Skip to content

Getting Started with Superpowers

Install once. Change how your AI works forever.


What is Superpowers? (Technical)

Superpowers is an agentic skills framework — a structured set of behavioral protocols that you load into an AI coding agent. It was created by Jesse Vincent, best known as the author of Request Tracker, one of the most widely deployed issue tracking systems in history.

The framework defines how an AI agent should think and behave during software development — not just what code to write, but when to stop and ask, how to plan, how to verify, and how to know when something is actually finished.

Technically, Superpowers is a collection of skills (also called plugins or rules, depending on your platform). Each skill is a focused set of instructions that activates under specific conditions. When you install Superpowers into your AI tool, you're giving it a set of professional software engineering disciplines built as machine-readable protocols.


Core Philosophy

Superpowers is built on three principles that run through every skill:

1. TDD First (Test-Driven Development)

No production code is written without a failing test first. This is not a guideline — it is an enforced rule. The AI must write a test that fails before it writes any implementation. If you find yourself tempted to skip this step, Superpowers explicitly requires you to delete any code written without a test and start over.

Why this matters: Tests written after code tend to test what the code does, not what it should do. Tests written first capture intent and catch assumptions early.

2. YAGNI (You Aren't Gonna Need It)

Build only what is explicitly required. No speculative abstractions, no "future-proofing," no extra configuration options "just in case." Every addition to a codebase has a maintenance cost. Superpowers keeps the AI focused on the task at hand.

3. Evidence Over Claims

Seeing a test pass is evidence. Saying it should work is a claim. Superpowers requires the AI to produce actual output — real test results, real command outputs, real file diffs — before asserting that anything is complete. "It should work" is not an acceptable completion statement.


Installation

Superpowers works across multiple platforms. Choose your tool:

Claude Code

/plugin install superpowers@claude-plugins-official

This installs the full Superpowers skill suite into your Claude Code session. Skills activate automatically based on context.

Cursor

/add-plugin superpowers

After adding the plugin, Cursor will load the skill definitions. You may need to restart your Cursor session.

Gemini CLI

gemini extensions install superpowers

Follow the prompts to authorize the extension. All skills become available in your next Gemini CLI session.

Codex

Add Superpowers as a custom instruction set via the Codex configuration file. Refer to the official Codex documentation for plugin installation paths.

OpenCode

opencode plugin add superpowers

Skills load automatically on session start.


Verifying Your Installation

After installation, ask your AI agent:

"What Superpowers skills are available?"

A correctly installed Superpowers suite will list the active skills. You should see skills including brainstorming, writing-plans, executing-plans, test-driven-development, and others.

If the agent doesn't recognize the skills, re-run the installation command and restart your session.


The 7-Step Core Workflow

Once installed, Superpowers guides development through a structured 7-step process. Here's a brief overview — each step has its own dedicated chapter.

StepSkillDescription
1BRAINSTORMExplore problem space
2ISOLATECreate safe work environment
3PLANWrite detailed task list
4EXECUTEImplement with subagents
5TESTTDD cycle: Red → Green → Done
6REVIEWTwo-stage quality gate
7COMPLETEVerified delivery

Step 1: Brainstorming

The AI explores the problem with you before touching any code. It asks clarifying questions one at a time, presents 2–3 approaches with trade-offs, and waits for your approval. A hard gate prevents code from being written until a design is approved.

→ Covered in detail in Brainstorming & Design

Step 2: Isolation

Work happens in a git worktree — an isolated copy of your repository. Changes stay contained until they're deliberately merged. Your main branch is always safe.

→ Covered as part of the execution workflow

Step 3: Writing Plans

Before execution begins, a detailed implementation plan is produced. Each task in the plan is self-contained, with complete code snippets, exact file paths, and exact commands. The plan is reviewed and approved.

→ Covered in detail in Writing Plans

Step 4: Execution & Subagents

Tasks from the plan are dispatched to fresh subagent processes — isolated AI instances with no memory of previous tasks. This prevents errors from one task contaminating the next. Each completed task goes through a two-stage review.

→ Covered in detail in Execution & Subagents

Step 5: Test-Driven Development

Every feature begins with a failing test. The RED-GREEN-REFACTOR cycle is followed strictly. No feature is marked complete without running the test suite and seeing actual passing output.

→ Covered in detail in Test-Driven Development

Step 6: Review

A two-stage review process checks both spec compliance (did we build what was asked?) and code quality (is the code maintainable and correct?). These two concerns are evaluated separately.

→ Part of the Execution & Subagents chapter

Step 7: Completion

After all verification gates pass, the agent presents a clean completion report, helps create a pull request, and ensures the work is properly integrated. Nothing is "done" until it's provably done.

→ Covered in the finishing-a-development-branch skill documentation


Quick Reference: Key Commands

ActionWhat to say to your AI
Start a new feature"I want to add [feature]. Use the brainstorming skill."
Write a plan"We've agreed on the design. Write an implementation plan."
Execute a plan"Execute this plan using subagents."
Run TDD"Implement [feature] using test-driven development."
Review work"Review this implementation before we complete."

What Comes Next

This guide walks through each step of the workflow in depth. The chapters are designed to be read in order for your first pass, then used as reference material afterward.


Remember: Superpowers is not about slowing down. It's about stopping the expensive cycle of write → break → fix → break again. The upfront discipline saves multiples of time downstream.