Slash commands are how you steer a Claude Code session without leaving the keyboard. Some are built in, managing context, cost, models, and connected tools, and some you write yourself. Knowing the right handful turns Claude Code from something you type paragraphs at into something you actually drive.
This is the shortlist: the built-in commands worth committing to muscle memory, and the custom ones most people should build. Command sets evolve between releases, so run /help in your version to see exactly what’s available, but the categories below are stable.
Built-in commands worth memorizing
These ship with Claude Code and handle the mechanics of a session.
Daily-driver built-in commands
| /context | Show how the context window is allocated |
|---|---|
| /clear | Reset the conversation and free the window |
| /compact | Summarize the conversation to reclaim space |
| /model | Switch models without restarting |
| /mcp | List connected MCP servers and their tools |
| /agents | Manage and inspect subagents |
| /hooks | Add and review hooks |
| /help | List every available command |
/context and /clear, the two you’ll lean on most
/context shows where your window is going, system, tools, conversation. When responses start feeling vague or the session slows, check it first. Often the culprit is a bloated conversation or too many connected MCP servers, which this guide on MCP context cost gets into.
/clear is the reset button. When a session has wandered and accumulated junk, clearing it and re-stating just the relevant context is faster than fighting a cluttered window. Make it a reflex between unrelated tasks.
/compact, continuity without the bloat
When you want to keep going but the window is filling, /compact summarizes the conversation so far and frees space while preserving the gist. The difference from /clear: compact keeps continuity, clear throws it away. Use compact mid-task, clear between tasks.
/model, switch tiers on the fly
/model changes the model without restarting the session. Drop to a cheaper, faster model for routine edits and bump back up for a hard problem. If you route through a proxy, the available names reflect your config, see the Claude Code Router setup and the cheap-models guide for that.
Commands for managing your setup
These don’t do work; they show you the state of your tools.
/mcp lists connected servers and the tools each exposes, your first stop when a server isn’t behaving or you’re auditing context cost. /agents opens subagent management, scaffolding new ones and showing which are available. /hooks walks you through adding hooks without hand-editing JSON. Together they’re how you inspect and tune the extension points covered in the power-user guide.
Custom commands you should build
Built-ins handle the session; custom commands handle your repeated prompts. The mechanics, a markdown file in .claude/commands/, are in creating custom slash commands. Here’s what’s worth building:
/review, your code-review prompt applied to current changes. The single most useful custom command for most teams./commit, generate a commit message in your house format./test, write tests for a given file following your conventions./explain, explain a file or module, taking the path as an argument, for onboarding or unfamiliar code./pr-desc, draft a pull request description from the branch’s diff./security, a focused security pass over the current changes.
Review the current changes for bugs, missing error handling, and
security issues. Point to a file and line for each finding and rank
them: blocker, should-fix, nitpick. If the change is clean, say so.A practical starter set
You don’t need every command on day one. Learn these built-ins and build these custom ones:
Slash command starter set
- Built-in: /context, /clear, /compact for managing the window
- Built-in: /model to switch tiers, /mcp and /agents to inspect tools
- Custom: /review for code review on demand
- Custom: /commit for your commit message format
- Custom: /explain with an argument for unfamiliar code
- Run /help to see everything your version offers
Verify the result
The built-in commands you’ll actually use boil down to a handful: /context and /clear to manage the window, /compact for continuity, /model to switch tiers, and /mcp and /agents to inspect your tools. Layer on a few custom commands for the prompts you repeat, /review and /commit first, and you’re driving Claude Code instead of typing at it.
To build the custom ones, follow creating custom slash commands; to see where commands fit among skills, hooks, and subagents, the power-user guide ties it together.
