Every Claude Code Slash Command, Explained: The Cheat Sheet Worth Saving

1. Session and Context Management

This is the category you'll use most often. These commands deal with conversation state, context size, and checkpoints.

/clear — Wipe the conversation clean

Erases the entire current conversation context. Use it when Claude starts confusing files from a previous session, or when you're moving on to a completely different task.

The difference from the similar-looking /compact matters. /clear erases all memory and starts from scratch, while /compact summarizes that memory to shrink it. New task: /clear. Same task running long: /compact.

/compact [instructions] — Save context by summarizing the conversation

Summarizes the conversation to reduce context usage. Use it when you're continuing the same task but the session has grown long.

Append instructions to control what gets kept.

  /compact keep only the API decisions
  /compact keep only the error-handling patterns
  /compact preserve only where the code changed and why

The general rule is to run /compact once context usage passes 80%. As of v2.0.64, compaction completes instantly.

/context — Check context usage

Shows how your context window is being used. Run it before /compact to see what's actually taking up the space.

Recent versions also display per-model token estimates for each skill, plus the plugin name for plugin-sourced skills. When you find yourself wondering \"why is my context filling up so fast?\", this is the command that answers it.

/rewind — Roll back to a checkpoint

Reverts to a previous checkpoint or state. Useful when the conversation is heading in the wrong direction. Its alias is /undo.

It does the same thing as the Esc + Esc keyboard shortcut, but you can also invoke it explicitly as a command.

/branch [name] — Fork the conversation

Forks the current conversation or workflow. The alias /fork shows up in some references.

If you've been heading in direction A and want to try direction B too, creating a branch lets you experiment without losing the original.

/goal — Set a completion condition, then run autonomously

Set a completion condition and Claude keeps working across multiple turns until it's met. Works in interactive mode, -p mode, and remote control.

Use it when a task has a clear definition of done and you want Claude to get there on its own. Elapsed time, turn count, and token usage appear in an overlay panel.

/resume [session name] — Resume a previous session

Reopens a previous session. Supports named sessions and session titles set with /rename or --name. Press Ctrl+A to show sessions from all projects.

/rename [name] — Rename a session

Renames the current session. When you're juggling several tasks at once, give each one a meaningful name so it's easy to find later.

/status — Check session status

Shows the current session and environment status. A good first command to run when you sit down to work.

/diff — View changes

Opens an interactive diff viewer. Use it to see what changed before a code review or a commit.

/copy [N] — Copy a response

Copies the most recent response — or a selected one — to the clipboard. Handy for moving code or generated text somewhere else.

/export [filename] — Export the conversation

Exports the conversation to a file, usually in Markdown. Useful for keeping a record of your work or sharing it with teammates.

/cost — Check costs

Opens cost information. Since v2.1.118 it has been folded into /usage, but it still works as a shortcut — it jumps straight to the cost tab inside /usage.

2. Settings and Environment

Commands that configure how Claude Code itself behaves.

/init — Generate CLAUDE.md

Generates CLAUDE.md, the project instructions file. One of the most important setup commands there is.

It's the first command to run when you start working on a new project with Claude Code. Claude scans the codebase and drafts a CLAUDE.md covering build commands, how to run tests, code conventions, and more. You can then edit it by hand.

/config — Open the settings panel

Opens Claude Code's settings. Editor modes (Vim and others) are managed in here too.

/memory — Edit memory files

Opens or edits Claude's memory files, including the CLAUDE.md workflow.

Messages that start with # get appended to CLAUDE.md, and /memory lets you inspect and edit that content directly.

/permissions — Manage permissions

Manages permission modes and recently denied commands. When Claude keeps asking for permission, or commands keep getting blocked, this is where you adjust the policy.

/model [model] — Switch models

Switches the active model. Balance speed against depth depending on the nature of the task.

 /model opus    # hard reasoning, architecture
  /model sonnet  # everyday coding
  /model haiku   # simple exploration

Gateway /v1/models discovery is enabled with the environment variable CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY=1.

/effort [level] — Adjust the reasoning level

Adjusts reasoning effort. The options are low, medium, high, xhigh, max, and auto.

xhigh only works on Opus 4.7 and above. Run /effort with no argument and an interactive slider appears.

Since March 2026, the default effort level for Max/Team subscribers on Opus 4.6 has changed from high to medium.

/fast [on|off] — Fast mode

Toggles fast mode on and off. It's not a different model — it runs the same Opus 4.6 with an API configuration optimized for speed.

Ideal for fast interactive iteration and live debugging. Turn it off when cost matters. One caveat: if you switch fast mode on mid-session, your entire prior context gets re-billed at fast-mode rates.

/output-style [style] — Adjust the response format

Changes the response formatting style. Switches between concise and detailed output.

/theme — Change the theme

Changes or creates themes. Supports \"auto-match terminal,\" named custom themes, JSON themes in ~/.claude/themes/, and plugin-provided themes.

/color [color] — Change the accent color

Changes the session's accent color. If remote control is connected, the accent color syncs to claude.ai/code as well.

