Command Reference Cheatsheet

Q: How do you start Claude Code and Codex from the command line?

claude starts interactive mode and claude -p runs a one-off, non-interactive query. codex starts an interactive REPL, and codex "prompt" starts with an initial prompt. Both accept --model to select the model.

Q: Which Cursor context references get used most?

@codebase searches the entire project and @web returns web search results — the top two of the Cursor list. @files, @folders, @symbols, and @git scope context to specific files, directories, functions, or Git history.

Q: Which Codex approval mode should be used?

untrusted prompts for commands outside the trusted set, on-request lets the agent request an elevated action, and never never prompts while disallowed sandbox operations fail. --full-auto is a deprecated compatibility shortcut, and --dangerously-bypass-approvals-and-sandbox (--yolo) is for hardened, isolated runners only.

Q: How are custom slash commands created in Claude Code?

Custom commands are markdown files in .claude/commands/. $ARGUMENTS passes parameters from the slash command, allowed-tools restricts which tools it can use, argument-hint shows hint text for the argument, @file.js references files inline, and !command executes bash commands inline.

Q: What replaced the Claude Code thinking-budget ladder?

The think / think harder / ultrathink budget ladder is gone. think, think hard, and think more are ordinary text, while ultrathink is a one-turn hint that adds a deeper-reasoning instruction without changing API effort. Durable depth is controlled with the thinking toggle and /effort.

Command reference for Cursor, Claude Code, and Codex collects the CLI flags, slash commands, and context symbols of all three tools in one page. Cursor uses @ references and a command palette, Claude Code exposes built-in slash commands plus claude CLI flags, and Codex combines slash commands with approval and sandbox flags.

This is the reference you keep open in a second tab. Every command, flag, and symbol across all three tools in one place.

Quick Command Overview

@ — Reference symbols, files, folders, URLs
# — Quick file reference
/ — Slash commands
Cmd/Ctrl+Shift+P — Command palette
cursor — CLI command

/ — Built-in slash commands
@ — Reference files in prompts
# — Quick CLAUDE.md update
claude — Start interactive mode
claude -p — One-off query (non-interactive)

/ — Slash commands
codex — Start interactive mode
codex "prompt" — Start with initial prompt
codex --ask-for-approval or -a — Set trust level
codex --model — Select model

Cursor Commands

@ Symbol References

Symbol	Description	Example
`@files`	Reference specific files	`@main.js explain this file`
`@folders`	Include entire directories	`@src/components analyze structure`
`@code`	Reference code blocks	`@calculateTotal how does this work?`
`@symbols`	Reference functions/classes	`@UserAuth explain implementation`
`@definitions`	Go to symbol definitions	`@getPriceDefinition show me`
`@docs`	Reference documentation	`@ReactDocs useState hook`
`@git`	Git history and changes	`@git recent commits`
`@link`	Reference URLs	`@link https://api.docs.com`
`@linter-errors`	Current linting issues	`@linter-errors fix all`
`@past-chats`	Previous conversations	`@past-chats auth discussion`
`@recent-changes`	Recent file modifications	`@recent-changes last hour`
`@web`	Web search results	`@web React 19 features`
`@cursor-rules`	Project rules	`@cursor-rules show current`
`@codebase`	Search entire project	`@codebase authentication flow`

Slash Commands

Command	Description
`/Reset Context`	Reset to default state
`/Generate Cursor Rules`	Auto-generate project rules from codebase
`/Disable Iterate on Lints`	Prevent auto-fixing lint errors
`/Add Open Files to Context`	Include all open tabs
`/Add Active Files to Context`	Include only visible tabs
`/loop`	Bundled skill (Cursor 3.5+): re-run a prompt on a local schedule until an outcome is reached or you stop it. Omit the interval and the agent picks when to wake

Command Palette (Cmd/Ctrl+Shift+P)

