Welcome to Superpowers
A practical guide to making AI coding agents work with discipline — not just fast.
What is Superpowers?
Superpowers is a collection of "skills" — structured behavioral rules — that you install into an AI coding agent (like Claude Code, Cursor, or Gemini CLI). Once installed, your AI assistant stops behaving like an eager intern who dives straight into code and starts behaving like a senior engineer who thinks before acting.
In plain terms: Superpowers teaches your AI to ask before coding, plan ahead, and verify results before claiming anything is done.
It doesn't make the AI smarter. It makes the AI more disciplined.
Think of it like the difference between a contractor who starts hammering nails the moment they arrive versus one who first walks the site, asks about the blueprints, checks for plumbing behind the walls, and then builds methodically. Both contractors know how to build. Only one of them won't cause an expensive mess.
The Problem It Solves
Most people who have tried AI coding tools have experienced this cycle:
Without Superpowers
| What you say | What the AI does |
|---|---|
| "Add a login feature" | Immediately writes 400 lines of code across 8 files |
| "Does it work?" | "Yes! I've implemented the complete login system." |
| "I ran the tests" | 3 tests fail, 2 files are missing imports |
| "Fix the tests" | Rewrites half the codebase and introduces 2 new bugs |
| "Is it done now?" | "Yes! Everything looks great." |
| Reality | Nothing works. You've lost an hour. |
The AI wasn't lying — it genuinely believed it was done. But it never ran the tests. It never verified. It just said it was done because that's what you wanted to hear.
With Superpowers
| What you say | What the AI does |
|---|---|
| "Add a login feature" | Asks 3 clarifying questions about requirements |
| You answer | Proposes 2 different approaches with trade-offs |
| You pick one | Writes a detailed plan for your approval |
| You approve | Implements step by step, running tests after each step |
| "Is it done?" | Runs the full test suite, shows you the actual output, then says done |
| Reality | It works. You have tests. The plan is documented. |
The difference isn't capability. It's workflow discipline.
How It Works (Simple)
Superpowers enforces a 7-step workflow for every significant coding task:
Step 1 — Brainstorming
Before any code is written, the AI explores the problem with you. It asks one question at a time, presents multiple approaches, and waits for your approval on the design. No code until you say so.
Step 2 — Isolation
Work happens in an isolated environment (a git worktree) so your main codebase is never affected by in-progress experiments. Like a separate sandbox.
Step 3 — Planning
The AI writes a detailed, step-by-step implementation plan. Each task is small enough that a fresh engineer could complete it in 2–5 minutes. The plan is reviewed and approved before any execution begins.
Step 4 — Execution
Tasks from the plan are executed — often by separate "subagent" processes — one at a time, with clear hand-offs and status reporting. No task bleeds into the next.
Step 5 — Testing
Tests are written before production code (Test-Driven Development). The AI cannot claim a feature is complete unless tests pass with real, observed output — not assumed output.
Step 6 — Review
After implementation, a two-stage review checks: (1) did we build what was specified? and (2) is the code quality acceptable? These are separate concerns and both must pass.
Step 7 — Completion
Only after all verification gates pass does the AI declare work complete. It then helps you integrate, create a pull request, or close the task cleanly.
Who Uses Superpowers?
Superpowers has been adopted widely across the developer community:
- 86,000+ GitHub stars — one of the most-starred agentic toolkits
- 6,800+ forks — teams actively customizing it for their workflows
- 24 contributors — active open-source community
- 6+ supported platforms — works with Claude Code, Cursor, Gemini CLI, Codex, OpenCode, and more
It's used by solo developers who want to stop babysitting their AI tools, by engineering teams who want consistent quality from AI-assisted development, and by product managers who want predictable delivery when working with AI agents.
Is It Right For Your Team?
Use the checklist below to see where Superpowers fits your situation:
For Developers
For Product Managers & Team Leads
For QA Engineers
If you checked 3 or more items in any section, Superpowers is built for you.
Getting Started — Where to Read Next
Different roles benefit from different parts of this guide:
| Role | Recommended Path |
|---|---|
| Developer | Start with Getting Started → Brainstorming → Writing Plans → all chapters |
| Product Manager | Read Brainstorming and Writing Plans to understand how requirements become plans |
| QA Engineer | Jump to TDD and Execution & Subagents for the verification and testing model |
| Team Lead / Architect | Read all chapters — Superpowers is a team-wide discipline, not just a developer tool |
Bottom line: Superpowers doesn't replace your judgment. It makes sure the AI asks for your judgment at the right moments, documents the decisions, and proves its work before calling anything done.