/scroll-speed — Adjust scroll speed

Adjusts mouse-wheel scroll speed, with a live preview. Useful when scrolling feels off in VS Code, Cursor, or your terminal.

/tui — Switch terminal UI modes

Switches the terminal UI mode. /tui fullscreen renders the same conversation without flicker.

/terminal-setup — Terminal setup guide

Configures terminal integration. It walks you through environment tweaks, like mapping the Option key to Meta on a Mac.

/keybindings — Configure shortcuts

Creates or edits ~/.claude/keybindings.json. Changes apply immediately — no restart required.

/hooks — Manage hook scripts

Manages hook scripts and hook events. In recent versions, hooks can call MCP tools with type: \"mcp_tool\", use the exec-style args: string[], and set continueOnBlock on PostToolUse.

/login, /logout — Authentication

Log in and log out. Use these to clean up active authentication sessions.

/doctor — Diagnostics

Runs diagnostics. Useful when something is off after an install or upgrade. Press f and Claude fixes the reported issues itself.

/help — Help

Displays help and the available commands. Commands change often, so this is the safest one to run when you're not sure.

/exit — Quit

Exits Claude Code.

3. Skills, Agents, Plugins, and MCP

Commands that extend Claude Code.

/agents — Manage subagents

Manages subagents. Running claude agents opens the new agents view, which lists running, blocked, and completed Claude Code sessions in one place.

/skills — List skills

Lists the available skills. Recent versions support filter-as-you-type search and show clearer context for plugin-sourced skills.

Skills are user-defined commands; drop them into the .claude/skills/ directory and they work as project-specific slash commands.

/plugin — Manage plugins

Opens plugin management. Plugin details show a component inventory, hook event names, MCP server names, and the estimated per-session token cost.

/reload-plugins — Hot-reload plugins

Hot-reloads installed plugins. Useful when testing a plugin. Recent versions import plugin-provided skills more reliably, and v2.1.110 added remote-control support.

/mcp — Manage MCP servers

Manages MCP (Model Context Protocol) servers. It also unlocks dynamic MCP slash commands, and reconnecting picks up .mcp.json edits without a restart.

The MCP slash command pattern

Commands are generated dynamically from connected MCP servers.

/mcp__[server]__[prompt] [args]

These commands only appear once the corresponding MCP server is connected.

/claude-api — Claude API/SDK workflows

Loads the Claude API or SDK helper workflow. Useful for API-related work. Recent versions cover Managed Agents as well.

/less-permission-prompts — Permission automation

Scans your transcripts for frequently used read-only Bash and MCP tool calls, then proposes a prioritized allowlist to add to .claude/settings.json. Added in v2.1.111.

When the constant permission prompts get annoying, run this once and allowlist all your frequently used, safe commands in one pass.

4. Coding, Review, and Planning

The commands you'll use most in real development work.

/plan [description] — Enter plan mode

Switches into plan mode: Claude only reads, never edits. Use it when you want to talk through the approach before a big change.

The Shift + Tab keyboard shortcut toggles the same mode.

/ultraplan — Precision planning

Runs a more detailed planning workflow. In recent versions, the remote-session flow auto-provisions a default cloud environment.

/review — Code review

Reviews your current code changes. One of the most heavily used coding commands.

/simplify — Multi-agent refactoring

A workflow where multiple agents refactor in parallel. Great for structural improvements.

If /review is a single review, /simplify is a parallel review from multiple perspectives. Use it to examine code quality from several angles at once.

/ultrareview [PR number] — Comprehensive cloud review

Runs a comprehensive cloud code review with parallel multi-agent analysis and critique. With no argument, it reviews the current branch; pass a PR number to review that PR instead.

/security-review — Security review

Runs a security-focused review. Useful before a merge or a release.

/debug [description] — Guided debugging

Runs a guided debugging workflow. Great for tracking a bug down step by step.

/batch — Bulk changes

Applies the same change across multiple files or worktrees. Useful for repetitive edits that span an entire project.

/loop [interval] — Recurring workflows

Runs recurring or scheduled workflows. Useful for periodic checks and automation. Its alias is /proactive.

/btw <question> — Side questions

Asks a side question without dragging in your full working context. Meant for quick checks.

You can ask without interrupting the work in progress. The response isn't added to the conversation history and doesn't change your context window.

/usage — Check usage

Shows plan limits, quota usage, costs, and stats. Since v2.1.118, /cost and /stats have been consolidated as tabs inside /usage.

/insights — Usage insights

Shows usage and session insights. Useful for reflecting on your own work patterns.

/focus — Toggle focus view

Toggles focus view on and off. It's no longer tied to Ctrl+O.

/powerup — Interactive feature lessons

Opens interactive feature lessons. Added in 2026, it helps you get up to speed quickly on new Claude Code features.

/release-notes — Release notes

Opens the release notes. Claude Code updates weekly, so this is the fastest way to keep up with recent changes.

/feedback — Send feedback

Sends feedback. Use it to file bug reports or feature requests directly.

