What Is Superpowers? The Skills Pack That Teaches Claude Code to Plan, Test First and Review
You install Superpowers in your coding agent (Claude Code, Codex, Cursor and others); before starting work, the agent asks what you want, gets the design approved, writes a plan, implements it test-first and has the work reviewed. It contains 15 skills; they kick in on their own, so you don't need to call them one by one. It is MIT licensed and free; Claude Code's official marketplace is pinned to an older version (6.4.1), and the author's marketplace is used for the latest version (7.0.0).
- What it does
- Teaches your coding agent a way of working
- Type
- Skills pack (15 skills, 1 hook)
- Where
- Claude Code, Codex, Cursor, Gemini CLI and 12 more environments
- In the official marketplace
- Pinned to version 6.4.1 (current: 7.0.0)
- Windows
- The hook needs Git for Windows (bash)
- Turkish
- Not stated in official sources
- Cost
- The pack is free, MIT licensed
What does Superpowers do?
Coding agents often start writing code before fully understanding what to do. Superpowers is a methodology pack that tells the agent to stop first and ask what you want, get the design approved, write a plan, implement test-first and review the work. A skill is a ready-made instruction set that tells the assistant how to do a task; Superpowers' skills kick in on their own for the right kind of work. The README describes them as "mandatory workflows, not suggestions".
Who it's for: developers who work regularly with Claude Code or a similar agent on code projects and are tired of the agent working without a plan or tests. If you don't write code, it won't help you.
What's inside Superpowers?
The pack carries 15 skills, a SessionStart hook and some helper scripts. There's no separate agent, slash command or MCP server (the old commands were removed in 5.1.0).
| Skill | When it kicks in |
|---|---|
| using-superpowers | At the start of every conversation: requires checking whether a skill applies and, if so, invoking it before answering |
| brainstorming | Before creative or building work: asks what you want, writes up the design and gets it approved |
| using-git-worktrees | When the work needs isolation: sets up a separate git worktree |
| writing-plans | For multi-step work: writes a plan split into tasks before touching code |
| subagent-driven-development | When executing the plan in this session: a fresh subagent and a review per task |
| executing-plans | When you say to execute the plan yourself in one session: the agent does the tasks itself, with a single review at the end |
| test-driven-development | For every feature and bug fix: a failing test first, then code |
| requesting-code-review | When a task is done or before merging: an independent review subagent |
| receiving-code-review | When review feedback arrives: verifies it before applying it blindly |
| finishing-a-development-branch | When the work is done and tests pass: offers merge, open a PR or leave it |
| systematic-debugging | On a bug or failing test: a root-cause process before proposing a fix |
| verification-before-completion | Before saying "done": runs the relevant command and shows the output |
| dispatching-parallel-agents | With 2 or more independent tasks: one subagent per task |
| diagnosing-superpowers | When a Superpowers session goes badly: produces an evidence-based report from the transcript |
| writing-skills | When writing a new skill: a test-first method adapted to skill writing |
- All at once or one by one: the pack is installed in one go, not picked individually; the skills kick in on their own depending on the type of work.
- Conflicts: don't install the same pack from two marketplaces at once.
How do you install Superpowers?
Each environment has its own install command and is installed separately. The commands are taken verbatim from the official README and marketplace entries.
› /plugin install superpowers@claude-plugins-official› /plugin marketplace add obra/superpowers-marketplace› /plugin install superpowers@superpowers-marketplace- Inside a session,
/plugin installdoesn't install right away; it opens the plugin's details panel. Choose the scope there: just you, everyone on this project, or just you and only this project. - To activate it right away in the open session, type
/reload-pluginsor restart Claude Code. - To install from the shell without opening a session:
claude plugin install superpowers@claude-plugins-official.
!Anthropic's official marketplace entry pins Superpowers to a specific commit (5bf4e78, version 6.4.1); the latest release on GitHub is 7.0.0. If you install from the official marketplace, you won't see the new features in the README. Don't install both.
- In the Codex app, click Plugins in the sidebar, find Superpowers in the Coding section and press the + next to it.
- In the Codex CLI, type
/plugins, type superpowers in the search box and choose Install Plugin.
In Codex, the subagent skills need the multi_agent = true setting in the config (the skill's own reference file).
› /add-plugin superpowers› gemini extensions install https://github.com/obra/superpowersThe README also describes how to install in Antigravity, Devin CLI, Factory Droid, GitHub Copilot CLI, Grok Build CLI, Kimi Code, OpenCode, Pi, Qwen Code, Hermes Agent and Muse. Examples: for Copilot CLI, copilot plugin marketplace add obra/superpowers-marketplace and copilot plugin install superpowers@superpowers-marketplace; for Qwen Code, qwen extensions install obra/superpowers.
