vibe-coding-cn/i18n/en/skills/01-ai-tools/headless-cli/references/claude-cli.md

Here is the English translation of the Markdown document:

Claude Code CLI Parameter Reference

Source: Official Documentation

Installation

npm install -g @anthropic-ai/claude-code

Authentication

Requires an Anthropic API Key or Claude Pro/Max subscription:

export ANTHROPIC_API_KEY="YOUR_API_KEY"

Core Commands

Command	Description	Example
claude	Starts an interactive REPL	claude
claude "query"	Starts with an initial prompt	claude "explain this"
claude -p "query"	Print mode, exits after execution	claude -p "review code"
claude -c	Continues the most recent conversation	claude -c
claude -c -p "query"	Continues conversation (Print mode)	claude -c -p "run tests"
claude -r "id" "query"	Resumes a specified session	claude -r "abc123" "continue"
claude update	Updates to the latest version	claude update
claude mcp	Configures the MCP server	claude mcp add server

CLI Parameters

Parameter	Description	Example
--model	Specifies the model	--model claude-sonnet-4
--output-format	Output format: text/json/stream-json	--output-format json
--max-turns	Limits the number of conversation turns	--max-turns 3
--dangerously-skip-permissions	Skips all permission confirmations (YOLO)	See below
--allowedTools	List of allowed tools	--allowedTools "Write" "Bash(git *)"
--disallowedTools	List of disallowed tools	--disallowedTools "Bash(rm *)"
--add-dir	Adds additional working directories	--add-dir ./apps ./lib
--verbose	Enables detailed logs	--verbose
--continue	Continues the recent conversation	--continue
--resume	Resumes a specified session	--resume abc123

Available Models

• claude-sonnet-4 - Balanced model (default)
• claude-opus-4 - Most powerful model
• claude-opus-4.5 - Latest and most powerful

Headless Mode Usage

# Print mode (non-interactive, exits after execution)
claude -p "review this code" --output-format text

# Piped input
cat input.txt | claude -p "explain these errors"

# YOLO mode (skips all permission confirmations)
claude --dangerously-skip-permissions "Your prompt"

# Alias setup
alias cc='claude --dangerously-skip-permissions'

# Continue conversation + Print mode (suitable for scripts)
claude -c -p "show progress"

Interactive Commands (Slash Commands)

Command	Description
/help	Displays all commands
/config	Configures settings
/allowed-tools	Configures tool permissions
/mcp	Manages MCP servers
/vim	Enables vim editing mode

Configuration Files

• User settings: ~/.claude/settings.json
• Project settings: .claude/settings.json

• Local settings: .claude/settings.local.json

  {
  "model": "claude-sonnet-4",
  "permissions": {
  "allowedTools": ["Read", "Write", "Bash(git *)"],
  "deny": ["Read(./.env)", "Bash(rm *)"]
  }
  }
  

Context Files (CLAUDE.md)

• Global: ~/.claude/CLAUDE.md
• Project: ./CLAUDE.md
• Subdirectory: Component-specific instructions

Deep Thinking Trigger Words

Increasing intensity:

• think - Basic thinking
• think hard - Deep thinking
• think harder - Deeper thinking
• ultrathink - Deepest thinking

Common Issues

1. Permission pop-ups: Use --dangerously-skip-permissions
2. Context too long: Use /compact or /clear
3. Reverting changes: Use /rewind