obra /
superpowers
A set of skills that makes coding agents plan, test and review before they write code
- 303.8k stars
- 9.7k stars since 3 Oct
- Shell
- Updated today
- Some experience
- Agent skills & plugins
Coding agents often jump straight into writing code before they understand what you want. Superpowers adds a set of skills, plus initial instructions that make the agent check for them, so it slows down and follows a process first.
When you start building something, the agent asks what you are trying to do and drafts a spec, which it shows you in short sections. After you approve the design, it writes a plan broken into small tasks with exact file paths and verification steps. Then it works through the tasks, using test-driven development and a review after each task (or one review at the end, in the cheaper mode). The README says agents can run for a couple of hours this way without leaving the plan.
The skills trigger on their own, so you do not call them by hand. The README stresses that the workflows are mandatory, not suggestions. The same repository works across many agent tools, and the project does not generally accept new skills from contributors.
What it does
Brainstorm before code
The brainstorming skill asks questions, explores alternatives and presents the design in sections for you to approve. It saves a design document.
Bite-sized implementation plans
The writing-plans skill splits approved designs into tasks of 2 to 5 minutes, each with file paths, complete code and verification steps.
Subagent-driven development
A fresh subagent handles each task, with a two-stage review for spec compliance and then code quality. An inline executing-plans option is cheaper.
Enforced test-driven development
The RED-GREEN-REFACTOR cycle is required: write a failing test, watch it fail, write minimal code, and delete code that was written before tests.
Debugging and session diagnosis
Skills for systematic root-cause debugging and verification before declaring success. A diagnosing-superpowers skill reads a session transcript to explain what went wrong.
What you could build
- Turn a rough feature idea into an approved design and a step-by-step plan before any code is written.
- Let a coding agent implement a multi-task feature in an isolated git worktree with tests and reviews at each step.
- Debug a stubborn bug with a structured four-phase root-cause process.
- Write and test your own agent skills by following the included writing-skills guide.
- Find out why a skill misfired in an agent session by asking the agent to diagnose the transcript.
Get started
Install is per agent tool, so pick the section for yours in the README and run its command. In Claude Code, use the official marketplace command, then start a new session and describe something you want to build to see brainstorming begin.
/plugin install superpowers@claude-plugins-officialgemini extensions install https://github.com/obra/superpowersdevin plugins install obra/superpowersCommands are copied from the project's README. Read it on GitHub for the full setup.
Anything better than superpowers?
Superpowers is the strongest pick for an enforced plan, test-first, review workflow, with the largest following and active maintenance. Look elsewhere if you want smaller, editable skills that don't own your process, or a much bigger bundle.
-
skillsAlternative You want small, editable skills (grilling, TDD, diagnosis, tickets, review) you can pick and tweak individually rather than adopting a fixed end-to-end workflow. Trade-off: You get no enforced pipeline or subagent-driven development, so you must decide when to invoke each skill.
-
compound-engineering-pluginAlternative You want a similar brainstorm, plan, work, review loop that also saves lessons in your repo so later tasks start with more context. Trade-off: Much smaller community (about 26k stars) and 36 skills, so more to learn, with less emphasis on strict test-first.
-
agent-skillsAlternative You prefer a spec-to-ship lifecycle with 25 skills and 9 slash commands that you trigger explicitly, including a ship stage. Trade-off: Smaller adoption and less emphasis on subagent-driven execution and enforced TDD.
-
ECCAlternative You want a much larger harness with hooks, rules, memory and many agents bundled in, not only workflow skills. Trade-off: Far heavier and more complex, with 293 skills to sift through and a steeper setup than Superpowers' focused workflow.