Quick Reference

View commands, flags, and help entry points supported by the current version.

Clear the current session context, suitable for starting fresh on a new question.

Compact long session history, retain key summaries, reduce context pressure.

View or declare the files, directories, and background to focus on for this turn.

View file changes for this turn, useful for verifying the scope of changes before committing.

View or switch models, useful for trading off between speed and reasoning intensity.

View the token count and estimated cost for the current session.

View login status, project path, model, permissions, and runtime environment.

Switch working directory so subsequent commands run in the specified project.

Check if CLI, permissions, dependencies, network, and project configuration are working correctly.

Initialize project documentation, generate or update CLAUDE.md.

View, edit, or refresh long-term memory and project memory.

Enter code review mode, focusing on risks, regressions, and test gaps.

View or adjust permission policies for tool calls, file writes, command execution, etc.

Open the configuration entry point to modify model, permissions, output, and project settings.

Log in or switch accounts to restore authorization and subscription status.

Create or switch to an isolated branch, suitable for experimental changes.

Roll back to an earlier conversation checkpoint and choose a different implementation path.

Resume a historical session to continue an unfinished task.

Rename the current session for easier retrieval in history.

Export session content, changes, or summaries for review and handover.

Adjust reasoning intensity; increase for complex design and review tasks.

Plan before executing, suitable for high-risk, multi-file, or unclear requirements tasks.

Insert side-channel information so the assistant remembers supplementary constraints without interrupting the current task.

Execute similar tasks in batches, e.g., batch review, batch document updates.

Periodically execute check tasks, suitable for monitoring logs, CI, or local services.

Schedule timed tasks or reminders so the assistant continues processing at a specified time.

View or run preset workflows, chaining multiple operations together.

Enter deep research mode, suitable for finding information, comparing solutions, and writing reports.

Debug following the 'Reproduce, Locate, Fix, Verify' rhythm.

Structured code review for PRs or local diffs.

Ask the assistant to reduce complexity, compress abstraction and repetitive logic.

Perform security checks around permissions, injection, keys, supply chain, and data leaks.

View, select, or manage available subagents.

View MCP servers, tools, and connection status.

Browse, install, or manage Plugins.

Reload local Plugins, commonly used when debugging Plugins.

Set the current session goal and budget, giving long tasks a clear endpoint.

Generate a brief progress and next-step summary for the current session.

Reduce the frequency of permission confirmations, suitable for high-trust local projects.

Switch or view terminal interaction interface settings.

Narrow the context scope so the assistant focuses on the current file or task.

View or adjust sandbox policies to control command and file system access.

View quota, usage, or subscription information.

Go to the Plugins discovery page to find installable extensions.

Browse installed Plugins and available commands.

Switch between plan mode, auto-edit, or normal input interaction modes.

Switch theme or terminal display style.

Interrupt the current output or ongoing operation.

Exit the current interactive session.

Show or hide the sidebar.

Open the context, file, or command selector.

Search command history or session history.

Open goal, task, or session navigation.

Clear the input line or open the search entry point, depending on the terminal version.

Cancel current input, close popover, or exit selection state.

Browse input history and candidates.

Complete commands, file paths, or candidates.

Execute a prompt once in non-interactive mode, suitable for scripts and CI.

Output results in JSON, for easy script parsing.

Stream JSON events, suitable for long tasks and real-time UIs.

Specify the model name.

Specify Permission Modes, e.g., default, plan, auto-accept edits, or bypass permissions.

Enable sandboxed execution to restrict command and file system side effects.

Run in a more conservative mode, suitable for untrusted projects or demo environments.

Limit maximum turns to avoid infinite loops in script tasks.

Don't save this session state, suitable for temporary batch processing.

Resume a specified or the most recent historical session.

Continue the most recent session.

Specify a subagent to execute tasks.

Load plugins from the specified directory.

Install or load plugins from a URL.

Start in a more lightweight mode, without loading some project decorators and additional context.

Run in an isolated worktree, isolating experimental changes.

Skip permission confirmations; only use in fully trusted isolated environments.

Add extra directories to the readable context.

Specify an MCP config file.

Enable channels provided by plugins or extensions.

Log in to an account or refresh authorization.

Clean up project-level cache, sessions, or index data.

List available subagents.

Output the subagent list in JSON format.

View plugin details, commands, and capability descriptions.

Enable an installed plugin.

Disable an installed plugin.

Project-level instructions: write project structure, commands, constraints, and collaboration rules.

Machine-private project notes, suitable for personal paths and temporary preferences.

Shared project settings, can be committed to the repository.

Machine-private settings, usually not committed.

Global personal instructions, effective across projects.

Global settings, suitable for default models and personal preferences.

