COMMAND REFERENCE▌

Summarize the visible conversation to free tokens. Use after long runs so Codex retains key points without blowing the context window.

Show the Git diff including files Git isn't tracking yet. Review Codex's edits before committing or running tests.

Toggle the Fast service tier for the current model/thread, or check current state.

Fork the current conversation into a new thread with a fresh ID, leaving the original transcript untouched for parallel exploration.

Set a persistent task goal (max 4,000 chars). Codex tracks it across turns. Subcommands: pause, resume, clear, view.

Include open files, current selection, and other IDE context in the next prompt without re-explaining what's open.

Generate an AGENTS.md scaffold in the current directory for persistent repo instructions.

List configured Model Context Protocol (MCP) tools. Pass 'verbose' for server diagnostics.

Attach a file or folder to the conversation via popup picker so Codex references it directly in the next turn.

Choose the active model (and reasoning effort when available). Opens a picker; press s to switch for the current session only.

Start a new conversation inside the same CLI session without clearing the terminal view.

Set what Codex can do without asking first — switch between Auto, Read Only, or custom permission profiles mid-session.

Choose a communication style for responses — friendly, pragmatic, or none — without rewriting your instructions.

Switch to plan mode and optionally send a prompt. Codex proposes an execution plan before implementation starts.

Resume a saved conversation from your session list. Continue work from a previous CLI session without starting over.

Ask Codex to review your working tree — summarizes issues, focusing on behavior changes and missing tests.

Start an ephemeral side conversation for a focused detour — the parent thread keeps running and its status stays visible.

Browse and use local skills to improve task-specific behavior by loading relevant skill context.

Display session configuration and token usage — active model, approval policy, writable roots, and context capacity.

Switch to Ask mode for read-only exploration — the agent searches and answers questions without editing any files.

Toggle auto-run (default) or set explicitly on/off/status. Controls whether tools run without asking for confirmation.

Create or edit custom slash commands for repetitive workflows.

Summarize conversation to free context window space while keeping a high-level record of what happened.

Share feedback with the Cursor team directly from the CLI.

Sign out from Cursor.

Toggle max mode on models that support it — enables extended reasoning and capability.

Browse, enable, and configure MCP servers. Subcommands: list, enable <name>, disable <name>.

Set or list available models for the current session.

Start a new chat session with empty context.

Switch to Plan mode — the agent researches your codebase, asks clarifying questions, and produces a reviewable implementation plan before writing code.