Superpowers for Claude Code: Setup, Skills and Use Cases (2026)

Superpowers is an open-source skills framework that gives AI coding agents a disciplined software development process. Installed as a plugin in Claude Code (and a dozen other coding agents), it makes the agent clarify what you want, write a plan, work test-first and review its own work before calling anything done, automatically, without you invoking each step. It is MIT-licensed, built by Jesse Vincent (GitHub: obra) and the team at Prime Radiant, and had close to 300,000 GitHub stars as of October 2026.

Below: how it works, what is inside, how to install it, and where it actually helps.

Why coding agents need it

Anyone who has spent real hours with Claude Code, Cursor or Codex knows the pattern: the agent writes something that runs, declares victory, and skips the parts that make code maintainable, such as tests, edge cases and a reviewable history. Superpowers fixes this at the process level instead of hoping a better prompt will. Its own description is "an agentic skills framework and software development methodology that works."

How the Superpowers workflow works

The skills trigger on their own as a session moves forward. The core loop has seven steps:

1. Brainstorming. Before any code, the agent asks what you actually want and why, then plays its understanding back for correction. Larger projects get a written design document.
2. Using git worktrees. After you approve the design, it creates an isolated workspace on a new branch, runs project setup and confirms the tests pass before it starts.
3. Writing plans. The work is broken into bite-sized tasks of 2 to 5 minutes each, with exact file paths and verification steps.
4. Subagent-driven development or executing plans. Either a fresh subagent handles each task with a review after each one (the most thorough option), or the plan runs inline in one session with a single review of the whole branch at the end (the cheapest option).
5. Test-driven development. Red, green, refactor: write a failing test, watch it fail, write the minimal code, watch it pass, commit.
6. Requesting code review. Between tasks the work is reviewed against the plan, and critical issues block progress.
7. Finishing the branch. When the tasks are done, it verifies the tests and offers to merge, open a pull request, keep or discard the work.

The project describes these as mandatory workflows, not suggestions: the agent checks for a relevant skill before every task.

Which skills are included

• Testing: test-driven development, with a reference of testing anti-patterns.
• Debugging: systematic debugging (a four-phase root-cause process), verification before completion, and a skill that diagnoses what went wrong in a Superpowers session itself.
• Collaboration: brainstorming, writing and executing plans, dispatching parallel agents, requesting and receiving code review, git worktrees, finishing a branch, and subagent-driven development.
• Meta: writing new skills, and an introduction to the skills system.

How to install Superpowers in Claude Code

Superpowers is in Anthropic's official Claude plugin marketplace. Inside Claude Code, run:

/plugin install superpowers@claude-plugins-official

Or install it from the project's own marketplace:

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

It also installs in Cursor, the Codex app and CLI, Gemini CLI, GitHub Copilot CLI, OpenCode, Qwen Code and several other coding agents; each one needs its own install. The current commands for every agent are in the project's README on GitHub at obra/superpowers.

Check that it works: start a fresh session and ask for something new, for example "Let's make a React to-do list." A working install starts with brainstorming questions instead of writing code straight away.

Superpowers use cases

• Building a new feature from a vague idea. Brainstorming and planning turn "add team accounts" into a reviewed design and a task list before any code exists, which is where most agent rework comes from.
• Long autonomous runs. With subagent-driven development, the project reports agents working for a couple of hours at a time without drifting from the plan, because every task is small, specified and reviewed.
• Fixing hard bugs. Systematic debugging makes the agent trace the root cause and verify the fix, instead of patching the first symptom it finds.
• Refactoring safely. Test-first work and a clean test baseline in an isolated worktree mean a refactor is proven by tests before it touches your main branch.
• Delivering client work with AI agents. For agencies, the biggest risk in agent-assisted delivery is code that looks finished and is not. A plan, tests and a review gate on every task make the work reviewable and the history easy to audit or roll back.
• Teams standardising how agents work. Because the skills apply the same process in every session, different developers get consistent output from their agents.

When it is not the right tool

• One-off scripts and throwaway prototypes. The upfront questions, plans and tests are overhead when the code will not live long.
• Tight token budgets. A fresh subagent and review per task is the most thorough mode and the most expensive one. The executing-plans mode trades some rigour for cost.

FAQ

Is Superpowers free? Yes. It is open source under the MIT license. Prime Radiant offers commercial support for enterprise use.

Does it only work with Claude Code? No. It supports Claude Code, Cursor, Codex, Gemini CLI, GitHub Copilot CLI, OpenCode, Qwen Code and other coding agents, installed separately in each.

Do I have to call the skills myself? No. The skills trigger automatically based on what you are doing in the session.

What if a session misbehaves? Ask the agent to figure out what went wrong with Superpowers in this session. It runs a diagnosing skill that reads the transcript and reports what happened, with evidence.

Does it send data anywhere? The optional visual companion loads a logo from Prime Radiant's site that includes the Superpowers version; it does not include project details. It can be turned off with the SUPERPOWERS_DISABLE_TELEMETRY environment variable.

If you are building with AI agents in production, the same discipline applies beyond coding tools: see why AI agents break in production, and the fixes that worked for us.

Key takeaways

• Superpowers gives coding agents a full development process: brainstorm, plan, test first, review, finish.
• The skills trigger automatically; you install it once per coding agent.
• In Claude Code: /plugin install superpowers@claude-plugins-official.
• It shines on real features, long autonomous runs, debugging and refactors, and is overkill for throwaway scripts.
• It is free, MIT-licensed and works across most major coding agents.