Project skills directory, for encapsulating reusable workflows and domain knowledge.

Project subagents directory, for defining roles like review, testing, and writing.

Split rules files, for maintaining project constraints by topic.

Project MCP server configuration.

User-level state and global metadata file.

Add an MCP server.

Add an MCP server with authentication headers.

List configured MCP servers.

View details of a specific MCP server.

Remove an MCP server.

Import MCP configuration from the desktop app.

View MCP connection status and available tools within a session.

MCP tool naming format, typically a 'server + tool' combination.

Triggers before tool execution, suitable for intercepting dangerous commands or supplementing audits.

Triggers after tool execution, suitable for formatting, logging, and syncing state.

Triggers when the user submits a prompt; can inject context or perform compliance checks.

Triggers when the main session stops, suitable for summarizing, notifying, or cleaning up resources.

Triggers when a subagent starts, suitable for preparing isolated directories or environment variables.

Triggers when a subagent stops, suitable for collecting results and cleaning up temporary files.

Hook type determines trigger timing.

Matchers determine which tools, commands, or contexts trigger the hook.

Receives input as command arguments, suitable for calling scripts.

Hooks declared within Skills, only active when the Skill is in use.

Stop event input fields, can be used to read session state and last output.

Subagent stop event input fields, for reading agent name, status, and artifacts.

Default mode, key operations require confirmation.

Plan mode, only discuss approaches, don't modify files directly.

Auto-accept file edits, but high-risk commands may still require confirmation.

More proactively executes routine tool calls, suitable for low-risk tasks.

Bypass permission confirmations, only suitable for trusted isolated environments.

Allow specific commands, paths, or tool patterns.

Deny specific commands, paths, or tool patterns.

Use sandbox boundaries to restrict process, network, and file system access.

Call a specific subagent by name in the conversation.

Specify the default subagent at startup.

Project subagent definition file.

Built-in agents, typically covering review, research, implementation, and testing scenarios.

Let subagents work in isolated worktrees to reduce interference.

Allow subagents to run in the background while the main session continues other tasks.

View, select, and manage available subagents.

Install plugins from a plugin source.

Install plugins from a GitHub repository.

List installed plugins.

Reload local plugins without restarting the CLI during debugging.

Load local plugin directory at startup.

Plugin manifest file declaring name, version, commands, and capabilities.

Plugin command naming convention to avoid conflicts with system commands.

Plugin marketplace or discovery page for finding team and community extensions.

Anthropic API key.

OpenAI or compatible gateway key.

DeepSeek or compatible model provider key.

Default reasoning intensity.

Current session ID, useful for log correlation.

Disable terminal alternate screen for easier copying of historical output.

Force synchronous output, suitable for scripts and log collection.

Control package manager automatic update behavior.

Enable gateway model discovery.

Hide the current working directory, suitable for screen recording or public demos.

Enable safety mode.

Disable built-in skills, only use explicitly configured capabilities.

Disable automatic update checks.

Prioritize PowerShell tools in Windows environments.

Disable workflow capabilities.

Base branch or reference used when creating a worktree.

Specify the Linux bubblewrap sandbox path.

Specify the socat path for sandbox network bridging.

Control whether parent directory settings are inherited.

Set the cleanup cycle for cache, temporary data, or history.

Configure PR link templates for jumping from the terminal to code hosting platforms.

Disable built-in skills.

Restrict the selectable model range, suitable for team-wide cost and compliance policies.

Identify links in the terminal footer and provide navigation.

First increase reasoning intensity, then enter plan mode, implement after confirmation.

First check diff, then check cost, export summary and compact context if necessary.

Delegate similar review tasks to batch processing commands.

Periodically run tests, logs, or health checks.

Create a new branch for experiments, then merge or restore the main line upon success.

Pipe diff to the CLI, output JSON for automated systems to read.

Generate documentation, add rules, then commit to the repository.

Reproduce, locate, add tests, fix, then run verification.

Codex project instructions: write project commands, code standards, and collaboration constraints.

Codex global config file, typically for model, sandbox, and approval policies.

Project or user-level Codex extensions directory, commonly for skills, plugins, and memory.

Control the file system sandbox scope for Codex command execution.

Control when Codex requires user approval for commands or network operations.

Set reasoning intensity; increase for complex tasks.

View commands and flags supported by the Codex CLI.

Non-interactively execute a task once, suitable for automation scripts.

View the current model, directory, sandbox, permissions, and tool status.

Switch or view the model currently used by Codex.

View or adjust the approval policy.

View current changes in code review mode.

View available MCP servers and tools for Codex.

View available skills and their trigger methods.

Compress long session context and continue the current task.

Commonly used when connecting OpenAI-compatible or low-cost models like DeepSeek.