obra/superpowers: An agentic skills framework & software development methodology that works.

An agentic skills framework & software development methodology that works. - obra/superpowers

Summary

obra/superpowers is a software development methodology and skill library for coding agents (Claude Code, Codex, Cursor, Gemini, and others) that injects a structured workflow into the agent at session start. Rather than letting the agent jump straight to code, it enforces a pipeline of skills—brainstorming, plan writing, test-driven implementation, code review, and branch finishing—that trigger automatically based on context. It ships as a plugin per supported agent harness rather than a standalone CLI.

What it is useful for

Superpowers addresses a familiar pain point with AI coding assistants: they often start writing code before the problem is understood, skip tests, or drift from an agreed plan. It is useful when you want an agent to:

• Ask clarifying questions and produce a written design before touching code. The brainstorming skill refines rough ideas, explores alternatives, and presents the spec in digestible sections.
• Produce an implementation plan with tasks scoped to 2–5 minutes, each with exact file paths, complete code, and verification steps.
• Enforce TDD as a RED-GREEN-REFACTOR discipline, including deletion of code written before tests.
• Run structured code review between tasks and a formal branch-finishing step that verifies tests and offers merge/PR/keep/discard options.
• Get consistent behavior across many coding harnesses—the same skills target Claude Code, Codex, Cursor, Devin, Gemini, Copilot, Grok, Kimi, OpenCode, Pi, Qwen, Hermes, Antigravity, and Muse.

How engineers can use it

Installation differs per harness; you install Superpowers separately for each agent you use. For Claude Code:

/plugin marketplace add obra/superpowers-marketplace
/plugin install superpowers@superpowers-marketplace

For other agents, the README documents parallel commands—e.g., gemini extensions install https://github.com/obra/superpowers, agy plugin install https://github.com/obra/superpowers, droid plugin install superpowers@superpowers, or pi install git:github.com/obra/superpowers. Full per-harness steps and harness-specific notes are in the README.

After installation, the workflow is implicit: start a new session, describe what you want to build, and the agent should auto-trigger brainstorming before writing code. The README documents a verification prompt—"Let's make a react todo list"—where a working install triggers brainstorming instead of jumping into code. If a session misbehaves, ask the agent to "figure out what went wrong with superpowers in this session"; it invokes the diagnosing-superpowers skill to inspect the transcript.

Documented limitations: contributions of new skills are not generally accepted, Hermes has no post-compaction hook (very long sessions may need a fresh restart), and an optional visual companion loads a versioned logo from the maintainers' site by default—disable with SUPERPOWERS_DISABLE_TELEMETRY. License is MIT.