output for automation
gemini -o json "List all TODO comments"

# Stream JSON for real-time processing
gemini -o stream-json "Analyze this codebase"

IDE Integration

  • VS Code Integration: Connect to workspace
  • Native diff viewing: Review changes in editor
  • Context sharing: IDE context available to Gemini
{
  "preferredEditor": "vscode"
}

layout: image-left image: https://images.unsplash.com/photo-1498050108023-c5249f4df085?ixlib=rb-4.0.3&auto=format&fit=crop&w=1920&q=80 backgroundSize: cover

Practical Applications

Real-World Workflows

Common use cases

Code Exploration

  • Understand unfamiliar codebases
  • Trace dependencies and data flow
  • Find patterns and conventions
  • Generate architecture documentation
gemini "Analyze the architecture of @./src/ and explain
how the components interact"

gemini "Trace the flow from the API endpoint to the database
in @./src/controllers/ and @./src/services/"

Test Generation

  • Generate unit tests for existing code
  • Identify edge cases automatically
  • Create integration test scaffolding
  • Mock setup and fixtures
gemini "Create comprehensive unit tests for @./src/utils.py
with pytest, including edge cases"

gemini "Generate integration tests for @./src/api/users.py
with proper mocking"

Documentation Generation

  • README files for projects
  • API documentation
  • Architecture diagrams (Mermaid)
  • Code comments and docstrings
gemini "Generate a comprehensive README.md for this project"

gemini "Add detailed docstrings to all public functions
in @./src/services/"

gemini "Create a Mermaid diagram showing the system architecture"

Refactoring & Modernization

  • Upgrade legacy code patterns
  • Apply modern language features
  • Improve code organization
  • Fix anti-patterns
gemini "Refactor @./src/legacy.py to use modern Python 3.12
features like type hints and match statements"

gemini "Convert this callback-based code to async/await
@./src/api.js"

Debugging Workflows

  • Analyze error messages and stack traces
  • Identify root causes
  • Suggest fixes with context
  • Test and verify solutions
gemini "This test is failing with @./tests/output.log.
Analyze the error and fix the issue in @./src/app.py"

gemini "Debug why the API returns 500 errors.
Check @./src/routes.py and @./src/middleware.py"

Git Workflows

  • Generate commit messages
  • Create pull request descriptions
  • Analyze diffs and changes
  • Resolve merge conflicts
# Analyze staged changes
!git diff --staged
"Generate a conventional commit message for these changes"

# Create PR description
"Create a pull request description summarizing the changes
from the last 5 commits"

CI/CD Integration

  • Non-interactive mode for pipelines
  • JSON output for parsing
  • Exit codes for success/failure
  • Automated code reviews
# In CI/CD pipeline
gemini "Review @./src/ for security issues" -o json > review.json

# Check exit code
if gemini "Verify all tests pass" -o json; then
  echo "All checks passed"
fi

layout: image-right image: https://images.unsplash.com/photo-1522071820081-009f0129c71c?ixlib=rb-4.0.3&auto=format&fit=crop&w=1920&q=80 backgroundSize: cover

Optional Advanced Module

  • Designed for teams moving from personal usage to production workflows
  • Can be taught live or assigned as follow-up practice
  • Focus areas: auth strategy, governance, skills/MCP, CI automation

Authentication Matrix (Team Use)

  • Login with Google: Best default for local interactive usage
  • GEMINI_API_KEY: Simple personal/team scripts via AI Studio
  • Vertex + ADC: Enterprise cloud projects and IAM-managed access
  • Service account credentials: CI/CD and headless automation

Auth Quick Checks

# API key route
export GEMINI_API_KEY="..."
gemini

# Vertex route (ADC)
unset GEMINI_API_KEY GOOGLE_API_KEY
gcloud auth application-default login
export GOOGLE_CLOUD_PROJECT="my-project"
export GOOGLE_CLOUD_LOCATION="us-central1"
gemini

Governance: Approval and Policies

  • Approval mode controls are per-session safety rails
  • Policies provide durable guardrails across users/projects
  • Keep high-risk tools constrained in team settings
  • Prefer explicit allow/deny patterns over ad-hoc approvals

