The Ultimate Claude Code Command Dictionary: A Complete Guide for Power Users

Thorough Explanation of Commands, Official Skills, and Practical Workflows for the Code Tab

Scope of "Commands" in This Article

This article covers commands executed in the input field of the Code tab in Claude Desktop. Specifically, these are called by typing "/" at the start of the input field or by selecting "Slash commands" from the "+" button next to the input field.

We will cover three types:

• Commands built into Claude Code itself
• Official Skills bundled by Anthropic
• Skills added from Projects or Plugins

Conversely, we will not cover starting Claude Code from the terminal, startup flags, non-interactive execution, or operation from external scripts.

In Claude Desktop, many operations like models, permission modes, plugins, connectors, diffs, and tasks can be done via the GUI. Therefore, some commands used in the terminal version may not appear in Desktop or are replaced by GUI operations.

The most reliable way to check which commands are available in your environment is to type "/" in the Code tab input field. The contents displayed will vary depending on your OS, subscription plan, execution environment, installed plugins, and project skills.

Chapter 1: How to Invoke Commands in Claude Desktop

In the Code tab of Claude Desktop, you can invoke commands in two ways.

First, by typing "/" at the beginning of the input field.

/

A list of candidates will appear, and you can narrow them down by continuing to type.

/ver

This will show candidates containing "ver," such as "/verify."

Second, by selecting "Slash commands" from the "+" button next to the input field. The list will display the following together:

• Built-in Commands
• Anthropic's official Bundled Skills
• User-created Skills
• Project-specific Skills
• Skills from installed Plugins

When selected, the command name is inserted into the input field highlighted. You can then write your request after it and send it.

Place Commands at the Start of Input

To be recognized as a command, it must generally be placed at the beginning of the message.

Correct example:

/plan Create a plan to migrate the authentication function to the Session Cookie method

The following might be interpreted as normal natural language:

First, use /plan to investigate the authentication function

Text written after the command name is treated as arguments for that command or as the request content.

Combining Multiple Skills at Once

In Claude Code v2.1.199 and later, you can line up to 6 skills at the start of the input.

/security-context /api-rules /test-policy Implement the authentication API

This loads all specified skills and passes the request to each. However, more is not always better. Overlapping similar skills can cause issues like:

• Conflicting instructions
• Increased context consumption
• Ambiguous priorities
• Execution of unnecessary steps
• Complex completion conditions

Only combine skills directly related to the current task.

Chapter 2: Commands for Organizing Sessions and Context

In Claude Desktop, conversation history, project folders, changes, and context are independent for each session. As work progresses, you'll need to decide whether to continue the current conversation, summarize it, or split it into another session. These commands help with that.

/clear

/clear

Ends the current conversation and starts a new one with an empty context. Old conversations aren't deleted; they remain in the Desktop sidebar. You can also name it:

/clear auth-investigation

Use /clear when:

• You've finished auth work and are moving to billing
• You don't want to carry over old hypotheses or wrong results
• You're starting a new issue
• The conversation has become too long and the goal has shifted

/compact

/compact

Keeps the same conversation but summarizes past dialogue and tool results to free up context. You can specify what to keep:

/compact Keep the auth method decision, changed files, failing tests, and pending tasks

Use /compact when:

• Working on the same feature for a long time
• Context usage is high
• There are many old search results or long logs
• You want to keep important design decisions

/context

/context

Check what is consuming the current context (History, CLAUDE.md, Rules, Skills, MCP definitions, Files, etc.). For details:

/context all

Run this if Claude starts confusing past and current requests, if too many skills are loaded, or if it starts forgetting important instructions.

/branch

/branch alternative-auth

Duplicates the current conversation at that point and moves you to the duplicate. The original remains, allowing you to try an alternative without losing the current plan. This branches the conversation, not the Git branch.

/fork

/fork Investigate the safety and cost of the JWT method separately from the current plan

Starts a Background Subagent inheriting the entire current conversation. Unlike /branch, you stay in the current conversation while the agent handles a side task. Results return to the main conversation once finished.

/btw

/btw What was the cookie name used in this auth process?

Asks a "side question" using the current context without saving the content to the main conversation history. Great for quick clarifications that shouldn't derail the main task.

/recap

/recap

Briefly summarizes the current session. Useful for catching up after a long process or when juggling multiple sessions.

/rename

/rename auth-cookie-migration

Renames the current session. Useful for keeping the sidebar organized.

