If you’ve been following my journey at Medevel.com, you know I’m obsessed with the intersection of AI engineering and open-source software. We talk about LLMs, Next.js, and Directus all the time. But today, I need to talk about the process.
For months, I’ve been wrestling with a common developer problem: AI agents (like Claude Code, Cursor, or Gemini CLI) are brilliant, but they can also be impulsive. They jump into code without a spec, ignore your existing tests, and, let’s be honest, sometimes create a DRY-violating mess that would make a junior dev blush.
Then I found Superpowers.
What is Superpowers? (Hint: It’s Not Just a Plugin)
Superpowers isn’t just a tool; it’s a complete software development methodology injected directly into your coding agent. It’s built on the philosophy that AI shouldn’t just write code, it should think like a senior engineer.
In our daily workflow, whether I’m building a new feature for the HealthyLife AI ecosystem or fine-tuning the TakeAsIstan marketplace, I’ve started relying on this specific framework to keep things from spiraling into technical debt.
The Workflow That Changed Everything
The magic of Superpowers is that it forces your agent to follow a strict, logical progression. It’s not a suggestion; it’s the law.
1. Brainstorming & Spec First
Before a single line of code is written, the brainstorming skill activates. The agent steps back and asks: “What are we actually trying to do here?” It teases out a spec, shows it to you in digestible chunks, and ensures you’ve signed off on the design before moving an inch.
2. The “Junior Engineer” Plan
Once the design is locked, it creates an implementation plan. The documentation describes the target for this plan as an “enthusiastic junior engineer with poor taste and no judgment.” By assuming the agent might fail, the plan becomes so precise (file paths, exact code blocks, verification steps) that it almost can’t fail.
3. True Red/Green TDD
This is where I fell in love. Superpowers enforces Test-Driven Development (TDD). It writes a failing test, watches it fail, writes the minimal code to pass, and commits. If the agent tries to write code before the test? Superpowers deletes it. It’s brutal, but it’s the only way to ensure 100% reliability.
Using Superpowers with OpenCode & Gemini
As a developer living in Turkey, I’m always looking for ways to optimize my stack for performance and open-source flexibility. Superpowers is incredibly versatile across platforms.
If you’re using OpenCode (one of my favorite open-source environments), getting started is a breeze. You just tell OpenCode to fetch the instructions directly:
Fetch and follow instructions from https://raw.githubusercontent.com/obra/superpowers/refs/heads/main/.opencode/INSTALL.md
And since I’m writing this for my fellow Gemini CLI users, you can install the extension directly from the repo:
gemini extensions install https://github.com/obra/superpowers
Why We Use This Daily
In my daily workflow, I’m often switching between being a medical doctor, a horse riding instructor, and a software developer. Context switching is my biggest enemy.
When I’m working on the Herbal Archive AI or the MOTOID project, I don’t have time to micro-manage an AI that forgets the project context. With the subagent-driven-development skill, I can set a plan, say “Go,” and let the agents work autonomously for hours. They inspect their own work, review it against the spec, and only come back to me when the task is done and verified.
How to Install (The Cheat Sheet)
| Platform | Command |
| Claude Code | /plugin install superpowers@claude-plugins-official |
| Cursor | /add-plugin superpowers |
| Gemini CLI | gemini extensions install https://github.com/obra/superpowers |
| GitHub Copilot | copilot plugin install superpowers@superpowers-marketplace |
The Philosophy: Simplicity over Complexity
The core of Superpowers is YAGNI (You Aren’t Gonna Need It) and DRY (Don’t Repeat Yourself). It’s about evidence over claims. If the agent says the bug is fixed, the systematic-debugging and verification-before-completion skills ensure there’s a test to prove it.
If you are a developer looking to move from “prompting” to “engineering,” you need to give your agent these skills. It turns a chat box into a legitimate development partner.
Are you using AI agents in your workflow yet? If you’ve tried Superpowers or are running it on OpenCode, drop a comment or reach out on Medevel. It’s time we stop writing code and start architecting solutions.
Special thanks to Jesse (obra) for building this incredible open-source treasure. Consider sponsoring his work if this tool helps you stay as productive as it helps me!