Command	Description
`Cursor Settings`	Open Cursor-specific settings
`Cursor: New Rule`	Create new project rule
`Cursor: Toggle Privacy Mode`	Enable/disable privacy
`Cursor: Export Chat`	Save conversation
`Cursor: Import Settings`	Import from VS Code
`Cursor: Toggle Max Mode`	Enable advanced features
`Cursor: Generate Commit Message`	AI commit message
`Cursor: Resolve Conflicts`	AI merge conflict resolution

Cursor CLI

# Basic usage
cursor [options] [paths...]          # Open files/folders
cursor .                             # Open current directory
cursor file.js                       # Open specific file
cursor folder/                       # Open folder
cat file.txt | cursor -              # Read from stdin

# File operations
cursor -d --diff <file1> <file2>     # Compare two files
cursor -m --merge <p1> <p2> <b> <r>  # Three-way merge
cursor -g --goto file.js:10:5        # Go to line:column
cursor -a --add folder               # Add folder to window
cursor --remove folder               # Remove folder from window

# Window management
cursor -n --new-window               # Force new window
cursor -r --reuse-window             # Force reuse window
cursor -w --wait                     # Wait for files to close

# Agent modes (v2.4+)
cursor --plan "describe task"        # Plan mode (read-only)
cursor --ask "question"              # Ask mode (read-only)

# Configuration
cursor --locale <locale>             # Set locale (e.g., en-US)
cursor --user-data-dir <dir>         # Custom data directory
cursor --profile <name>              # Use/create profile
cursor --add-mcp <json>              # Add MCP server

# Extensions
cursor --list-extensions             # List installed
cursor --install-extension <id>      # Install
cursor --uninstall-extension <id>    # Uninstall
cursor --update-extensions           # Update all

# Troubleshooting
cursor -v --version                  # Show version
cursor -h --help                     # Show help
cursor -s --status                   # Show diagnostics
cursor --disable-extensions          # Safe mode
cursor --verbose                     # Verbose output

Claude Code Commands

Built-in Slash Commands

Command	Purpose	Example
`/add-dir`	Add working directories	`/add-dir ../backend`
`/bug`	Report bugs to Anthropic	`/bug unexpected behavior with...`
`/cd`	Move the session to a new working directory without breaking the prompt cache (v2.1.169)	`/cd ../backend`
`/clear`	Clear conversation history	`/clear`
`/code-review`	Run the dedicated code-review workflow; add `--fix` when supported to apply findings	`/code-review high`
`/compact`	Summarize conversation	`/compact focus on auth logic`
`/config`	View/modify configuration	`/config`
`/cost`	Show token usage stats	`/cost`
`/doctor`	Check installation health	`/doctor`
`/effort`	Set reasoning effort for the selected model	`/effort high`
`/goal`	Set a completion condition and keep working across turns until an evaluator confirms it (v2.1.139+); `/goal clear` ends it early	`/goal all tests in tests/auth pass`
`/help`	Get usage help	`/help`
`/init`	Initialize CLAUDE.md	`/init`
`/login`	Switch accounts	`/login`
`/logout`	Sign out	`/logout`
`/loop`	Bundled skill: re-run a prompt on an interval, at a self-paced delay, or as a maintenance routine. Alias `/proactive`	`/loop 5m check the deploy`
`/mcp`	Manage MCP servers	`/mcp`
`/memory`	Edit CLAUDE.md files	`/memory`
`/model`	Select AI model; `/model fable` switches to Claude Fable 5 (v2.1.170+)	`/model fable`
`/permissions`	Update permissions	`/permissions`
`/review`	Request code review	`/review`
`/simplify`	Review changed code for reuse, quality, and efficiency; distinct from `/code-review`	`/simplify`
`/status`	View system status	`/status`
`/terminal-setup`	Configure terminal	`/terminal-setup`
`/usage`	Detailed token metrics	`/usage`
`/workflows`	Watch and manage dynamic workflow runs	`/workflows`

Claude CLI Flags

# Interactive mode
claude                              # Start REPL
claude "explain this project"       # Start with prompt