Chapter 3: Commands for Controlling Plans and Goals

/plan

/plan Create a plan to migrate auth from JWT to Session Cookies

Moves Claude into Plan mode to investigate and plan before editing code. It helps identify implementation details, test configurations, and risks before execution.

/goal

/goal Continue until all tests pass, type errors are 0, and the changed registration screen is manually verified

Sets a clear completion condition. Claude will continue through multiple turns until an evaluator confirms the goal is met. Goals should be objective and observable.

/loop

/loop 5m Check CI status and investigate if it fails

Repeats a request at set intervals within the current session. Useful for waiting for CI, deployments, or monitoring logs.

Chapter 4: Commands for Parallel Processing and Large-Scale Work

/batch

/batch Migrate the src/ directory from the old Logging API to the new one

An official skill that breaks large changes into independent units and implements them in parallel using Background Subagents and Git Worktrees. Best for repetitive changes across many files or packages.

/deep-research

/deep-research Investigate the implementation status of the WebAssembly Component Model as of 2026

A dynamic workflow that distributes research across many subagents and integrates the results. Much more powerful than a standard web search.

Chapter 5: Commands for Running Implemented Applications

Claude Desktop can use a Browser Pane to actually check your application. The core commands are /run, /verify, and /run-skill-generator.

/run

/run

Starts the project application and operates the changed parts. It checks if the app starts, if buttons work, and if there are console errors.

/verify

/verify

Proves that changes meet requirements through execution results. While /run focuses on operation, /verify focuses on whether the "completion conditions" are met (e.g., can you actually log in?).

/run-skill-generator

/run-skill-generator

Learns project-specific startup and verification methods to generate custom skills for /run and /verify. This saves tokens and time in future sessions.

Chapter 6: Commands for Improving Code Quality

/simplify

/simplify

Reviews changed code for simplification and reuse. It focuses on reducing unnecessary complexity and reusing existing helpers.

/code-review

/code-review high --fix

Reviews changes for bugs and quality improvements. You can specify depth (low, high, xhigh, ultra) and even have it apply fixes or post comments to GitHub.

/security-review

/security-review

Analyzes changes specifically from a security perspective (Injection, XSS, CSRF, etc.). Essential for auth or payment-related changes.

/review

/review 123

A fast, single-pass review of a GitHub Pull Request. Good for quick first checks.

Chapter 7: Project Initialization and Specialized Tasks

/init

/init

Investigates the project and generates an initial CLAUDE.md file containing project overviews, tech stacks, and coding patterns.

/claude-api

/claude-api migrate

Loads official references for Anthropic APIs, SDKs, and tool use. Use this when writing code that interacts with Anthropic's services.

/dataviz

/dataviz Design a dashboard comparing monthly sales and churn

An official skill for designing effective data visualizations, focusing on chart types, accessibility, and information priority.

Chapter 8: Commands for Diagnosing Claude Code Itself

/doctor

/doctor

Diagnoses the Claude Code installation, settings, and context configuration. It suggests and applies fixes for performance issues or context bloat.

/debug

/debug MCP Server connection drops every time

Diagnoses runtime problems with Claude Code itself (MCP connections, hooks, permissions), not your application code.

/fewer-permission-prompts

/fewer-permission-prompts

Analyzes past sessions to create permission rules for safe, read-only operations you frequently approve, reducing friction.

Chapter 9: Commands for Changing Settings Directly

While GUI is preferred in Desktop, you can use /config with key-value pairs:

/config theme=dark

Note: /config without arguments (interactive panel) does not work in the Desktop Code tab.

Chapter 10: Custom Skills and Plugin Skills

Project-specific skills (shared via repository), User skills (personal preferences), and Plugin skills also appear in the "/" menu. Skills are only loaded when called, making them more context-efficient than putting everything in CLAUDE.md.

Chapter 11: What to Use GUI for Instead of Commands

In Claude Desktop, use the GUI for:

• Model/Effort selection
• Permission modes
• Plugin/Connector management
• Diff and Task monitoring
• Session switching

Chapter 12: Practical Command Workflows

Implementing a New Feature Safely

1. /plan (Design)
2. Implement
3. /simplify (Cleanup)
4. /code-review high (Bug check)
5. /security-review (Security check)
6. /verify (Execution proof)

Fixing Complex Bugs

1. /goal (Set conditions)
2. /fork (Investigate hypotheses)
3. /code-review high
4. /verify

Summary: Commands Are an Interface for Control

Commands in Claude Desktop aren't just terminal alternatives. They are the interface for instructing Claude on when to plan, when to parallelize, and what evidence constitutes completion. Master these to turn Claude Desktop into a true development environment.