{"id":163001,"date":"2026-03-20T16:33:41","date_gmt":"2026-03-20T13:33:41","guid":{"rendered":"https:\/\/computingforgeeks.com\/codex-cli-cheat-sheet\/"},"modified":"2026-03-22T03:24:15","modified_gmt":"2026-03-22T00:24:15","slug":"codex-cli-cheat-sheet","status":"publish","type":"post","link":"https:\/\/computingforgeeks.com\/codex-cli-cheat-sheet\/","title":{"rendered":"OpenAI Codex CLI Cheat Sheet – Commands, Shortcuts, Tips"},"content":{"rendered":"\n

Codex CLI<\/a> is OpenAI’s terminal-based coding agent. Built in Rust, it reads your codebase, edits files, runs commands, and manages git workflows from the command line. Codex runs with OS-native sandboxing (Seatbelt on macOS, Bubblewrap on Linux) to safely execute commands without risking your system. This cheat sheet covers every command, shortcut, configuration option, and workflow pattern for Codex CLI.<\/p>\n\n\n\n

\"OpenAI<\/figure>\n\n\n\n

Installation<\/h2>\n\n\n\n

Install Codex CLI via npm, Homebrew, or direct binary download:<\/p>\n\n\n\n

# npm (all platforms)\nnpm install -g @openai\/codex\n\n# Homebrew (macOS)\nbrew install --cask codex\n\n# Verify\ncodex --version<\/code><\/pre>\n\n\n\n
Authenticate with your ChatGPT account or API key:<\/p>\n\n\n\n
# Browser-based login (default)\ncodex\n\n# API key login\nprintenv OPENAI_API_KEY | codex login --with-api-key\n\n# Headless\/SSH login\ncodex login --device-auth<\/code><\/pre>\n\n\n\n
Keyboard Shortcuts<\/h2>\n\n\n\n
Shortcut<\/th>Action<\/th><\/tr><\/thead>
Ctrl+C<\/code><\/td>Cancel current operation (press twice to quit)<\/td><\/tr>
Ctrl+D<\/code><\/td>Exit Codex (press twice to force quit)<\/td><\/tr>
Ctrl+L<\/code><\/td>Clear terminal screen<\/td><\/tr>
Ctrl+G<\/code><\/td>Open prompt in external editor (vim, etc.)<\/td><\/tr>
Enter<\/code><\/td>During execution – inject new instructions<\/td><\/tr>
Tab<\/code><\/td>Queue a follow-up prompt while agent is working<\/td><\/tr>
Esc + Esc<\/code><\/td>Edit your previous message (when composer is empty)<\/td><\/tr>
Esc<\/code><\/td>Close navigation drawer<\/td><\/tr>
Up\/Down<\/code><\/td>Navigate draft history<\/td><\/tr>
@<\/code><\/td>Fuzzy file search – attach files to conversation<\/td><\/tr>
!<\/code><\/td>Execute local shell command directly<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n
Slash Commands<\/h2>\n\n\n\n
Session Management<\/h3>\n\n\n\n
Command<\/th>What It Does<\/th><\/tr><\/thead>
\/clear<\/code><\/td>Reset UI and conversation<\/td><\/tr>
\/compact<\/code><\/td>Summarize conversation to free token space<\/td><\/tr>
\/new<\/code><\/td>Start fresh conversation in same session<\/td><\/tr>
\/resume<\/code><\/td>Resume a saved conversation<\/td><\/tr>
\/fork<\/code><\/td>Clone current conversation to new thread<\/td><\/tr>
\/copy<\/code><\/td>Copy latest Codex output to clipboard<\/td><\/tr>
\/diff<\/code><\/td>Display git diff including untracked files<\/td><\/tr>
\/status<\/code><\/td>Show session config and token usage<\/td><\/tr>
\/exit<\/code> \/ \/quit<\/code><\/td>Exit Codex CLI<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n
