20251120 - 终端中的人工智能指南 - ai-in-the-terminaldocs14-cheat-sheet.md at main · theNetworkChuckai-in-the-terminal¶

分类: Clippings
创建: 2025-11-20
标签: 人工智能, 终端工具, 开发, 项目管理, Git

ai-in-the-terminal/docs/14-cheat-sheet.md at main · theNetworkChuck/ai-in-the-terminal¶

摘要 (Summary)¶

此页面是关于如何在终端中使用人工智能工具的指南，涵盖了安装命令、启动命令、会话命令、项目设置、上下文更新、文件操作、Git集成和常见工作流程等主题。

要点 (Key Facts)¶

安装 Gemini CLI 和 Claude Code 的命令。
启动命令和会话命令的示例。
项目设置和上下文管理的步骤。
Git 集成和多终端工作流程的操作。
错误排除和紧急命令。

正文 (Content)¶

Open in github.dev Open in a new github.dev tab Open in codespace

and

Initial commit: AI in the Terminal companion guide

8710b0b ·

Quick reference for all AI terminal tools covered in the video.

Installation Commands¶

Gemini CLI¶

# Linux/macOS/WSL (npm)
npm install -g @google/generative-ai-cli

# With sudo (if permission error)
sudo npm install -g @google/generative-ai-cli

# macOS (Homebrew)
brew install gemini-cli

Claude Code¶

# All platforms (npm)
npm install -g @anthropic-ai/claude-code

# With sudo
sudo npm install -g @anthropic-ai/claude-code

# Installation command
npm install -g @openai/codex-cli

# Or follow OpenAI documentation

opencode¶

# Installation (one command - from video)
curl -fsSL https://opencode.sh/install.sh | sh

# Reload shell
source ~/.bashrc
# or
source ~/.zshrc

Launch Commands¶

Basic Launch¶

gemini              # Launch Gemini CLI
claude              # Launch Claude Code
codex               # Launch Codex
opencode            # Launch opencode

# Claude Code
claude -r                               # Resume previous session
claude --dangerously-skip-permissions   # Skip safety prompts
claude -r --dangerously-skip-permissions  # Both flags combined

# Gemini CLI
gemini -p "your prompt here"           # Headless mode (one-shot)

# opencode
opencode --model claude-sonnet-4       # Specify model

In-Session Commands¶

Gemini CLI¶

/init               # Create gemini.md context file
/tools              # Show available tools
/help               # Show help
exit                # Exit Gemini (or Ctrl+C)

Claude Code¶

Keyboard shortcuts:

Tab                 # Toggle thinking mode
Shift+Tab           # Toggle planning mode
Ctrl+C              # Interrupt/exit
Ctrl+O              # View agent details (when agent running)

opencode¶

Project Setup¶

# Standard workflow
mkdir my-project
cd my-project

# Launch AI and create context
gemini
> /init

# Verify context file created
ls *.md

# Create project
mkdir my-project
cd my-project

# Initialize all three context files
# Terminal 1
gemini
> /init

# Terminal 2
claude
> /init

# Terminal 3
codex
> /init

# Sync context files
claude
> Sync claude.md to gemini.md and agents.md

# Let AI create it
> /init

# Manual creation
nano gemini.md      # Edit manually
nano claude.md
nano agents.md

Update Context¶

# Ask AI to update
> Update gemini.md with: [your updates]

# Manual edit
nano gemini.md

# Use Claude to sync
claude
> Read claude.md and make gemini.md and agents.md identical

View Context¶

# View file contents
cat gemini.md
cat claude.md
cat agents.md

# In Claude, check context usage
> /context

Agent Management¶

Deploy Agents¶

> @agent-name do a task
> @homelab-guru research NAS options
> @brutal-critic review my script

Multi-Agent Tasks¶

> @agent1 do task A, @agent2 do task B, and compile results

> /output-style                # View available styles
> /output-style new            # Create new style
> /output-style                # Select active style

> /output-style new

# Then describe the style:
Name: my-expert
Description: You are an expert in [domain]...

File Operations¶

Create Files¶

# Ask AI to create
> Create a file named project-plan.md with [content]

# Manual creation
nano project-plan.md

Read Files¶

# AI reads files automatically when mentioned
> Read project-plan.md and summarize

# Reference with @
> Review @project-plan.md

Update Files¶

# AI updates
> Update project-plan.md to add [new section]

# Manual edit
nano project-plan.md

List Files¶

# Terminal command
ls
ls -la

# AI command
> What files are in this directory?
> Show me all markdown files

Git Integration¶

Initialize Repository¶

git init
git add .
git commit -m "Initial commit"