# One-off queries (non-interactive)
claude -p "explain this function"   # Query and exit
cat logs.txt | claude -p "analyze"  # Process piped content

# Session management
claude -c                           # Continue last conversation
claude -c -p "add tests"           # Continue non-interactively
claude -r                           # Choose session to resume
claude -r "abc123"                  # Resume specific session

# Utility
claude update                       # Update to latest
claude mcp                          # Configure MCP servers

Flag	Description	Example
`-p, --print`	Non-interactive mode	`claude -p "fix this bug"`
`-c, --continue`	Continue last conversation	`claude -c`
`-r, --resume`	Resume specific session	`claude -r "session-id"`
`--model`	Specify model	`claude --model opus`
`--dangerously-skip-permissions`	Skip all permission prompts	For CI/CD only
`--output-format`	Output format (text/json/stream-json)	`claude -p "query" --output-format json`
`--input-format`	Input format (text/stream-json)	`claude -p --input-format stream-json`
`--verbose`	Verbose output	`claude --verbose`
`--debug`	Enable debug mode (includes MCP)	`claude --debug`
`--safe-mode`	Start with all customizations (CLAUDE.md, plugins, skills, hooks, MCP servers) disabled, for troubleshooting (v2.1.169; also via the `CLAUDE_CODE_SAFE_MODE` env var)	`claude --safe-mode`
`--add-dir`	Add additional directories	`claude --add-dir ../shared`
`--max-turns`	Limit agentic turns	`claude -p "task" --max-turns 3`
`--permission-mode`	Set permission mode	`claude --permission-mode plan`
`--session-id`	Use specific session ID	`claude --session-id "uuid"`
`--from-pr`	Start with PR context	`claude --from-pr 123`
`-v, --version`	Show version	`claude --version`
`-h, --help`	Show help	`claude --help`

Custom Slash Commands

Create in .claude/commands/ as markdown files:

---
allowed-tools: [Edit, Read, Bash]
description: Fix GitHub issue
argument-hint: issue-number
---

Fix GitHub issue #$ARGUMENTS:

1. Fetch details: !gh issue view $ARGUMENTS
2. Analyze the problem
3. Implement solution
4. Add tests
5. Create PR with fix

Features:

$ARGUMENTS — Pass parameters from the slash command
allowed-tools — Restrict which tools the command can use
argument-hint — Show hint text for the argument
@file.js — Reference files inline
!command — Execute bash commands inline

MCP Commands

# Configure MCP servers
claude mcp add <name> -s <scope> -- <command> [args]
claude mcp remove <name>
claude mcp list

# In-session MCP
/mcp                           # List available MCP tools
/mcp__github__create-pr        # Use a specific MCP tool

Codex Commands

Codex CLI

# Interactive mode
codex                               # Start interactive REPL
codex "explain this project"        # Start with initial prompt

# Configuration
codex --model <model>               # Specify model
codex -a <mode>                     # untrusted | on-request | never
codex --full-auto                   # Deprecated compatibility shortcut
codex --dangerously-bypass-approvals-and-sandbox  # alias --yolo; no approvals, no sandbox

# Utility
codex --version                     # Show version
codex --help                        # Show help

Codex CLI Flags

Flag	Description	Example
`--model`	Specify model	`codex --model gpt-5.6-sol`
`-a, --ask-for-approval`	Approval policy; sandbox is configured separately	`codex -a on-request`
`--full-auto`	Deprecated compatibility shortcut; configure approval and sandbox explicitly	`codex -a on-request --sandbox workspace-write`
`--dangerously-bypass-approvals-and-sandbox`	Alias `--yolo`. No approvals, no sandbox — isolated runners only	`codex --yolo`
`--sandbox`	Sandbox mode	`codex --sandbox workspace-write`
`--version`	Show version	`codex --version`
`--help`	Show help	`codex --help`

