Installing Superpowers in 5 Minutes
Installation is five minutes if you have Node.js. This lesson walks you through the exact commands and explains what each step actually does under the hood — because understanding the mechanism is what lets you debug it when something goes wrong.
Prerequisites
You need three things before starting:
Node.js 18 or later. Claude Code is a Node.js CLI application. Verify with node --version. If you need to install it, use the official installer at nodejs.org or brew install node on macOS.
An Anthropic account with Claude Code access. You can sign up at claude.ai. Claude Code is included in Pro and above, or available via API key on pay-as-you-go.
A terminal. Any terminal works: macOS Terminal, iTerm2, Windows Terminal, or the integrated terminal in VS Code or JetBrains. All commands in this course use Bash/Zsh syntax.
Step 1: Install Claude Code
This installs the claude CLI globally. Verify:
You should see a version number like 1.x.x. If the command is not found, your Node.js bin directory may not be on your PATH. Run npm bin -g to find the global bin path and add it to your shell profile.
Step 2: Authenticate
This opens a browser window where you authenticate with your Anthropic account. After authenticating, your credentials are stored locally at ~/.claude/credentials.json. The CLI never stores your password — it stores an OAuth token.
If you prefer API key authentication (for CI environments or shared machines):
Add this to your shell profile (~/.zshrc, ~/.bashrc) to make it permanent.
Step 3: Install the Superpowers Plugin
This downloads the official Superpowers plugin from the Claude plugin registry. Under the hood, it:
- Downloads the plugin package to
~/.claude/plugins/cache/claude-plugins-official/superpowers/<version>/ - Creates an entry in
~/.claude/plugins/installed.json - Makes all skill SKILL.md files discoverable by the
Skilltool
Verify the installation:
You should see a versioned directory (e.g., 5.0.7). Inside it:
Each directory is one skill. The SKILL.md file is the skill itself. The patterns/ subdirectories contain code examples that load on demand (covered in detail in lesson 6).
Step 4: Open a Project
Navigate to your project directory and start Claude Code:
Claude Code starts an interactive session. The first thing it does is load your CLAUDE.md file (if it exists) and the auto-memory index from ~/.claude/projects/<project-hash>/memory/MEMORY.md. You will see a > prompt when it is ready.
Your First Skill Invocation
Type the following at the prompt:
This invokes the ORACLE skill. You will see Claude announce:
Running task-intake to classify and plan approach.
Then it will run through the 5-phase classification protocol — but since you did not give it a task, it will ask you what you are working on. Type a real task description and watch it classify complexity, select a skill chain, and assign a model tier.
This is the most important skill in the system. Every subsequent lesson builds on understanding what happened here.
How the Skill System Works Mechanically
Demystifying the loading mechanism prevents a lot of confusion later.
Discovery. When Claude Code starts, it scans the plugin directories for all SKILL.md files. It reads only the frontmatter of each file — the name, description, and type fields between the --- delimiters. This creates a lightweight skill registry. The full skill content is not loaded at startup.
Invocation. When you type /skill-name or when Claude decides a skill applies to the current task, the Skill tool loads the full SKILL.md content for that skill. This is when the ~700-1000 token content enters the context window. Skills are lazy-loaded — only the ones relevant to the current task incur token cost.
Pattern files. Some skills (ml-engineering, ai-engineering, embedded-systems) have patterns/ subdirectories with code examples. These are explicitly referenced inside the SKILL.md but not automatically loaded. The skill text says "Load patterns: patterns/data-pipeline.md" — Claude then decides whether to load that pattern file based on what the task requires. This keeps the base skill lean while making the detailed examples available when needed.
The ASCEND skill. This is special: it is loaded automatically at the start of every session by a session hook. It tells Claude the rules for skill discovery and invocation — the meta-skill that governs all other skills. You never invoke it manually.
Anatomy of a SKILL.md File
Every skill follows the same structure:
The description field is critical — it is what Claude reads when deciding whether this skill applies to the current task. A good description is specific and includes the exact scenarios where the skill applies.
The type field indicates skill category:
process: Workflow skills (task-intake, debugging, TDD, verification)domain: Domain expertise skills (ml-engineering, ai-engineering, embedded-systems)implementation: Execution skills (brainstorming, writing-plans, subagent-driven-development)
CLAUDE.md vs Skills
You may already have a CLAUDE.md file in your project. This is different from skills.
CLAUDE.md is project-specific instruction. It describes your stack, conventions, API patterns, and anything else Claude should know about your specific project. It is always loaded at session start. Think of it as "the README Claude actually reads."
Skills are universal protocols. They describe how to approach categories of work — ML pipelines, debugging sessions, multi-agent tasks — regardless of which project you are in. Skills are loaded on demand by relevance.
They complement each other. CLAUDE.md provides project context. Skills provide process and domain expertise. Both are in the context window at the same time for any given session.
The Auto-Memory System
One more thing installs alongside the plugin: the memory directory. At ~/.claude/projects/<project-hash>/memory/, Claude Code will maintain persistent memory files across sessions. You will see:
MEMORY.md— the index file, loaded at every session start- Individual
.mdfiles for each memory entry (user preferences, project context, feedback, references)
This directory is created automatically the first time Claude writes a memory. You do not need to create it. Lesson 9 covers the memory system in full.
Verification Checklist
Before moving to the next lesson, verify:
- [ ]
claude --versionreturns a version number - [ ]
claude logincompleted without errors - [ ]
ls ~/.claude/plugins/cache/claude-plugins-official/superpowers/shows a version directory - [ ] Opening a project with
claudeand typing/task-intaketriggers the skill announcement
If /task-intake shows an error instead of the skill content, check that the plugin installed correctly:
You should see superpowers in the installed plugins. If not, re-run claude plugin install superpowers.
Key Takeaway
The Superpowers system is a directory of SKILL.md files that load on demand into Claude's context window. Skills are lazy-loaded — only invoked skills incur token cost. The ASCEND skill loads at session start and governs the rest. CLAUDE.md provides project context; skills provide process and domain expertise. Together, they give Claude the structure it needs to apply the right approach to every task.