Claude Code ships with 93 commands. Most developers never use more than a handful. The barrier isn't documentation — it's knowing where to start. This guide cuts straight to the 16 commands that deliver the highest immediate return, ranked by practical impact rather than alphabetical order.
Setup and Context Commands: Build the Right Foundation
Before Claude writes a single line of code for you, it needs to understand your project. This is where the gap between mediocre and excellent AI-assisted development opens up.
- /init — The first command you should run in any new project. Claude reads your existing files and auto-generates a
CLAUDE.mdfile summarizing your architecture, patterns, and constraints. When Claude understands your stack from the start, output quality improves noticeably from the very first prompt. - /context — Shows a breakdown of what's consuming your token window. A surprising finding: most users have already spent 15–20% of available context before writing their first real prompt. Checking this before long sessions is non-negotiable.
- /compact — Compresses conversation history to free up context without losing the thread. The golden rule: use it at 50% consumption, not at 95% when the model is already starting to degrade in quality.
- /clear — A full session reset. Essential when switching between unrelated tasks. Leftover context from a previous session can corrupt new results in ways that aren't always obvious until something breaks.
Flow Control Commands: Take Back Precision
One of the most common frustrations with AI agents is that they "do too much at once." These commands put you back in control.
- /plan — Switches Claude into read-only mode. It maps every step before touching a single file. Five minutes of upfront planning prevents an hour of painful refactoring. Combined with a well-configured
CLAUDE.md, this is arguably the highest-ROI command on the entire list. - /rewind — Undoes the last action when Claude has drifted off course. Quick alternative: pressing Esc twice opens the rollback menu without typing anything.
- Shift+Tab — Not a slash command, but equally critical. Cycles through three modes: normal, auto-accept, and plan. Mastering this shortcut changes the entire rhythm of a working session.
- /permissions — Pre-authorizes specific safe commands rather than skipping permission checks entirely. The aggressive alternative is --dangerously-skip-permissions, which removes all permission prompts and lets Claude execute without interruption. Powerful for automated pipelines and controlled dev environments, but use it deliberately.
Power Commands: Scale and Depth
Once you have the basic flow locked in, these commands unlock capabilities that go well beyond a simple coding assistant.
- /model — Switches between Claude Opus, Sonnet, and Haiku mid-session. The logic is straightforward: Opus for tasks requiring deep reasoning, Haiku for speed on repetitive or mechanical tasks. You don't need to use the same model for every step of a workflow.
- /agents — Creates and manages subagents that work in parallel. This is the most powerful command on the list for complex projects. Splitting code generation and code auditing into independent subagents, for instance, both reduces wall-clock time and improves final output quality.
- /chrome — Gives Claude direct control of your browser. It can navigate, click, fill forms, and read console logs. For QA work and end-to-end testing, this fundamentally changes what's possible without leaving your terminal.
- /loop — Runs a prompt on a recurring schedule for extended periods — up to three consecutive days. Built for supervising long-running builds or monitoring continuous integration processes.
- /simplify — Spins up three parallel agents to review your code from different angles simultaneously. It surfaces architectural problems, duplicated logic, and anti-patterns that a single-pass review would miss.
Utility Commands: Small Footprint, Real Value
- /btw — Lets you ask a secondary question without interrupting the main task in progress. Small on the surface, but it eliminates the friction of opening a new session just to resolve a quick doubt.
- "Ultra think" — Not a slash command but a natural language instruction that forces maximum reasoning depth in a single turn. Save it for complex architectural decisions where you need the best possible answer, not the fastest one.
Where to Start Tomorrow
If you had to pick just three commands to focus on this week, choose /init for every new project, /plan before any non-trivial task, and /context to understand how you're spending your token budget. The rest will follow naturally once those three are automatic.
Claude Code has 93 commands because the use cases are genuinely diverse. You don't need to know all of them. You need to know the right ones for your work. This list is that starting point.
