• claude
  • Claude Code Commands: The 16 That Give You 80% of the Power

    Most developers approach AI tools like chatbots.

    That’s the mistake.

    Tools like Claude Code aren’t just “ask-and-reply” interfaces—they’re programmable environments. Once you stop chatting and start operating them like systems, your productivity jumps dramatically.

    This cheat sheet image captures a huge surface area, but the reality is simple:

    👉 You only need a small set of commands to unlock most of the value.

    Let’s break down the 16 commands that matter, and more importantly—how to think about them as a system.


    🧠 The Mental Shift: From Chatbot → System

    Beginners:

    • Ask → Wait → Copy → Repeat

    Power users:

    • Plan → Execute → Monitor → Optimize

    Claude Code is designed for the second approach.


    ⚙️ The Core 16 Commands (That Actually Matter)

    🏁 1. Session & Setup

    /init

    Creates CLAUDE.md, your project memory layer.

    Think of it as:

    • Persistent context
    • Team knowledge
    • Coding conventions

    👉 If you’re not using this, you’re starting from zero every time.


    /clear

    Resets the session.

    Use it when:

    • Switching tasks
    • Context is polluted
    • Responses start degrading

    🧭 2. Thinking Before Coding (Critical)

    /plan

    Forces Claude to think before acting.

    This is one of the highest ROI commands.

    Use it for:

    • Feature design
    • Refactors
    • Architecture decisions

    👉 Senior engineers don’t code first. They model first.


    "ultra think"

    Maximizes reasoning depth.

    Use it when:

    • Problems are complex
    • You need trade-offs, not just answers

    🧠 3. Context Management (The Hidden Bottleneck)

    /context

    Shows what’s consuming tokens.

    Why it matters:

    • Context = memory = performance
    • Overflow = worse answers

    /compact

    Compresses context mid-session.

    Use it when:

    • Conversations get long
    • You want to keep momentum without resetting

    /rewind

    Undo mistakes in reasoning or execution.

    👉 Think of it as Git for your AI session.


    🔀 4. Flow Control & Exploration

    /btw

    Ask side questions without breaking flow.

    Example:

    • You’re implementing caching
    • You ask: “btw what’s the tradeoff with Redis vs in-memory?”

    Shift + Tab

    Switch modes instantly.

    This is about speed of interaction, not capability.


    🤖 5. Multi-Agent Power

    /agents

    Runs multiple agents in parallel.

    Use it for:

    • Generating alternatives
    • Comparing implementations
    • Reviewing code from different perspectives

    /simplify

    Spawns 3 agents to review your code.

    Think:

    • Code review
    • Refactoring suggestions
    • Complexity reduction

    👉 This is like having a mini team on demand.


    🌐 6. Automation & Execution

    /chrome

    Controls a browser:

    • Click
    • Fill forms
    • Navigate

    Use cases:

    • Testing flows
    • Scraping structured data
    • Automating repetitive tasks

    /loop

    Runs prompts repeatedly.

    Perfect for:

    • Monitoring systems
    • Periodic checks
    • Background workflows

    🔐 7. Permissions & Autonomy

    /permissions

    Pre-approve safe operations.

    This removes friction.


    --dangerously-skip-permissions

    Full auto mode.

    ⚠️ Use carefully:

    • No interruptions
    • No confirmations
    • Maximum speed, minimum guardrails

    👉 This is where Claude shifts from assistant → autonomous agent.


    🧩 8. Model Control

    /model

    Switch between models (Opus / Sonnet / Haiku).

    Use strategically:

    • Opus → deep thinking
    • Sonnet → balanced
    • Haiku → fast + cheap

    🔥 Putting It Together: A Real Workflow

    Here’s what a power-user flow looks like:

    1. /init → load project memory
    2. /plan → define approach
    3. "ultra think" → deepen reasoning
    4. /agents → explore multiple implementations
    5. /model → switch to faster model for execution
    6. /compact → optimize context
    7. /loop → automate follow-ups
    8. --dangerously-skip-permissions → run at full speed

    👉 This is not chatting. This is orchestration.


    🚫 Why Most Developers Struggle

    Because they do this:

    “Write me a function…”

    Instead of this:

    “Plan → Compare → Execute → Validate → Optimize”

    Claude Code isn’t underperforming.

    It’s being underused.


    💡 Final Takeaway

    You don’t need to memorize the entire cheat sheet.

    You need to master:

    • Planning (/plan)
    • Context (/context, /compact)
    • Control (/agents, /loop)
    • Autonomy (/permissions)

    Everything else is leverage.


    🚀 TL;DR

    • Most people use Claude Code like a chatbot
    • Power users run it like a system
    • These 16 commands unlock ~80% of the value
    • The real advantage is not commands—it’s workflow

    Save this. You’ll need it. ♻️

    3 mins