OpenAI Codex Command Guide

1. Install, launch, and update

Command	What it does
npm i -g @openai/codex	Installs Codex CLI globally.
codex	Launches the interactive Codex CLI/TUI.
npm i -g @openai/codex@latest	Updates Codex CLI to the latest published version.

OpenAI’s CLI docs say Codex runs on macOS, Windows, and Linux, and that first launch prompts you to sign in.

---

2. CLI slash commands

In the interactive CLI, type / to open slash commands. While Codex is working, you can also queue commands with Tab.

Main slash commands

Slash command	What it does	Notes
/agent	Switch the active agent thread.	Used for multi-agent workflows.
/apps	Manage apps.	Current-source command.
/approve	Approve one retry of the latest auto-review denial.	Current-source command.
/approvals	Legacy alias for permissions.	Docs say it still works, but /permissions is preferred.
/clear	Clear the conversation and start a new one.	Different from Ctrl+L, which only clears the terminal view.
/clean	Alias for /stop.	Current-source alias.
/collab	Change collaboration mode.	Experimental/current-source command.
/compact	Summarize conversation context to save tokens.	Useful before long sessions.
/copy	Copy the last assistant response as Markdown.	Not available on every platform.
/debug-config	Show configuration layers and the source of each effective value.	Useful when config behavior is confusing.
/debug-m-drop	Memory debug command.	Internal/debug; avoid for normal guide users.
/debug-m-update	Memory debug command.	Internal/debug; avoid for normal guide users.
/diff	Show Git diff, including untracked files.	Good before review or commit prep.
/exit	Exit Codex.	Alias of /quit.
/experimental	Toggle experimental features.	Used for things like goals.
/fast	Toggle Fast mode.	Supports /fast on, /fast off, /fast status.
/feedback	Send feedback and logs to maintainers.	CLI/app/IDE surfaces include this.
/fork	Fork the current conversation into a new thread.	Useful for alternate solution paths.
/goal	Set, view, pause, resume, or clear a long-running goal.	Experimental; details below.
/hooks	View and manage lifecycle hooks.	Current-source command.
/ide	Include current IDE selection, open files, or IDE context.	Current-source command.
/init	Create an AGENTS.md instruction file.	Common first command in a repo.
/keymap	Configure TUI keyboard shortcuts.	Supports inline arguments.
/logout	Log out of Codex.	CLI command also exists.
/mcp	List MCP tools and resources.	/mcp verbose shows details.
/memories	Configure memory use and memory generation.	Current-source command.
/mention	Mention a file.	Alternative to @ file search.
/model	Select model and reasoning effort.	Also available via CLI flags.
/new	Start a new chat during the current session.	Current-source command.
/permissions	Choose what Codex is allowed to do.	Preferred over legacy /approvals.
/personality	Set communication style.	Docs list friendly, pragmatic, and none.
/plan	Switch to Plan mode.	Useful before coding or making changes.
/plugins	Browse or manage plugins.	Current-source command.
/ps	List background terminals.	Pair with /stop.
/quit	Exit Codex.	Same as /exit.
/raw	Toggle raw scrollback mode.	Useful for copy-friendly terminal selection.
/realtime	Toggle realtime voice mode.	Experimental/current-source command.
/rename	Rename the current thread.	Supports inline arguments.
/resume	Resume a saved chat.	CLI command also exists.
/review	Review current changes and identify issues.	Can be used inline, for example /review staged changes.
/rollout	Print rollout file path.	Debug-build/current-source command.
/sandbox-add-read-dir	Allow the sandbox to read an additional directory.	Platform-gated; docs note Windows usage.
/settings	Configure realtime microphone/speaker settings.	Current-source command.
/setup-default-sandbox	Set up an elevated agent sandbox.	Specialized/current-source command.
/side	Start a side conversation as an ephemeral fork.	Details below.
/skills	Use skills.	Current-source command; skills are replacing older custom prompts.
/status	Show current session config and token usage.	CLI/app/IDE surfaces include this.
/statusline	Configure status-line items.	Useful for persistent session info.
/stop	Stop running background terminals.	/clean aliases to this.
/subagents	Switch active subagent thread.	Current-source sibling/alias for multi-agent use.
/test-approval	Test approval request behavior.	Debug-build/current-source command.
/theme	Choose syntax highlighting theme.	Current-source command.
/title	Configure terminal title items.	Useful when many sessions are open.
/vim	Toggle Vim mode.	Current-source command.