5. Git, PRs, and Releases

Commands for branches, commits, pull requests, and release workflows.

/commit — Commit

Generates a commit message and commits. One of the most popular commands people build for themselves.

/pr — Create a PR

Creates a pull request. A common custom/community command.

/push — Push

Pushes the current branch.

/pr-comments [PR] — Fetch PR comments

Fetches GitHub PR comments. The internal underscore form is /pr_comments.

/install-github-app — Set up the GitHub app

Sets up the GitHub app integration.

/fix-pipeline — Repair the CI pipeline

Attempts to repair a failing CI pipeline. A common custom/community command.

/lint — Run linting

Runs lint commands. Use it for code style checks.

/merge-to-main — Merge to main

Handles the merge-to-main workflow.

/vitest — Vitest tests

Runs a Vitest-based testing workflow.

6. Integrations, Remote, and Devices

Commands for working across IDEs, browsers, desktop, mobile, and remote environments.

/ide — IDE integration

Opens or manages IDE integration.

/chrome — Chrome integration

Opens the Chrome integration. Availability varies by environment.

/desktop — Desktop app

Hands work off to the desktop app.

/rc or /remote-control — Remote control

Starts remote control. /rc is the short form, /remote-control the full name. Remote-session features keep expanding in recent versions.

/remote-env — Configure remote environments

Configures remote environments. The latest remote-session planning features can auto-provision a default cloud environment.

/schedule — Cloud schedules

Manages scheduled cloud tasks.

/teleport — Transfer a session

Transfers or bridges a session.

/voice — Voice mode

Enables push-to-talk voice mode.

/add-dir <path> — Add a directory

Adds another directory to the working scope. Useful for monorepos and adjacent projects.

7. Diagnostics and Internal Commands

Mostly internal, debug-only, or experimental commands. Regular users won't reach for them often, but they're worth knowing about.

/team-onboarding — Team onboarding guide

Generates a team ramp-up guide from your local Claude Code usage history. Added in v2.1.101. When a new teammate joins, it automatically produces a \"here's how our team uses Claude Code\" document.

/sandbox — Sandbox controls

Opens sandbox controls.

/stickers — Stickers

Public, but decidedly non-essential.

Internal and leaked commands

The following commands show up in source references but may not exist in standard public builds. Just know they're out there.

• /summary — Session summary
• /session — Session management UI
• /share — Share a session
• /env — Environment settings
• /files — List the files in context
• /issue — Create a GitHub issue
• /passes — Multi-pass workflows
• /bughunter — Bug-hunting workflow
• /advisor — Architecture advice
• /tasks — Manage background tasks
• /onboarding — First-run onboarding

8. At a Glance, by Category

The 10 you'll use most
/init # first command on a new project /clear # switching to a new task

/compact # save context in long sessions /context # check context usage /model # switch models to fit the task /plan # plan mode before big changes /review # code review /btw # side questions /usage # check cost and usage /help # the safest command when unsure

<Recommended combos by workflow>

Starting a new project
/init → generate CLAUDE.md /permissions → review the permission policy /add-dir → add the directories you need

Starting a big task
/plan → plan mode /ultraplan → more precise planning /effort xhigh → maximum reasoning level

Managing a long session
/context → check current usage /compact → compress past 80% /clear → switch to a new task

Code review
/diff → check the changes /review → baseline review /simplify → multi-agent review /security-review → security review /ultrareview → comprehensive cloud review

Git work
/commit → commit /push → push /pr → create a PR /pr-comments → check PR comments

9. A Realistic Order for Learning the Commands

Memorizing 50-plus commands at once isn't realistic. This sequence works far better.

Week 1: the 5 survival commands.
/help, /clear, /compact, /init, /exit. These alone cover basic work.

Week 2: 5 everyday commands.
/model, /plan, /review, /diff, /context. Your daily work gets noticeably more efficient.

Week 3: 5 for integration and automation.
/mcp, /agents, /skills, /hooks, /permissions. This is where Claude Code becomes part of your working environment.

Week 4: 5 advanced commands.
/ultraplan, /ultrareview, /security-review, /goal, /btw. These handle big tasks and complex reviews.

Learn 20 commands in a month this way, and after that just look things up with /help as needed. The remaining 30 are worth knowing about and searching for on demand.

10. Building Your Own Commands

Beyond the built-ins, you can create commands of your own.
Just add a Markdown file to the .claude/commands/ directory.

Personal commands

Put them in ~/.claude/commands/ and they work in every project.

Project-shared commands

Put them in .claude/commands/ and they're shared with your team through git.

Example: building a /ship command

Save the following to ~/.claude/commands/ship.md.

Review the current diff and run the tests.
  If they pass, commit with a clear message and push to main.

Now type /ship and that workflow runs exactly as written. Turning repetitive tasks into commands lets you customize Claude Code around your own work patterns.

Arguments work too. Use $ARGUMENTS in a command definition and whatever you pass at invocation drops into place.

Handle issue #$1 with priority $2.

Call it as /fix-issue 123 high and it becomes “Handle issue #123 with priority high.”