Claude Code is fast. It gets dramatically faster the moment you stop typing prose to it and start using slash commands instead. These ten cover 90% of what most builders need.
The starter five
/init — Generate a CLAUDE.md for your project. Run this first in any new repo. Claude reads the codebase, infers conventions (build command, test command, file layout, naming), and writes a tailored CLAUDE.md. That file becomes context for every future session — it's the difference between Claude guessing and Claude knowing.
/clear — Reset the conversation. Use it between unrelated tasks. Lingering context makes Claude reference irrelevant files and hallucinate "fixes" to code it half-remembers. A fresh /clear is cheaper than a half-poisoned context.
/compact — Summarize the current conversation in place. Run this when a long session is approaching context limits but you don't want to lose the thread. Claude will compress what came before and continue from a smaller surface.
/cost — Show how many tokens (and dollars) the session has used. Useful when you're testing a new workflow and want to know what it would cost to run daily.
/help — Lists everything available in your install, including any project-specific commands defined in .claude/commands/.
The shipping commands
/plan — Enter plan mode. Claude explores the codebase, asks clarifying questions, and produces an implementation plan you approve before any code is written. Critical for non-trivial work — a 30-second plan review beats 30 minutes of redoing the wrong thing.
/review — Review the current changes (or a PR). Claude reads the diff, checks for bugs, design issues, and consistency. Pair with /security-review for code that touches auth, payments, or user data.
/security-review — Same shape as /review, but focused on OWASP-class issues: injection, missing auth checks, leaked secrets, unsafe deserialization. Run it before any merge that touches a sensitive surface.
The leverage commands
/agents — List or invoke specialized subagents. Subagents have their own tools and instructions. Examples: a code-reviewer that gives independent second opinions, a general-purpose agent for parallel research, an Explore agent for fast read-only searches. Delegate to them when you want fresh eyes or to keep the main context lean.
/skills — Skills are reusable instruction packs (think: a "security-review" skill, a "deploy" skill, an "init" skill). Run /skills to see what's installed. Add more with npx skills add <repo> — the softaworks/agent-toolkit marketplace has 50+ free skills covering documentation, design, planning, and testing.
/hooks — Configure shell commands that run on Claude events: before a tool call, after an edit, on session end. Use cases: auto-format on save, run tests on every edit, push notifications when long jobs finish, block dangerous commands in sensitive directories. Hooks turn Claude Code from a tool into a teammate that follows your team's rules.
The order to learn them
If you're new to Claude Code, run them in this order on your next real project:
/init— give Claude context about your codebase/plan— let Claude design before it codes/review+/security-review— gate every meaningful change/clear— between tasks/agentsand/skills— add leverage as your workflows mature/hooks— automate the parts you find yourself typing twice
The first three are non-negotiable. The rest are accelerators.
Where to go next
- Read Setting up CLAUDE.md so Claude understands your project — it's the highest-ROI follow-up.
- Browse the agent-toolkit skill marketplace for ready-made workflows.