Configuration and Model<\/h3>\n\n\n\n
Command<\/th>What It Does<\/th><\/tr><\/thead>
\/model<\/code><\/td>Choose model and reasoning effort level<\/td><\/tr>
\/fast<\/code><\/td>Toggle fast mode for GPT-5.4<\/td><\/tr>
\/permissions<\/code><\/td>Set what Codex can do without asking<\/td><\/tr>
\/theme<\/code><\/td>Choose syntax highlighting theme<\/td><\/tr>
\/personality<\/code><\/td>Set communication style: friendly, pragmatic, or none<\/td><\/tr>
\/statusline<\/code><\/td>Configure footer status bar items<\/td><\/tr>
\/experimental<\/code><\/td>Toggle experimental features<\/td><\/tr>
\/debug-config<\/code><\/td>Print config layer diagnostics<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n
Code and Project Tools<\/h3>\n\n\n\n
Command<\/th>What It Does<\/th><\/tr><\/thead>
\/init<\/code><\/td>Generate AGENTS.md scaffold with project context<\/td><\/tr>
\/review<\/code><\/td>Code review of working tree changes<\/td><\/tr>
\/plan<\/code><\/td>Switch to plan mode (read-only exploration)<\/td><\/tr>
\/mention<\/code><\/td>Attach files to conversation<\/td><\/tr>
\/mcp<\/code><\/td>List configured MCP tools<\/td><\/tr>
\/agent<\/code><\/td>Switch active agent thread<\/td><\/tr>
\/ps<\/code><\/td>Show background terminals and output<\/td><\/tr>
\/feedback<\/code><\/td>Submit logs to maintainers<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n
CLI Subcommands<\/h2>\n\n\n\n
Command<\/th>What It Does<\/th><\/tr><\/thead>
codex<\/code><\/td>Start interactive TUI session<\/td><\/tr>
codex exec<\/code> \/ codex e<\/code><\/td>Non-interactive mode (like claude -p<\/code>)<\/td><\/tr>
codex review<\/code><\/td>Non-interactive code review<\/td><\/tr>
codex resume<\/code><\/td>Resume a previous session<\/td><\/tr>
codex fork<\/code><\/td>Fork a previous session into new thread<\/td><\/tr>
codex apply<\/code> \/ codex a<\/code><\/td>Apply latest diff as git apply<\/code><\/td><\/tr>
codex login<\/code><\/td>Manage authentication<\/td><\/tr>
codex mcp<\/code><\/td>Manage MCP servers<\/td><\/tr>
codex mcp-server<\/code><\/td>Run Codex as an MCP server (stdio)<\/td><\/tr>
codex sandbox<\/code><\/td>Run commands within Codex sandbox<\/td><\/tr>
codex features list<\/code><\/td>List all feature flags<\/td><\/tr>
codex cloud<\/code><\/td>Browse and apply Codex Cloud tasks<\/td><\/tr>
codex app<\/code><\/td>Launch desktop app (macOS)<\/td><\/tr>
codex completion bash<\/code><\/td>Generate shell completions<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n
CLI Flags Reference<\/h2>\n\n\n\n
Core Flags<\/h3>\n\n\n\n
Flag<\/th>Purpose<\/th><\/tr><\/thead>
-m, --model <MODEL><\/code><\/td>Set model (gpt-5.4, gpt-5.4-mini, gpt-5.3-codex)<\/td><\/tr>
-i, --image <FILE><\/code><\/td>Attach image(s) to initial prompt<\/td><\/tr>
-C, --cd <DIR><\/code><\/td>Set working root directory<\/td><\/tr>
--add-dir <DIR><\/code><\/td>Add writable directories<\/td><\/tr>
-p, --profile <NAME><\/code><\/td>Use named config profile<\/td><\/tr>
--search<\/code><\/td>Enable live web search<\/td><\/tr>
--oss<\/code><\/td>Use local open-source model (LM Studio\/Ollama)<\/td><\/tr>
