OpenAI Codex CLI cheat sheet

Every command, flag and slash command for codex — OpenAI's open-source terminal coding agent (Rust) — a sandboxed TUI plus a headless `exec` mode. 17+ commands, verified July 2026.

Older guides get these wrong: --full-auto is removed (use --sandbox instead); the slash command is /permissions, not /approvals; and codex login --api-key is deprecated — pipe it: printenv OPENAI_API_KEY | codex login --with-api-key.

Install

curl -fsSL https://chatgpt.com/codex/install.sh | sh                  # macOS / Linux
powershell -c "irm https://chatgpt.com/codex/install.ps1 | iex"       # Windows
npm install -g @openai/codex                                         # npm
brew install --cask codex                                            # Homebrew

Start & run

codex                           # interactive TUI in the current dir
codex "refactor auth and add tests"   # TUI with an initial prompt
codex exec "fix the failing tests"    # headless / non-interactive
codex resume --last             # resume the most recent session
codex review                    # non-interactive review of current changes
codex login  ·  codex logout    # auth management

Key flags

Flag	What it does
`-m, --model <MODEL>`	Model to use (e.g. `gpt-5.6`)
`-s, --sandbox <MODE>`	`read-only` · `workspace-write` · `danger-full-access`
`-a, --ask-for-approval <p>`	`untrusted` · `on-request` · `never` (TUI)
`-C, --cd <DIR>`	Set the agent's working root
`-i, --image <FILE>`	Attach image(s) to the prompt
`-c, --config key=value`	Override any `config.toml` value inline
`--json`	(exec) stream JSONL events to stdout
`-o, --output-last-message <FILE>`	(exec) write the agent's final message to a file
`--dangerously-bypass-approvals-and-sandbox`	No approvals, no sandbox — isolated envs only

In-session slash commands

Command	What it does
`/model`	Choose model and reasoning effort
`/permissions`	Choose approvals + sandbox
`/init`	Create an `AGENTS.md` for the project
`/compact`	Summarize the conversation to avoid the context limit
`/review`	Review current changes and find issues
`/diff` · `/status`	Show git diff · session config + token usage
`/mcp`	List configured MCP tools
`/new`	Start a new chat within the conversation

Config

# ~/.codex/config.toml   (CODEX_HOME overrides ~/.codex)
model = "gpt-5.6"
approval_policy = "on-request"      # untrusted | on-request | never
sandbox_mode  = "workspace-write"   # read-only | workspace-write | danger-full-access

[mcp_servers.docs]                  # external tools for the agent
command = "npx"
args = ["-y", "@modelcontextprotocol/server-filesystem", "/path"]

# auth: OPENAI_API_KEY, or `codex login` with a ChatGPT plan

Common one-liners

codex exec "fix the failing unit tests" --sandbox workspace-write
codex exec -m gpt-5.6 -o result.txt "summarize the changes on this branch"
codex exec --json --skip-git-repo-check "run the linter and report findings"
codex -i mockup.png "implement this screen"            # attach a UI mock

More CLI cheat sheets

Stop memorizing flags

Run OpenAI Codex CLI — and every other agent — from one app

TheVibeFather orchestrates OpenAI Codex CLI alongside every other coding CLI in one native macOS app. Bring the models you already pay for; keep your keys.

Get started See how it works