OpenAI’s public slash-command docs list the primary CLI commands, while the current CLI source enumerates additional commands such as /theme, /skills, /memories, /rename, /vim, /hooks, /raw, and debug/platform-gated items.

---

3. High-value slash command patterns

/goal

Use /goal for a long-running objective that Codex should continue working toward.

/goal Complete the migration to TypeScript and keep going until tests pass.

Useful forms:

/goal <objective>
/goal
/goal pause
/goal resume
/goal clear

OpenAI documents /goal as experimental and notes it requires the goals feature to be enabled. The docs also recommend clear goals with an objective, done condition, scope, and constraints.

---

/side

Use /side to open a temporary side conversation without permanently derailing the main thread.

/side explain why this test is failing

Typical uses:

/side summarize the current plan
/side what are the risks in this migration?
/side explain this error without changing files

OpenAI describes /side as an ephemeral fork of the conversation. It is not available inside nested side or review modes.

---

/mcp

Use /mcp to inspect MCP tools and resources available to Codex.

/mcp
/mcp verbose

/mcp verbose shows expanded MCP details.

---

/fast

/fast on
/fast off
/fast status

Fast mode toggles faster behavior where supported.

---

/personality

/personality friendly
/personality pragmatic
/personality none

OpenAI’s docs list those three personality modes.

---

4. CLI keyboard and input shortcuts

Shortcut or prefix	What it does
/	Open slash-command menu.
@	Fuzzy-search files; use Tab or Enter to insert a path.
!<command>	Run a local shell command.
Enter while Codex is running	Inject instructions into the current turn.
Tab while Codex is running	Queue a follow-up instruction or command for the next turn.
Ctrl+G	Open the prompt composer in your external editor using VISUAL or EDITOR.
Ctrl+L	Clear the terminal view only.
/clear	Clear the conversation and start a new chat.
Esc twice with an empty composer	Edit the previous user message; continuing to press Esc walks backward.
Enter after editing a previous message	Fork from that point.

OpenAI’s CLI tips document the @, !, Enter, Tab, Ctrl+G, and Esc behaviors. The slash docs distinguish /clear from Ctrl+L.

---

5. Top-level Codex CLI commands

Command	Alias	What it does
codex	—	Launches the interactive TUI.
codex app	—	Opens the Codex desktop app.
codex app-server	—	Runs the experimental app server.
codex apply	codex a	Applies a patch generated by Codex.
codex cloud	codex cloud-tasks	Works with cloud tasks.
codex completion	—	Generates shell completion scripts.
codex debug app-server send-message-v2	—	Debug app-server message command.
codex debug models	—	Debug model listing.
codex exec	codex e	Runs Codex non-interactively.
codex execpolicy	—	Tests command policy behavior.
codex features	—	Lists, enables, or disables feature flags.
codex fork	—	Forks a saved session.
codex login	—	Logs in.
codex logout	—	Logs out.
codex mcp	—	Manages MCP servers.
codex mcp-server	—	Runs Codex MCP server.
codex plugin marketplace	—	Manages plugin marketplaces.
codex resume	—	Resumes a saved session.
codex sandbox	—	Runs a command under Codex sandboxing helpers.
codex update	—	Updates Codex.

OpenAI’s CLI reference lists these command groups and marks several as stable or experimental.

---

6. Global CLI flags

These can be used with top-level codex where applicable.