-c, --config <key=value><\/code><\/td>Override config.toml values<\/td><\/tr>
--version<\/code><\/td>Show version number<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n
Security and Approval Flags<\/h3>\n\n\n\n
Flag<\/th>Purpose<\/th><\/tr><\/thead>
-s, --sandbox <MODE><\/code><\/td>Sandbox: read-only, workspace-write, danger-full-access<\/td><\/tr>
-a, --ask-for-approval <MODE><\/code><\/td>Approval: untrusted, on-request, never<\/td><\/tr>
--full-auto<\/code><\/td>Auto-approve within sandbox (on-request + workspace-write)<\/td><\/tr>
--yolo<\/code><\/td>No sandbox, no approvals (dangerous)<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n
Non-Interactive (codex exec) Flags<\/h3>\n\n\n\n
# Run a task non-interactively\ncodex exec \"fix the failing tests\"\n\n# Output JSON for scripting\ncodex exec --json \"list all API endpoints\"\n\n# Save final response to file\ncodex exec -o result.txt \"summarize the architecture\"\n\n# Enforce JSON schema on output\ncodex exec --output-schema schema.json \"extract metadata\"\n\n# Resume previous session with follow-up\ncodex exec resume --last \"now add error handling\"\n\n# Full-auto mode in CI\/CD\ncodex exec --full-auto \"run tests and fix failures\"\n\n# Set spending budget\ncodex exec --full-auto --config model=gpt-5.4-mini \"fix linting errors\"<\/code><\/pre>\n\n\n\n
Sandbox Modes<\/h2>\n\n\n\n
Codex uses OS-native sandboxing – Seatbelt on macOS, Bubblewrap on Linux, restricted tokens on Windows:<\/p>\n\n\n\n
Mode<\/th>File Access<\/th>Network<\/th>Use Case<\/th><\/tr><\/thead>
read-only<\/strong><\/td>Read anywhere, no writes<\/td>Disabled<\/td>Safe exploration, code review<\/td><\/tr>
workspace-write<\/strong><\/td>Read\/write in project dir<\/td>Disabled<\/td>Normal coding (default for git repos)<\/td><\/tr>
danger-full-access<\/strong><\/td>Full filesystem access<\/td>Enabled<\/td>System admin tasks (use with caution)<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n
Approval Policies<\/h2>\n\n\n\n
Policy<\/th>Behavior<\/th><\/tr><\/thead>
untrusted<\/strong><\/td>Only known-safe read commands auto-approved, everything else prompts<\/td><\/tr>
on-request<\/strong><\/td>Model decides when to ask (default with –full-auto)<\/td><\/tr>
never<\/strong><\/td>Never asks – failures returned to model silently<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n
Common combinations:<\/p>\n\n\n\n
# Safe exploration (read-only, always ask)\ncodex -s read-only -a untrusted\n\n# Normal coding (default for git projects)\ncodex -s workspace-write -a untrusted\n\n# Full auto (shortcut for workspace-write + on-request)\ncodex --full-auto\n\n# No safety rails (dangerous)\ncodex --yolo<\/code><\/pre>\n\n\n\n
Models and Reasoning<\/h2>\n\n\n\n
Model<\/th>Best For<\/th>Speed<\/th><\/tr><\/thead>
gpt-5.4<\/strong><\/td>Complex reasoning, architecture, multi-file changes<\/td>Slower<\/td><\/tr>
gpt-5.4-mini<\/strong><\/td>Quick edits, simple tasks, lower cost<\/td>Fast<\/td><\/tr>
gpt-5.3-codex<\/strong><\/td>Specialized coding tasks<\/td>Medium<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n
