Getting started with superpowers

This is the companion to the episode. The promise is simple: we help you get started developing with agents, step by step, across a three-part series. Part one, this video, is all about superpowers, one free tool that gives your coding agent a real developer's process from day one. First the theory, then we install it and build a real feature so you can watch the theory run.

What a coding agent is

A coding agent takes your request and works in a loop. It writes code, runs it, reads the result, and tries again until the task is done. It runs in any harness; we use Claude Code. You stay in charge: you approve the design and steer, and the agent does the typing.

The problem superpowers solves

Left on its defaults, an agent writes code with no design, skips tests, and tells you it is done when it is not. That is where beginners plateau. The code runs once in the demo, then falls over on the second case, and you had no way to catch it before shipping.

Meet superpowers

• superpowers is a free, open-source plugin for coding agents, made by Jesse Vincent. A plugin is an add-on that changes how the agent behaves; this one installs a whole way of working. Install is one command, and there is nothing to configure after.
• It is built from skills. A skill is a single readable markdown file that teaches the agent one process, with checklists and worked examples. You can read the entire system in an afternoon, which makes the repo a textbook as much as a tool.
• It is widely used: about a quarter-million stars on GitHub, and it ships in Anthropic's official plugin marketplace.

The workflow

superpowers runs every feature through five stages, and each stage is a skill you can open and read:

• Brainstorm. It asks questions first. No code until you agree on a design.
• Worktree. It works on a separate copy of your project, so your main code stays safe.
• The plan. It breaks the design into small tasks, each with the exact files and steps.
• Subagents. One fresh agent implements each task, and a separate one reviews the work.
• Finish. Tests pass, you merge, and it cleans up after itself.

The Iron Laws

Three rules run underneath every stage. They are the discipline that normally takes years to build:

• Test first. Write a failing test, then just enough code to pass it.
• Root cause first. Find the real cause before you touch the code.
• Prove it works. Run the check and read the output before you call it done.

What you walk away with

• The upgrade is a process you can install today, for free, in the agent you already use.
• You now know each stage by name, so you can steer the work instead of watching it happen.

Where to go next

• The links below are the repo, plus the community and live streams behind the work.
• This is part of a short series. Subscribe if you want the parts that follow.