Claude Code - Power Up

/powerup is interactive lessons teaching Claude Code features with animated demo, introduced in v2.1.90

Open one, read it, try it, mark it done.

To practice high recognition, learn how Claude Code developers teach and describe their features in concise language.

Talk to your codebase

• @ files, line refs.

Type @ anywhere in your prompt to fuzzy-find and attach a file. Claude read it before answering - no more pasting code.

Reference specific lines with @src/app.ts:12. Also, @folder/

Steer with modes

Press shift+tab to cycle permission modes. auto mode - excellent!

😑 Auto mode requires Team/Enterprise or API plan.

Undo anything

Press Esc Esc (double tap) to open /rewind and roll back to any prior state.

Also: /clear wipes conversations but keeps files. /branch forks the conversation to try two approaches.

Run in the background

Long builds and test suites do not have to block you.

Add & to any bash command and it runs in the background.

Run /tasks to see everything in flight.

Teach Claude your rules

Drop a CLAUDE.md file in your repo and Claude reads it at the start of every session. Put your conversations there: test commands, style rules, do-not-touch directories.

Run /init to generate a starter CLAUDE.md from your codebase.
Run /memory to edit it inline.

Works at three levels: repo, your home directory(all projects), and per-directory overrides.

Extend with tools

MCP servers give Claude new tools: read your Slack, query your database, control your browser.

Run /mcp to browser and connect servers.

Automate your workflow

Save a prompt to .claude/skills/deploy/SKILL.md and it becomes /deploy — type it, Claude runs it. Run /skills to see what you have.

Hooks run your own scripts on events:

• before a tool call
• after a response
• on session start
  use them to enforce rules, log activity, or inject context.

Run /hooks to see what fires when.

Run /install-github-app to let Claude review PRs when tagged.

Multiply yourself

Claude can spawn copies of itself to work in parallel. Ask it to "use subagents to search these 5 directories" and watch the fan-out.

Define specialized agents in .claude/agents/ — a test runner, a code reviewer, a docs writer — each with its own tools and instructions. Run /agents to manage them.

Subagents run in isolated context. For true parallel sessions on separate branches, launch with claude --worktree.

Code from anywhere

Run /remote-control and this terminal becomes visible on your phone and at claude.ai/code. Watch output, send prompts, approve tool calls — all from another device while this session keeps running.

Dial the model

Run /model to switch models. Opus for hard problems, Sonnet for most work, Haiku for quick questions. Each trades speed for depth.

/effort controls how long Claude thinks before answering. high for tricky bugs, low when you just need a quick edit.

Also: /fast toggles fast mode — same model, faster output.