Reasoning effort levels control how deeply the model thinks:<\/p>\n\n\n\n
Level<\/th>Use Case<\/th><\/tr><\/thead>
minimal<\/code><\/td>Trivial lookups<\/td><\/tr>
low<\/code><\/td>Simple questions, quick fixes<\/td><\/tr>
medium<\/code><\/td>Default – balanced<\/td><\/tr>
high<\/code><\/td>Complex debugging, architecture<\/td><\/tr>
xhigh<\/code><\/td>Hardest problems<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n
Switch with \/model<\/code> in interactive mode or -m gpt-5.4-mini<\/code> on the command line.<\/p>\n\n\n\n
AGENTS.md – Project Instructions<\/h2>\n\n\n\n
AGENTS.md is Codex’s equivalent of Claude Code’s CLAUDE.md. It provides project-specific instructions loaded at the start of every session.<\/p>\n\n\n\n
vi AGENTS.md<\/code><\/pre>\n\n\n\n
Example project instructions:<\/p>\n\n\n\n
# My Project\n\n## Build Commands\n- Test: `npm test`\n- Build: `npm run build`\n- Lint: `npm run lint -- --fix`\n\n## Rules\n- Use TypeScript strict mode for all new files\n- Write unit tests for every new function\n- Never commit directly to main branch\n- All API endpoints must have OpenAPI docs<\/code><\/pre>\n\n\n\n
Generate one automatically with \/init<\/code>.<\/p>\n\n\n\n
Configuration (config.toml)<\/h2>\n\n\n\n
Codex uses TOML config files. Precedence (highest to lowest): CLI flags, profile values, project config (.codex\/config.toml<\/code>), user config (~\/.codex\/config.toml<\/code>), system config (\/etc\/codex\/config.toml<\/code>), defaults.<\/p>\n\n\n\n
vi ~\/.codex\/config.toml<\/code><\/pre>\n\n\n\n
Common settings:<\/p>\n\n\n\n
# Model and reasoning\nmodel = \"gpt-5.4\"\nmodel_reasoning_effort = \"high\"\n\n# Security defaults\nsandbox_mode = \"workspace-write\"\napproval_policy = \"on-request\"\n\n# Personality\npersonality = \"pragmatic\"\n\n# Web search\nweb_search = \"cached\"\n\n# Performance\nservice_tier = \"fast\"\n\n# Developer instructions (like AGENTS.md but in config)\ndeveloper_instructions = \"Always use TypeScript. Prefer functional patterns.\"\n\n# MCP servers\n[mcp_servers.github]\ncommand = [\"npx\", \"-y\", \"@modelcontextprotocol\/server-github\"]\n\n# Named profiles\n[profiles.fast]\nmodel = \"gpt-5.4-mini\"\nmodel_reasoning_effort = \"low\"\nservice_tier = \"fast\"\n\n[profiles.thorough]\nmodel = \"gpt-5.4\"\nmodel_reasoning_effort = \"xhigh\"<\/code><\/pre>\n\n\n\n
Use profiles with codex -p fast<\/code> or codex -p thorough<\/code>.<\/p>\n\n\n\n
MCP Servers<\/h2>\n\n\n\n
Codex supports MCP<\/a> for connecting to external tools. Configure in ~\/.codex\/config.toml<\/code>:<\/p>\n\n\n\n
# STDIO transport (local process)\n[mcp_servers.github]\ncommand = [\"npx\", \"-y\", \"@modelcontextprotocol\/server-github\"]\nenabled = true\nstartup_timeout_sec = 30\n\n# HTTP transport (remote server)\n[mcp_servers.remote-db]\nurl = \"https:\/\/my-mcp-server.example.com\"<\/code><\/pre>\n\n\n\n
Codex can also run as an MCP server itself:<\/p>\n\n\n\n
codex mcp-server<\/code><\/pre>\n\n\n\n
Skills System<\/h2>\n\n\n\n
Skills extend Codex with reusable task-specific capabilities. They are scanned from multiple locations:<\/p>\n\n\n\n