Regular Commits¶

# After work session
git add .
git commit -m "Session summary: [what you did]"
git push

# Using session-closer agent
> @script-session-closer run

# Agent automatically:
# - Summarizes session
# - Updates context files
# - Commits to git

Multi-Terminal Workflows¶

# Terminal 1: Claude
cd ~/my-project
claude

# Terminal 2: Gemini
cd ~/my-project
gemini

# Terminal 3: Codex
cd ~/my-project
codex

tmux (Advanced)¶

# Start tmux session
tmux new -s ai-work

# Create windows
Ctrl+B, C           # New window
Ctrl+B, [0-9]       # Switch to window N
Ctrl+B, "           # Split horizontal
Ctrl+B, %           # Split vertical

Model Management¶

# View available models
> /model

# Switch to specific model
> /model claude-sonnet-4
> /model grok-fast
> /model llama-3.2

Configuration¶

# Edit config file
nano ~/.config/opencode/opencode.jsonc

# Example config for local model
{
  "model": "llama-3.2"
}

Authentication¶

Session Management¶

Advanced Techniques¶

Headless Mode¶

# Gemini one-shot command
gemini -p "Research ZFS performance and save to report.md"

# Claude with pipe
echo "Analyze this file" | claude

# Chain commands
gemini -p "Research topic" && claude -r

# Claude agent using Gemini
> @gemini-research search for latest Docker security updates

# Agent runs:
gemini -p "search for latest Docker security updates"

Obsidian Integration¶

Troubleshooting Commands¶

Permission Issues¶

# Reinstall with sudo
sudo npm install -g @google/generative-ai-cli

# Fix permissions
sudo chown -R $USER /usr/local/lib/node_modules

# Reload shell
source ~/.bashrc
source ~/.zshrc

# Verify installation
which gemini
which claude
which opencode

# Check PATH
echo $PATH

# Verify file exists
ls *.md

# Check file contents
cat gemini.md

# Recreate if needed
> /init

Clear Cache/Reset¶

# Claude Code
rm -rf ~/.config/claude-code

# Gemini CLI
rm -rf ~/.config/gemini-cli

# opencode
rm -rf ~/.config/opencode

Common Workflows¶

# Step 1: Research (Gemini)
gemini
> Research [topic] and compile findings

# Step 2: Write (Claude)
claude
> Read research.md and write comprehensive blog post

# Step 3: Review (Codex)
codex
> Review blog-post.md for accuracy and clarity

Project Kickoff¶

# Create project
mkdir project-name
cd project-name

# Initialize
gemini
> /init
> Help me plan this project: [description]

# Track with git
git init
git add .
git commit -m "Project initialized"

# Start
cd ~/current-project
claude

# Check status
> Where are we in the project?

# Work
> [Your tasks]

# End session
> @script-session-closer run
# Or manually:
git add .
git commit -m "Session: [summary]"

Quick Tips¶

Keyboard Shortcuts¶

Ctrl+C              # Interrupt AI / Exit
Ctrl+D              # Exit (alternative)
Tab                 # Toggle thinking (Claude)
Shift+Tab           # Toggle planning (Claude)
Ctrl+O              # Check agent status (Claude)

Speed Tips¶

# Use dangerous mode (when safe)
claude --dangerously-skip-permissions

# Use headless for quick tasks
gemini -p "quick question"

# Deploy agents for parallel work
> @agent1 task A, @agent2 task B

Organization Tips¶

# One directory per project
~/projects/project-a/
~/projects/project-b/

# Consistent context file naming
gemini.md, claude.md, agents.md

# Use git for everything
git commit regularly

Emergency Commands¶

# Find process ID
ps aux | grep gemini
ps aux | grep claude

# Kill it
kill -9 [PID]

# Or use Ctrl+C twice rapidly

Reset Everything¶

# Remove config directories
rm -rf ~/.config/gemini-cli
rm -rf ~/.config/claude-code
rm -rf ~/.config/opencode

# Reinstall
npm install -g @google/generative-ai-cli
npm install -g @anthropic-ai/claude-code
curl -fsSL https://opencode.sh/install.sh | sh

# Gemini CLI
https://ai.google.dev/gemini-api/docs/cli

# Claude Code
https://docs.anthropic.com/claude/docs/claude-code

# opencode
https://github.com/stackblitz-labs/opencode

# Codex
https://platform.openai.com/docs/tools/codex

Print this page for quick reference!

← Back to README