Flag	Values / shape	What it does
--add-dir	path	Adds a directory as writable or readable context.
--ask-for-approval, -a	untrusted, on-request, never	Sets approval policy.
--cd, -C	path	Sets working directory.
--config, -c	key=value	Overrides a config value.
--dangerously-bypass-approvals-and-sandbox	boolean	Disables approvals and sandboxing.
--yolo	boolean	Alias for bypassing approvals and sandboxing.
--disable	feature	Disables a feature flag.
--enable	feature	Enables a feature flag.
--image, -i	path or comma-separated paths	Adds image input.
--model, -m	model name	Selects model.
--no-alt-screen	boolean	Avoids alternate screen behavior.
--oss	boolean	Uses the local open-source model provider.
--profile, -p	profile name	Uses a config profile.
--remote	ws://... or wss://...	Connects to a remote endpoint.
--remote-auth-token-env	env var name	Reads remote auth token from an environment variable.
--sandbox, -s	read-only, workspace-write, danger-full-access	Sets sandbox policy.
--search	boolean	Enables web search where supported.
PROMPT	string	Starts Codex with an initial prompt.

OpenAI’s CLI reference lists these global options.

---

7. Important subcommand details

codex exec

Use this for non-interactive automation.

codex exec "fix the failing tests"
codex exec -m gpt-5.1-codex --sandbox workspace-write "review this repo"

Common options include:

Option	What it does
--cd	Set working directory.
--color	always, never, or auto.
--dangerously-bypass-approvals-and-sandbox, --yolo	Disable approvals and sandboxing.
--ephemeral	Run without saving the session.
--full-auto	Deprecated full-auto behavior.
--ignore-rules	Ignore rules.
--ignore-user-config	Ignore user config.
--image	Include images.
--json, --experimental-json	Emit JSON events.
--model	Select model.
--oss	Use local OSS provider.
--output-last-message, -o	Save last message to a file.
--output-schema	Constrain output to a schema.
--profile	Use config profile.
--sandbox	Set sandbox.
--skip-git-repo-check	Skip Git repo check.
-c, --config	Override config.
PROMPT or -	Prompt string or stdin.

OpenAI’s CLI reference documents these codex exec flags, plus codex exec resume.

---

codex cloud

Common forms:

codex cloud "investigate this issue"
codex cloud list

Useful options:

Option	What it does
--attempts	Sets 1–4 attempts.
--env	Selects environment ID.
--cursor	Pagination cursor for listing.
--json	JSON output.
--limit	Number of cloud tasks to list.

OpenAI’s reference documents codex cloud, codex cloud exec, and codex cloud list.

---

codex features

codex features list
codex features enable <feature>
codex features disable <feature>

Use this to inspect or change feature flags.

---

codex mcp

Command	What it does
codex mcp add <name> -- <command...>	Adds a stdio MCP server.
codex mcp add <name> --url <url>	Adds a URL-based MCP server.
codex mcp get <name>	Shows one MCP server.
codex mcp list	Lists MCP servers.
codex mcp login <name>	Authenticates an MCP server.
codex mcp logout <name>	Logs out of an MCP server.
codex mcp remove <name>	Removes an MCP server.

Useful options include --env, --bearer-token-env-var, --json, and --scopes.

---

codex plugin marketplace

Command	What it does
codex plugin marketplace add <source>	Adds a plugin marketplace.
codex plugin marketplace remove <marketplace-name>	Removes a marketplace.
codex plugin marketplace upgrade [marketplace-name]	Upgrades one or all marketplaces.

Options include --ref and --sparse for marketplace add.

---

codex resume and codex fork

codex resume
codex resume --last
codex resume --all
codex resume <SESSION_ID>

codex fork
codex fork --last
codex fork --all
codex fork <SESSION_ID>

These resume or fork previous sessions.

---

codex sandbox

codex sandbox runs a command inside Codex’s sandbox helper. Subcommands vary by platform, including macOS seatbelt and Linux/Windows modes.

