Open-source agentic skills framework that brings spec-first, TDD-enforced, subagent-driven software development to Claude Code, Cursor, Codex, and 8 other AI coding harnesses.
What it does
Superpowers is a complete software development methodology and composable skills framework for coding agents, built by Jesse Vincent and Prime Radiant. It layers automatic, mandatory workflows (spec, plan, TDD, subagent dispatch, code review) on top of Claude Code, Cursor, Codex, Gemini CLI, and eig
Superpowers imposes a full software engineering discipline on coding agents before they write a single line of code. Most agents jump straight to implementation; Superpowers forces a Socratic brainstorm, a human-approved spec, a bite-sized implementation plan, strict RED-GREEN-REFACTOR TDD, and two-stage code review, in that order, every time.
Superpowers is a composable skills framework and development methodology, released by Jesse Vincent and Prime Radiant, licensed MIT. It works across 11 coding harnesses (Claude Code, Cursor, Codex App, Codex CLI, Gemini CLI, GitHub Copilot CLI, Kimi Code, Antigravity, Factory Droid, OpenCode, and Pi) by installing harness-specific plugins or extensions from the GitHub repository at github.com/obra/superpowers.
Each skill lives in a SKILL.md file with YAML frontmatter that tells the agent when to activate it. Per the using-superpowers bootstrap skill, the agent must invoke any skill with even a 1% chance of relevance, and skill invocations are mandatory, not optional. The priority chain is clear: user instructions in CLAUDE.md or AGENTS.md override skills, skills override default system prompts, and the agent has no way to rationalize itself out of using an applicable skill.
The core workflow runs seven stages: brainstorming (spec refinement), git worktree setup, plan writing (tasks sized 2-5 minutes each, with exact file paths and verification steps), subagent-driven execution, test-driven development, code review between tasks, and branch finishing. Each stage activates automatically; the agent checks for relevant skills before any task.
The multi-harness coverage is a genuine strength, but it is also the main maintenance constraint: the README states that skill updates must work across all supported coding agents, and new skill contributions are generally not accepted. Teams running more than one harness must install Superpowers separately per harness, and any customization through personal skill overrides (shadowing) must stay compatible with all targets. If your team has a custom SDLC that conflicts with TDD or YAGNI, skills can be overridden by user instructions, but the defaults will push back hard.
Superpowers is the sharpest off-the-shelf way to make a coding agent behave like a disciplined engineer rather than an autocomplete engine. Install it once per harness from github.com/obra/superpowers and the methodology runs automatically from the first prompt.
Features
Field notes
Reviewed Jun 26, 2026
Best for
Builder outcomes
Watch out
Tested with
Superpowers is a complete software development methodology and composable skills framework for AI coding agents, released as MIT open-source by Jesse Vincent and Prime Radiant. It installs as a plugin or extension into your coding harness and automatically enforces a structured workflow: Socratic brainstorm, human-approved spec, detailed plan, strict TDD, subagent-driven execution, and code review, in that order. The skills trigger automatically based on context, so you do not need to prompt for them manually.
Installation is harness-specific and must be done separately for each agent you use. The README at github.com/obra/superpowers documents every supported harness: Claude Code (official Anthropic plugin marketplace or Superpowers marketplace), Cursor (plugin marketplace via /add-plugin), Codex App/CLI (official Codex plugin marketplace), Gemini CLI (gemini extensions install), GitHub Copilot CLI (Superpowers marketplace), Kimi Code (built-in plugin manager), Antigravity, Factory Droid, OpenCode, and Pi. Follow the harness-specific instructions in the README for the correct command.
Superpowers is MIT licensed and fully open-source, hosted at github.com/obra/superpowers. It is free to install and use across all supported harnesses. Prime Radiant offers commercial support, additional tooling, and managed spending for enterprise users via sales@primeradiant.com, but those services are entirely optional.
Superpowers works best for developers who want a coding agent to behave like a disciplined engineer rather than an autocomplete engine. It is especially valuable when you need hours of autonomous agent work that stays on plan, when your team wants TDD and code review enforced automatically, or when you are building something complex enough to need a spec-first approach. Per the README, it is not uncommon for the agent to work autonomously for a couple of hours at a time without deviating from the approved plan.
A custom system prompt gives you one-off instructions; Superpowers gives you a versioned, composable library of mandatory workflow skills that activate automatically based on task context. The key difference is the 1% rule: per the using-superpowers bootstrap, the agent must invoke any skill with even a 1% chance of relevance, making the methodology non-negotiable at runtime rather than advisory. Custom instructions can still override any skill, so Superpowers and your own CLAUDE.md or AGENTS.md coexist, with user instructions always taking highest priority.
Superpowers requires a separate install per coding harness, so teams using multiple agents face compounding maintenance. Upstream contributions of new skills are generally not accepted, meaning you cannot easily extend the shared library. The framework enforces TDD, YAGNI, and DRY by default; teams with different SDLC conventions will need to override skills via user instruction files. Telemetry is on by default (a version-tagged logo load from Prime Radiant's servers) and must be explicitly disabled with the SUPERPOWERS_DISABLE_TELEMETRY environment variable.
Agent skill collection for engineers who use Claude Code or Codex, small, composable prompts that enforce TDD, domain modeling, and architecture review.
Best for Engineers already using Claude Code or Codex daily who want to stop architectural and quality driftOpen-source collection of reusable agent skills, shared AGENTS.MD rules, and portable helper scripts for Codex and Claude agents, by steipete.
Best for Developers running multiple Codex or Claude agent projects who need one canonical source for shared rules and skillsAgent skill that audits your codebase with a capable model and writes executable plans for cheaper models to implement, by shadcn.
Best for Teams running AI-assisted development at scale who want high-quality code review without paying for a capable model on every execution step39 mental-model skills for Claude Code, first principles, Bayesian, OODA, pre-mortem, and more, all invoke-by-name with no setup.
Best for Engineers, founders, and analysts who want Claude Code to reason through decisions using named, established mental modelsAnthropic's official skill library for Claude: modular SKILL.md bundles that add repeatable, domain-specific workflows across docs, dev, design, and enterprise tasks.
Best for Teams who need Claude to follow repeatable, specialized workflows across many sessions without re-promptingA catalog of repeatable AI agent workflows with built-in feedback loops, plus an installable skill for finding, adapting, and designing your own loops.
Best for Teams using Codex, Cursor, or Claude Code who want agents to iterate reliably instead of making one-shot guesses