Governance: Hooks and Trust

  • /hooks list shows active lifecycle hooks
  • Use hooks for auditing, validation, and policy enforcement
  • Folder trust impacts settings, skills, and context loading
  • Treat untrusted repos with stricter modes and sandboxing

Skills + MCP at Scale

  • Skills package repeatable expert workflows for teams
  • MCP connects external systems (docs, data, browsers, APIs)
  • Resource URIs can be injected via @server://resource/path
  • Use includeTools/excludeTools to narrow risky MCP exposure

Skills + MCP Commands

# Skills lifecycle
/skills list
/skills disable skill-name
/skills enable skill-name

# MCP visibility and maintenance
/mcp list
/mcp auth server-name
/mcp refresh

Automation Patterns (Non-Interactive)

  • Prefer -o json for machine-readable parsing
  • Use exit codes for pipeline gates
  • Keep prompts deterministic and repo-scoped
  • Store logs/artifacts for auditability

CI Example: Gate + Report

gemini -o json "Review @./src/ for security issues" > review.json

if gemini -o json "Run checks and report pass/fail"; then
  echo "Gate passed"
else
  echo "Gate failed" && exit 1
fi

layout: image-right image: https://images.unsplash.com/photo-1522071820081-009f0129c71c?ixlib=rb-4.0.3&auto=format&fit=crop&w=1920&q=80 backgroundSize: cover

Best Practices

Professional Workflows

Tips for success

Effective Prompting

  • Be specific about what you want to achieve
  • Provide context about goals and constraints
  • Use file references to ground the conversation
  • Iterate for complex tasks
  • Include examples when possible

Safety Guidelines

  • Start in default mode - Get comfortable first
  • Enable checkpointing - Safety net for mistakes
  • Review generated code - Don't blindly accept
  • Use sandbox for exploratory work
  • Commit often - Git is your safety net

Project Setup

  1. Create comprehensive GEMINI.md
  2. Set up project-specific settings.json
  3. Configure relevant MCP servers
  4. Create custom commands for common tasks
  5. Establish team conventions

GEMINI.md Best Practices

  • Keep it focused - Relevant project info only
  • Update regularly - Reflect current state
  • Use imports - Modularize large contexts
  • Include examples - Show expected patterns
  • Document conventions - Style guides, patterns

Team Collaboration

  • Share GEMINI.md in version control
  • Standardize settings.json across team
  • Create shared custom commands
  • Document AI-assisted workflows
  • Review AI-generated code together

Common Pitfalls

  • Overly broad prompts → Be specific
  • Missing context → Use GEMINI.md
  • Skipping review → Always verify output
  • YOLO mode too early → Build trust first
  • Ignoring checkpoints → Enable early

Troubleshooting

  • Authentication issues: Check GEMINI_API_KEY
  • Rate limits: Use appropriate tier
  • Tool failures: Check MCP server status
  • Context not loading: Run /memory refresh
  • Debug mode: Use gemini -d for details
# Debug mode for troubleshooting
gemini -d

# Check memory/context
/memory show

# Refresh context files
/memory refresh

Quick Access

Gemini CLI Docs

github.com/google-gemini/gemini-cli

Course Repository

github.com/kousen/gemini-training

Important Links

📚 Official Documentation

https://github.com/google-gemini/gemini-cli

🔑 Get API Key

https://aistudio.google.com/apikey

📦 MCP Server Registry

https://modelcontextprotocol.io/registry

💻 Course Materials

https://github.com/kousen/gemini-training

Command Reference: Basic Usage

# Interactive mode
gemini

# One-shot mode
gemini "prompt"

# Interactive with initial context
gemini -i "context"

Command Reference: Safety

# Run in sandbox mode
gemini --sandbox

# Auto-approve all tool calls
gemini --approval-mode yolo

# Auto-approve file edits only
gemini --approval-mode auto_edit

Command Reference: Sessions

# Resume last session
gemini --resume

# List available sessions
gemini --list-sessions

# Delete a session
gemini --delete-session 2

Command Reference: Output

# JSON output for scripting
gemini -o json "prompt"

# Streaming JSON events
gemini -o stream-json "prompt"

# Debug mode
gemini -d

Thank You!

Questions?

Kenneth Kousen Author, Speaker, Java & AI Expert

kousenit.com | @kenkousen