gitwtfhub

obra/superpowers

🔥 On Fire196,696ShellAudience · developerComplexity · 3/5ActiveLicenseSetup · easy

TL;DR

A methodology and skill set that makes AI coding agents slow down, plan carefully, and execute disciplined development—spec approval, test-driven code, code review, all before merge.

Mindmap

mindmap
  root((Superpowers))
    What it does
      Clarify requirements
      Build implementation plan
      Execute with discipline
      Review before merge
    Key skills
      Brainstorming spec
      Test-driven development
      Code review process
      Git worktree isolation
    Use cases
      Non-trivial features
      Quality-critical tasks
      Team-like workflows
    Supported agents
      Claude Code
      Cursor
      GitHub Copilot CLI
      Gemini CLI
    Tech stack
      Shell scripts
      Git
      Subagent orchestration

Why anyone would actually build this

REASON 1

Build a feature with an AI agent that first clarifies requirements and gets your approval before coding.

REASON 2

Run test-driven development workflows where the agent writes failing tests, then implementation, then refactors.

REASON 3

Have an AI agent propose and execute a detailed plan for a non-trivial task, with code review at each step.

REASON 4

Integrate disciplined AI-assisted development into your existing coding agent (Claude, Cursor, Copilot, etc.).

Stack

ShellGitClaudeGeminiGitHub Copilot

Spinning it up

Difficulty · easy Time to first run · 5min
Use freely for any purpose, including commercial use, as long as you keep the copyright notice.

Wtf does this do

Superpowers is a complete software development methodology for AI coding agents, built as a set of composable skills plus initial instructions that ensure the agent actually uses them. The idea: instead of letting your AI agent jump straight into writing code when you ask it to build something, Superpowers makes the agent slow down, clarify what you actually want, propose a plan, and then execute it under a disciplined process.

It works by intercepting the moment you fire up your coding agent. The agent runs a brainstorming step that teases out a spec from conversation, then shows it back in chunks short enough for you to read and approve. Once approved, it builds an implementation plan detailed enough for an inexperienced engineer to follow, emphasizing test-driven development (red/green/refactor), YAGNI, and DRY. When you say go, it launches a subagent-driven development process where each engineering task is handled by a fresh subagent and reviewed for spec compliance and code quality before moving on. Skills cover brainstorming, plan-writing, git worktree isolation, executing plans, test-driven development, code review (both requesting and receiving), debugging, and a finishing-branch workflow that decides between merge, PR, keep, or discard.

You'd use Superpowers when you want a coding agent to behave like a careful engineer rather than a code-spewing autocomplete — especially for non-trivial tasks where missing requirements or skipping tests would cost you later. It supports many coding-agent harnesses including Claude Code, Codex CLI and App, Factory Droid, Gemini CLI, OpenCode, Cursor, and GitHub Copilot CLI, with separate install instructions for each. The codebase is shell-based, MIT-licensed, and built by Jesse Vincent. The full README is longer than what was provided.

Yoink these prompts

Prompt 1
I want to use Superpowers with Claude Code to build a feature. Walk me through the brainstorming and spec-approval flow.
Prompt 2
How do I set up Superpowers to enforce test-driven development when my AI agent writes code?
Prompt 3
Show me how Superpowers uses git worktrees to isolate each engineering task and keep the main branch clean.
Prompt 4
I'm using Cursor. How do I install and configure Superpowers to add code review and planning steps before my agent commits?
Prompt 5
What does the subagent-driven development process in Superpowers look like, and how does it decide between merge, PR, keep, or discard?
View the repo → Decode another repo

Generated 2026-05-18 · Model: sonnet-4-6 · Don't trust strangers blindly — verify against the repo.