How to tell it's installed: in Claude Code, check /plugin › Installed, or run claude plugin list in the shell. The README's test: in a new session, type "Let's make a react todo list"; if it works, brainstorming kicks in on its own before any code is written.
How do you use Superpowers?
After installing, you don't need to type any special command: when you ask for a task, the relevant skill kicks in on its own. If you want to call one manually, according to the Claude Code docs plugin skills are invoked as /plugin-name:skill-name (for example /superpowers:brainstorming).
If the many steps create friction, say "don't use the Superpowers workflow for this task": according to the skill's own text, your instructions (CLAUDE.md and your direct request) take precedence over the skills.
While working, the skill files create the docs/superpowers/ (design and plan documents), .superpowers/ and .worktrees/ folders in the project; details in the "What does it access?" section.
Which tools does Superpowers work in?
| Claude CodeYes | Via the official marketplace and the author's marketplace; claude.ai/code cloud sessions don't load a plugin installed on your machine. |
|---|---|
| Codex (app and CLI)Yes | The README lists the install steps. |
| Cursor, Gemini CLI, GitHub Copilot CLIYes | The README gives an install command for each. |
| Antigravity, Devin CLI, Factory Droid, Grok Build CLI, Kimi Code, OpenCode, Pi, Qwen Code, Hermes Agent, MuseYes | The README describes the install route; see each environment's own docs for details. |
| claude.ai chat and CoworkUnclear | The Claude listing page only shows "Install in Claude Code"; whether it can be used in chat is not stated in the official documentation. |
"Not in the docs" doesn't mean it doesn't work. The list is compiled from the README and the Claude Code documentation.
What does Superpowers write to my files, and what does it run?
- docs/superpowers/specs and plans: design and plan documents (the design document is committed)
- .worktrees/: separate worktrees; .superpowers/: subagent working files (kept out of git)
- Project setup in a new worktree: npm install if there's a package.json, cargo build if there's a Cargo.toml, and so on; then it runs the tests
In Claude Code, a SessionStart hook runs a local bash script at the start of every session (on startup, clear and after compaction) that adds the text of the using-superpowers skill to the agent's context. According to the Claude Code docs, command hooks run with your full permissions, without isolation; reading the hooks folder before installing is the approach the docs recommend.
The skills and the hook run locally. The only part that sends data is brainstorming's optional visual companion: the logo in its frame is loaded from primeradiant.com with the version number; it can be turned off with three environment variables.
It cleans up only the worktrees it created itself; if there are unsaved files it doesn't force-delete, it asks you. It deletes a branch only if you explicitly ask.
GitHub's repository and global advisory lists and the OSV database list no entries; the repository has no SECURITY.md. No published advisory is listed; this doesn't mean it is safe.
!This information is compiled from the official repository; it is not a security audit. The Claude Code docs state that Anthropic doesn't audit the contents of plugins in the official marketplace.
Is Superpowers free?
The pack itself is free and MIT licensed: no account, API key or MCP server needed. The real cost comes from the agent it runs in: Claude Code requires a paid Claude plan or an API account on most surfaces (the desktop app's Code tab requires a paid subscription), and other agents' own plans apply.
Working with subagents can use more tokens because it opens a new implementer and review session for each task. The README gives no cost figure; after installing, the "Always-on" line in the claude plugin details superpowers command shows how many tokens the plugin adds to every session.
- Using, copying, modifying and distributing it, including commercially (MIT)
- Removing the copyright and license notice from copies
Source: LICENSE (MIT, copyright holder Jesse Vincent) and README. The README also says Prime Radiant can be contacted for enterprise use. This is not legal advice.
