Slash commands are the fastest way to drive Claude Code from the terminal. Type / at any prompt and you get a menu of built-ins plus any custom skills you’ve installed. Most of the day-to-day stuff happens through about eight of them.
This page is a working reference: what each command does, when to reach for it, and notes from real use. Updated for the current Claude Code release.
The eight you’ll actually use
/init
Generates a CLAUDE.md at the root of the current repo, summarizing the codebase so future Claude Code sessions have project context. Run it once when you start using Claude Code on a new repo. Re-run it after big architectural changes.
/review
Performs a code review of the pending changes on your current branch. Reads the diff, checks for common bugs, suggests improvements. Best used right before you push a PR — it catches the things you stopped seeing two hours ago.
/security-review
Like /review but specifically focused on security issues — auth bugs, injection risks, secret leaks, dependency-level concerns. Worth running on any branch that touches authentication or user input handling.
/clear
Wipes conversation history. Useful when context has gotten muddy and the model is conflating an earlier task with the current one. Cheaper and faster than starting a new session.
/help
Shows the current command list. Always up to date — your starting point if you forget something.
/add-dir
Adds an additional directory to the working set. Use when your task crosses repo boundaries — e.g., a backend repo and a shared types repo.
/cost
Shows current session token + cost. Honest reality check on long agent runs.
/agents
Manages subagents — specialized agents with their own scoped tools. The advanced workflow once you’ve outgrown single-shot prompting.
Less common, still useful
/permissions
Edit which tools Claude Code is allowed to use without asking. Useful when you’re doing a lot of file edits and tired of confirming each one.
/model
Switch the active model mid-session. Drop down to a smaller model for grunt work, jump up to a bigger one for tricky refactors.
/compact
Compresses long conversation history into a summary. Less aggressive than /clear — preserves the gist while freeing context.
/mcp
Inspect connected MCP servers, list their tools, debug connections. Where you go when an MCP integration is misbehaving.
/hooks
Manage Claude Code hooks — scripts that run on specific events (file save, tool call, session end). Powerful, niche.
/config
Open the Claude Code settings UI in your terminal — keybindings, default model, theme, telemetry.
/pr
Drafts a pull request body from the current branch’s commits + diff. Pairs well with /review right before pushing.
Custom commands (skills)
Anything in your ~/.claude/skills/ or installed plugins shows up under /. The /init, /review, and /security-review built-ins are themselves skills — you can write your own with the same shape.
For how the skill system actually works, see Anatomy of a Skill. For description-writing patterns that make custom slash commands trigger reliably, see Writing skill descriptions.
Workflow combos
A few slash-command sequences worth memorizing:
Pre-PR cleanup. /review → /security-review → /pr. Catches bugs, catches security, drafts the PR body — all before you open the browser.
Onboarding to a new repo. /add-dir for adjacent repos → /init → ask Claude to summarize the architecture. CLAUDE.md persists for future sessions.
Long agent run going sideways. /cost to confirm you’re burning budget → /compact to reset context → re-state the goal in one sentence.
Related
For the bigger picture of building agentic workflows in Claude Code, see Anatomy of a Skill and Build your first MCP server.