Common options include:

Option	What it does
--cd	Sets working directory.
--config	Overrides config.
--include-managed-config	Includes managed config.
--permissions-profile	Selects permissions profile.
--allow-unix-socket	macOS seatbelt option.
--log-denials	macOS seatbelt option.
COMMAND...	Command to run in the sandbox.

OpenAI’s reference documents platform-specific sandbox helper options and warns that bypassing sandboxing should be used carefully.

---

8. Codex app commands

App slash commands

Slash command	What it does
/feedback	Send feedback.
/mcp	Show MCP information.
/plan-mode	Toggle or use plan mode.
/review	Review changes.
/status	Show status.

The Codex app also exposes enabled skills in the slash-command list and supports $ for skills.

App keyboard shortcuts

Shortcut	Action
Cmd+Shift+P or Cmd+K	Command menu.
Cmd+,	Settings.
Cmd+O	Open folder.
Cmd+[	Navigate back.
Cmd+]	Navigate forward.
Cmd++ or Cmd+=	Increase font size.
Cmd+- or Cmd+_	Decrease font size.
Cmd+B	Toggle sidebar.
Cmd+Option+B	Toggle diff panel.
Cmd+J	Toggle terminal.
Ctrl+L	Clear terminal.
Cmd+N or Cmd+Shift+O	New thread.
Cmd+F	Find in thread.
Cmd+Shift+[	Previous thread.
Cmd+Shift+]	Next thread.
Ctrl+M	Dictation.

OpenAI’s app docs list these shortcuts for macOS.

App deeplinks

Deeplink	What it opens
codex://settings	Settings.
codex://skills	Skills.
codex://automations	Automations.
codex://threads/<thread-id>	Specific thread.
codex://new	New thread.

codex://new supports query parameters such as prompt, originUrl, and path.

---

9. IDE extension commands

IDE slash commands

Slash command	What it does
/auto-context	Configure automatic context.
/cloud	Use cloud behavior.
/cloud-environment	Select or configure cloud environment.
/feedback	Send feedback.
/local	Use local behavior.
/review	Review changes.
/status	Show status.

OpenAI’s IDE docs say typing / in the Codex chat input shows available IDE slash commands.

VS Code command IDs

Command ID	What it does
chatgpt.addToThread	Add current selection to thread.
chatgpt.addFileToThread	Add file to thread.
chatgpt.newChat	New chat; default shortcut Cmd+N on macOS and Ctrl+N on Windows/Linux.
chatgpt.implementTodo	Implement a TODO.
chatgpt.newCodexPanel	Open a new Codex panel.
chatgpt.openSidebar	Open the Codex sidebar.

OpenAI documents these as VS Code command-palette commands that can be bound to shortcuts.

---

10. Custom slash commands and reusable prompts

OpenAI says custom prompts are deprecated in favor of skills, but existing custom prompts can still expose slash commands in the CLI and IDE.

Create a custom prompt slash command

Directory:

~/.codex/prompts

Example file:

~/.codex/prompts/draftpr.md

Example contents:

---
description: Draft a pull request description
argument-hint: FILES=<files> PR_TITLE=<title>
---

Review the following files and draft a concise PR description.

Files: $FILES
Title: $PR_TITLE
All arguments: $ARGUMENTS

After restarting Codex, it appears as:

/prompts:draftpr

Custom prompt files support front matter such as description and argument-hint, positional placeholders like $1 through $9, $ARGUMENTS, named placeholders such as KEY=value, and $$ for a literal dollar sign.

---

11. Safety defaults for a practical guide

For a user-facing Codex guide, I’d include this rule of thumb:

Situation	Recommended setting
Normal repo work	--sandbox workspace-write
Read-only investigation	--sandbox read-only
Trusted automation in isolated environment	Consider stricter explicit approvals first.
Dangerous local bypass	Avoid --dangerously-bypass-approvals-and-sandbox / --yolo unless inside a disposable or isolated environment.

OpenAI’s CLI reference includes safety guidance around sandboxing and warns about bypassing approvals and sandbox restrictions.