Master Claude Code with the /powerup Command: A Complete Interactive Tutorial

Verify Your Version

Run claude --version to ensure you have v2.1.90 or newer; otherwise update with npm update -g @anthropic-ai/claude-code and confirm the version again.

Open /powerup

Start Claude Code ( claude) and type /powerup. The UI shows ten animated lessons, each covering a core feature. Progress is tracked with a bar that fills after each lesson.

01 | Talk to Your Codebase

Use the @ symbol to tell Claude which file or directory to read, e.g. @src/components/Timer.tsx. This limits context usage and avoids Claude scanning the entire project. You can also specify multiple files: @./src/App.tsx @./src/styles/global.css. Permissions can block sensitive paths via .claude/settings.json with a permissions.deny array.

02 | Steer with Modes

Press Shift+Tab to cycle through four modes:

Normal – Claude asks for confirmation before every edit or command.

accept edits – Edits are applied automatically; commands still require confirmation.

Plan – Claude only proposes a plan; you approve before execution.

auto – Fully automatic; both edits and commands run without prompts.

Recommended progression: start with Normal, move to accept edits for daily work, use Plan for complex refactors, and reserve auto for long repetitive tasks.

03 | Undo Anything

Press Esc twice or run /rewind to open a rollback menu showing recent operations. Select a point to revert both file changes and conversation context. For large changes, commit with Git first ( git add . && git commit -m "snapshot") and you can always restore with git checkout ..

04 | Run in the Background

Ask Claude in plain language to run a command in the background, e.g.

please run npm run build in the background and tell me when it finishes

. Use /tasks to list running background jobs.

05 | Teach Claude Your Rules (CLAUDE.md)

Create a CLAUDE.md file at the project root (or .claude/CLAUDE.md) to provide persistent project context such as description, tech stack, coding conventions, and directory layout. Generate it automatically with /init or write it manually. Keep it concise; overly long files waste context.

Personal memory can be added via /memory or by saying “please remember …”.

06 | Extend with Tools (MCP)

Manage Model Context Protocol (MCP) servers with /mcp. Example servers include Brave Search, PostgreSQL, Playwright, and Filesystem. Add a server via the CLI: claude mcp add 12306 -- npx -y 12306-mcp, list with claude mcp list, and remove with claude mcp remove 12306 -s user. MCP lets Claude call external tools such as web search or database queries.

07 | Automate Your Workflow (Skills & Hooks)

Skills are plugin packages defined by a SKILL.md file placed in ~/.claude/skills/ (global) or .claude/skills/ (project). Install the official frontend-design skill with /plugin install frontend-design@claude-plugins-official and reload with /reload-plugins. The skill improves front‑end code generation.

Hooks run custom scripts before or after tool actions. Define them in .claude/settings.json using PreToolUse or PostToolUse. Example: automatically run Prettier after every Edit or Write operation by parsing the JSON input with jq and invoking prettier --write. Hooks can also trigger tests, security checks, or automatic commits.

08 | Multiply Yourself (Subagents)

Create isolated Claude instances with /agents. Subagents have their own context, useful for unbiased code reviews, large‑scale analysis, or batch modifications. Configure scope, description, tools, model, and color. Invoke a subagent with a command like use code‑quality‑reviewer agents to audit this project. Results are summarized back to the main session.

09 | Code from Anywhere (Remote Control & Teleport)

Run /remote-control (or /rc) in a terminal session to expose it to the web UI (claude.ai). Then you can control the same session from a phone or another computer. Conversely, use /teleport (or /tp) in a local terminal to pull a web‑based session into the terminal. The /teleport feature requires a Claude.ai subscription.

10 | Dial the Model (Model & Effort)

Switch models with /model. Options: Sonnet (balanced), Opus (most capable), Haiku (fastest, lightweight).

Adjust thinking depth with /effort. Levels: low, medium, high, xhigh, max. Recommended default is xhigh for most coding tasks; use max only after confirming a benefit. The temporary keyword ultrathink boosts effort for a single request.

High‑Frequency Maintenance Commands

/context

– Show current token usage. /compact – Summarize the conversation to free context. /clear – Erase the entire dialogue. claude --resume – Restore a previous session. claude --continue (or claude -c) – Continue the most recent session.

Conclusion

The /powerup tutorial provides a structured learning path covering everything from basic file references to advanced extensions like MCP, Skills, Hooks, Subagents, cross‑device syncing, and model tuning. New users should start with /powerup to get up to speed, while experienced users can explore hidden shortcuts such as background execution ( ! command &) and rapid rollback (Esc + Esc).