Codex Approval Modes

Mode	Description	Use When
`untrusted`	Prompts for commands outside the trusted set; sandbox remains separate.	Cautious exploration
`on-request`	Lets the agent request an elevated action when needed.	Interactive development
`never`	Never prompts; disallowed sandbox operations fail.	Trusted unattended jobs in a deliberately chosen sandbox
`--full-auto`	Deprecated compatibility shortcut.	Migrate to explicit approval + sandbox settings
`--dangerously-bypass-approvals-and-sandbox` (`--yolo`)	Truly approves everything — no approvals, no sandbox.	Hardened/isolated runners only

Codex Slash Commands

Command	Description
`/model`	Switch model
`/new`	Start new conversation
`/compact`	Summarize conversation
`/goal`	Set a persistent objective Codex works toward across turns and sessions. Stable and enabled by default since CLI 0.133.0
`/goal edit` / `pause` / `resume` / `clear`	Revise, suspend, continue, or drop the active goal (`edit` added in 0.131.0)

Advanced Patterns

Cursor: Context Chaining

Combine multiple @ references for comprehensive context:

@UserAuth @DatabaseSchema @recent-changes
explain how authentication works with the database

@src/api/*.js @tests/*.test.js
ensure all API endpoints have corresponding tests

Claude Code: Piping and Automation

# Analyze logs
tail -f app.log | claude -p "monitor for errors"

# Automated code review
git diff | claude -p "review these changes for bugs"

# CI/CD integration
claude -p "fix linting errors" --dangerously-skip-permissions

# JSON output for scripts
claude -p "list all TODO comments" --output-format json

# Scheduled tasks
0 * * * * claude -p "generate status report" > report.md

Codex: Automation Patterns

# GitHub issue automation
# (configured through ChatGPT/Codex integrations)
# Assign issue to @codex to trigger cloud agent

# Local interactive
codex "implement the feature described in issue #42"

# Trusted unattended job with explicit controls
codex -a never --sandbox workspace-write "add input validation to all API endpoints"

# Quick review
codex "what would break if I remove this function?"

Most Used Commands

Cursor Top 5

@codebase — Search entire project
@web — Web search for docs
/Generate Cursor Rules — Auto-generate rules
Cmd/Ctrl+K — Inline edit
cursor . — Open project

Claude Code Top 5

claude — Start session
/clear — Fresh context
/compact — Summarize conversation
claude -p "task" — Quick one-off task
/cost — Check usage

Codex Top 5

codex — Start session
codex "prompt" — Start with task
/model — Switch model
codex -a on-request — Interactive approval policy
codex --sandbox workspace-write — Set sandbox separately

Frequently asked questions

How do you start Claude Code and Codex from the command line?

claude starts interactive mode and claude -p runs a one-off, non-interactive query. codex starts an interactive REPL, and codex "prompt" starts with an initial prompt. Both accept --model to select the model.

Which Cursor context references get used most?

@codebase searches the entire project and @web returns web search results — the top two of the Cursor list. @files, @folders, @symbols, and @git scope context to specific files, directories, functions, or Git history.

Which Codex approval mode should be used?

untrusted prompts for commands outside the trusted set, on-request lets the agent request an elevated action, and never never prompts while disallowed sandbox operations fail. --full-auto is a deprecated compatibility shortcut, and --dangerously-bypass-approvals-and-sandbox (--yolo) is for hardened, isolated runners only.

How are custom slash commands created in Claude Code?

Custom commands are markdown files in .claude/commands/. $ARGUMENTS passes parameters from the slash command, allowed-tools restricts which tools it can use, argument-hint shows hint text for the argument, @file.js references files inline, and !command executes bash commands inline.

What replaced the Claude Code thinking-budget ladder?

The think / think harder / ultrathink budget ladder is gone. think, think hard, and think more are ordinary text, while ultrathink is a one-turn hint that adds a deeper-reasoning instruction without changing API effort. Durable depth is controlled with the thinking toggle and /effort.