Complete Command & Slash Command Reference

This lesson is a comprehensive reference for every Codex CLI flag, configuration key, and slash command. Bookmark it and return as needed — this is not meant to be memorized, but to be consulted when you need the exact syntax or option name.

CLI Flags — Launch Options

These flags are passed when starting Codex from the terminal. They control the session's initial configuration.

codex [prompt]                  Start interactive session (optional initial prompt)
codex --suggest [prompt]        Suggest mode — ask permission for everything (default)
codex --auto-edit [prompt]      Auto-edit mode — auto-apply file edits, ask for commands
codex --full-auto [prompt]      Full-auto mode — execute everything without asking
codex exec [prompt]             One-shot execution — run task and exit
codex resume                    Resume the most recent session
codex fork                      Fork from a previous session into a new one
codex cloud [prompt]            Delegate task to a Codex Cloud sandbox
codex cloud exec [prompt]       One-shot cloud execution
codex cloud status              Check status of cloud tasks
codex cloud diff <id>           View diff from a cloud task
codex review --base <branch>    Code review against a base branch
codex review --json             Output review results as JSON
codex mcp list                  List installed MCP servers
codex mcp add <name> <cmd>      Add an MCP server
codex mcp remove <name>         Remove an MCP server
codex features list             List feature flags
codex features enable <flag>    Enable a feature flag
codex features disable <flag>   Disable a feature flag
codex --login                   Authenticate via ChatGPT OAuth
codex --model <model>           Set model for the session
codex --reasoning-effort <lvl>  Set reasoning effort (low/medium/high/xhigh)
codex -i <path>                 Attach an image file for multimodal input
codex --mcp-server              Run Codex as an MCP server
codex --version                 Print version

Slash Commands — In-Session Actions

These commands are typed inside an active Codex session. They control the session state and trigger built-in workflows.

/model [name]     Switch model mid-session (e.g., /model o3)
/plan             Enter plan mode — analyze and propose before executing
/compact           Compress context — summarize conversation history
/diff              Show the current diff of all uncommitted changes
/review [file]     Review current changes or a specific file for issues
/mention <file>    Add a file to context without modifying it
/copy              Copy the last response to clipboard
/clear             Clear conversation history and start fresh
/fork              Branch into a new session from the current point
/status            Show session info: model, tokens used, approval mode
/help              Show all available commands and shortcuts

Approval Modes

The three approval modes control Codex's autonomy level. Choose based on trust and task risk.

suggest     (default) Ask permission for file edits AND shell commands
            Safest mode. You see every change before it happens.
            Best for: unfamiliar codebases, high-risk changes, learning.

auto-edit   Auto-apply file edits, ask permission for shell commands
            Good balance of speed and safety.
            Best for: routine coding in a familiar codebase.

full-auto   Execute everything without asking
            Maximum speed, requires trust in the sandbox.
            Best for: well-scoped tasks with clean git state.

Sandbox Settings

Sandbox configuration controls what Codex can access at the OS level.

Sandbox Mode        Description
─────────────────────────────────────────────────────────────
workspace-write     Can write to project directory only (default)
read-only           Can only read files, no writes allowed
no-sandbox          Disable sandbox entirely (not recommended)

Execpolicy Rules (in config.toml)
─────────────────────────────────────────────────────────────
"command" = "allow"             Allow the command in sandbox
"command" = "allow-network"     Allow with network access
"command" = "deny"              Block the command entirely

config.toml Reference

The configuration file lives at ~/.codex/config.toml and controls default behavior.

# Model defaults
[profile.default]
model = "codex-1"              # Default model
approval = "suggest"            # Default approval mode
reasoning_effort = "medium"     # For o-series models

# Project-doc size limit
project_doc_max_bytes = 100000  # Max AGENTS.md size before truncation

# MCP servers
[mcp.github]
command = "npx"
args = ["-y", "@modelcontextprotocol/server-github"]
env = { GITHUB_TOKEN = "ghp_..." }

# Sub-agent definitions
[agents.test-writer]
model = "gpt-4.1"
description = "Writes test suites"

# Exec policy
[execpolicy]
"npm install" = "allow-network"
"git *" = "allow-network"
"curl *" = "allow-network"