Common problems and fixes
| What happens | What to do |
|---|---|
| On Windows, "SessionStart hook error" appears at session start, or the agent doesn't remember Superpowers at all | The hook runs with bash: Git for Windows must be installed. In version 6.4.1 from the official marketplace, if bash isn't found the hook exits silently and no session-start context is added. The 7.0.0 release notes list fixes for this. |
| Skills don't trigger after installation, or the plugin doesn't appear | Check whether it's listed under /plugin › Installed and in claude plugin list. If you installed it inside a session, run /reload-plugins or open a new session. Don't install the plugin from two marketplaces at once. |
| Reported by users Slowness, tokens or session limits running out quickly, many steps even for small tasks | A common report in GitHub, Hacker News and YouTube comments. On the official side: the README describes executing-plans as the cheapest route; you can also say "don't use the Superpowers workflow for this task". |
| The /brainstorm, /write-plan, /execute-plan commands from older tutorials don't exist | Removed in 5.1.0. Skills kick in on their own or are called like /superpowers:brainstorming. The Claude listing page still describes the old commands. |
| After installing from the official marketplace, I don't see the new features from the README | The official marketplace entry is pinned to 6.4.1. For 7.0.0, use the author's marketplace (superpowers-marketplace); don't install both. |
| A docs/superpowers/ or .worktrees/ folder appeared where I didn't expect it | The write locations are fixed in the skills: docs/superpowers/specs and plans, .superpowers/, .worktrees/. No configurable environment variable is documented; Reported by users there are requests on GitHub to make this configurable. |
How do you disable or uninstall Superpowers?
- In Claude Code: Select Superpowers under
/plugin› Installed; choose Disable or Uninstall. In the shell:claude plugin disable superpowers@<marketplace>orclaude plugin uninstall superpowers@<marketplace>; the marketplace name is claude-plugins-official or superpowers-marketplace. - If you installed it at project scope Claude Code asks whether to disable it just for you or remove it for everyone.
- Folders created: docs/superpowers/, .superpowers/ and .worktrees/ are the project's own files; the official docs say they aren't deleted on uninstall. You can list worktrees with
git worktree listand remove them yourself. - Other environments: The README only gives install and update commands; to uninstall, use the environment's own plugin management.
To opt out without disabling it completely, it's enough to say "don't use the Superpowers workflow for this task" or pick the "skip questions" option in brainstorming's first message.
Frequently asked questions
What is Superpowers?
A free, MIT-licensed skills pack that teaches your coding agent a software development methodology. With 15 skills, the agent asks what you want before starting, gets the design approved, writes a plan, implements test-first and has the work reviewed.
How do I install Superpowers in Claude Code?
In Claude Code, type /plugin install superpowers@claude-plugins-official and choose the scope in the panel that opens. This official marketplace entry is pinned to version 6.4.1; for the latest version (7.0.0), first type /plugin marketplace add obra/superpowers-marketplace, then /plugin install superpowers@superpowers-marketplace.
Is Superpowers free?
The pack is free and MIT licensed. Costs come from the plan of the agent it runs in and the tokens used; Claude Code requires a paid Claude plan or an API account on most surfaces. Working with subagents can use more tokens.
Which tools does Superpowers work in?
The README describes installation for Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot CLI and Antigravity, Devin CLI, Factory Droid, Grok Build CLI, Kimi Code, OpenCode, Pi, Qwen Code, Hermes Agent, Muse. It is installed separately in each environment.
Do I need to call the skills one by one?
No, the skills kick in on their own depending on the type of work. If you want to call one manually, type something like /superpowers:brainstorming. Your instructions take precedence over the skills: you can say not to use it for a task.
What does Superpowers write to my files?
It creates the docs/superpowers/ (design and plan documents), .superpowers/ and .worktrees/ folders in the project and opens a separate git branch and worktree. In the worktree it can run project setup (such as npm install) and the tests. It asks you before deleting anything.
Does Superpowers collect data?
The skills and the hook run locally. The only part that sends data is brainstorming's optional visual companion: the logo in the frame is loaded from primeradiant.com with the version number, and it can be turned off with three environment variables.
Does Superpowers work on Windows?
The hook requires bash: on Windows, Git for Windows must be installed. In version 6.4.1 from the official marketplace, if bash isn't found the hook exits silently and no session-start context is added; the 7.0.0 notes list fixes for this.
How do I disable or uninstall Superpowers?
In Claude Code, choose Disable or Uninstall under /plugin > Installed; in the shell, claude plugin uninstall superpowers@<marketplace>. To opt out without disabling it completely, just tell it not to use it for a task.
Tasks and setup kits that use Superpowers
Related guides
Sources
- Superpowers README
- Superpowers release 7.0.0
- Superpowers hook definition
- Superpowers LICENSE (MIT)
- Claude official plugin marketplace entry
- Claude Code: installing and managing plugins
- Author's marketplace
Not clear in the official sources: Whether Codex, Cursor and other marketplaces actually list Superpowers (only the README says so) · whether it can be used in claude.ai chat · real token and cost figures (the README gives none) · reply quality in Turkish · when the official marketplace pin will be updated. Each is covered in the relevant section, along with what you can do.





