Superpowers

Superpowers is a complete software-development methodology for coding agents, built as a set of composable skills. It’s the engine behind the spec-driven habit: instead of letting the agent jump straight to code, it runs a disciplined loop. Open source, by Jesse Vincent — github.com/obra/superpowers.

The method

When you start building, Superpowers makes the agent:

1. Brainstorm — ask clarifying questions about what you actually want.
2. Design — present the approach in digestible chunks for your approval.
3. Plan — write a detailed implementation plan, broken into 2–5 minute tasks.
4. Execute — subagent-driven development with code-review checkpoints.
5. Test-first — RED → GREEN → REFACTOR; write the test before the code.

Underneath it all: reduce complexity and verify with evidence before declaring anything done. (That last one is the antidote to “looks done to me.”)

The skills it ships

Skills trigger automatically when relevant — you don’t memorise commands. The core set:

This is also the clearest real-world example of the Skills concept: a library of SKILL.md procedures that compose into a methodology.

Install it (in this kit’s Claude Code)

In the browser VS Code (localhost:8080), open the Claude Code terminal and run:

/plugin install superpowers@claude-plugins-official

(Other agents have their own install line — Cursor, Gemini CLI, Pi, etc. — see the repo.)

Use it

You don’t call the skills by hand — just start building and they kick in. Tell Claude Code “implement the next roadmap TODO in latex-starter” and watch it brainstorm → plan → execute → verify instead of dumping code.

Why it matters: this is the single habit most worth copying to your team. It’s what turns an agent from “fast but sloppy” into “spec-first and verifiable” — the through-line across all three tracks.

Try it: install it, then ask Claude Code to add the explain_steps tool to the starter (roadmap Stage 2). Make it brainstorm and write a plan before it touches code.