Superpowers for Claude Code is a free, MIT licensed plugin of 15 agent skills plus one startup hook that makes the agent follow a fixed process: ask what you want and why, write a short design, turn it into a plan of small tasks, build each task test first with a review after it, and verify before it says the work is done. Install it with /plugin install superpowers@claude-plugins-official. It suits multistep features where you can spare tokens and a few minutes of questions; for one line fixes it mostly adds overhead.
The project is obra/superpowers on GitHub, built by Jesse Vincent and the team at Prime Radiant. It first shipped in October 2025, and on November 19, 2026 it had 321,694 stars and sat on GitHub's trending page with about 2,000 new stars that day. The current release is v7.0.2 from November 3, 2026. Despite the name, it runs on more than a dozen coding agents, not only Claude Code.
What Superpowers actually is
Superpowers is a set of agent skills: Markdown SKILL.md files that an agent loads when a task matches the skill's description. What makes it different from a loose skill collection is the bootstrap. A SessionStart hook prints the full text of one skill, using-superpowers, into the session at startup, after /clear and after compaction. Claude Code's hooks documentation says text returned this way goes into Claude's context window as a system reminder.
That bootstrap skill is blunt. It tells the agent that if there is even a 1% chance a skill applies, it must invoke it before any response, including clarifying questions. It then lists the excuses agents give for skipping a skill ("this is just a simple question", "let me explore the codebase first") and answers each one. The rest of the package is the process those rules point to.
The 15 skills in v7.0.2 fall into four groups:
| Group | Skills |
|---|---|
| Planning | brainstorming, writing-plans, executing-plans, subagent-driven-development, dispatching-parallel-agents |
| Quality | test-driven-development, systematic-debugging, verification-before-completion |
| Review and branches | requesting-code-review, receiving-code-review, using-git-worktrees, finishing-a-development-branch |
| Meta | using-superpowers, writing-skills, diagnosing-superpowers |
How to install Superpowers in Claude Code
There are two routes, and both install the same code.
- Anthropic's official marketplace. Run
/plugin install superpowers@claude-plugins-officialinside Claude Code. The official marketplace file pins Superpowers to commit22f0b46, which is the v7.0.2 release, and the plugin listing on claude.com describes it. - The author's marketplace. Run
/plugin marketplace add obra/superpowers-marketplace, then/plugin install superpowers@superpowers-marketplace. This marketplace also carries related plugins from the same author, such as a Chrome DevTools browsing skill and an Elements of Style writing skill.
Restart or start a new session after installing so the hook runs. The README's quick check is to ask for something like "Let's make a react todo list": a working install starts the brainstorming skill before any code gets written.
One detail from our test: the claude.com listing still describes slash commands such as /execute-plan and a separate code reviewer agent. The v7.0.2 package we installed had 15 skills, one hook and zero agents according to claude plugin details, so trust the repository README over the listing copy when they disagree. If you want to know what else a plugin can do inside Claude Code, our Claude Code mods guide explains how plugins, hooks, skills and mods differ.
Installing Superpowers in other coding agents
Each harness needs its own install. The commands below come from the project README as of v7.0.2.
| Agent | Install |
|---|---|
| OpenAI Codex app or CLI | Listed in the official Codex plugin marketplace; open Plugins (or /plugins in the CLI) and search for Superpowers |
| Cursor | /add-plugin superpowers in Agent chat, or search the plugin marketplace |
| GitHub Copilot CLI | copilot plugin marketplace add obra/superpowers-marketplace, then copilot plugin install superpowers@superpowers-marketplace |
| Gemini CLI | gemini extensions install https://github.com/obra/superpowers |
| Qwen Code | qwen extensions install obra/superpowers |
| OpenCode | Ask OpenCode to fetch and follow the repo's .opencode/INSTALL.md |
| Antigravity | agy plugin install https://github.com/obra/superpowers |
| Pi | pi install git:github.com/obra/superpowers |
| Hermes Agent | hermes plugins install obra/superpowers --enable |
The README also covers Devin CLI, Factory Droid, Grok Build CLI, Kimi Code and Muse. Two caveats are worth knowing. Antigravity has no session start hook, so Superpowers relies on the model reading the using-superpowers description and loading it. Hermes has no hook after compaction, so a very long Hermes session can lose the bootstrap; the README's fix is to start a fresh session.
The Superpowers workflow, step by step
The README describes seven stages, each one triggered by a skill:
- Brainstorming. Before building anything, the agent asks what you want and why. In v7.0.2 it asks a question only when the answer would change what gets built, and offers options with a recommendation when there is a real choice. Larger projects get a written design saved to
docs/superpowers/specs/and committed. - Git worktree. After you approve the design, it creates an isolated workspace on a new branch, by default under
.worktrees/in the project, and checks that the tests pass before any change. - Writing plans. It breaks the work into tasks of two to five minutes each, with exact file paths, code and a way to verify each one. Plans go to
docs/superpowers/plans/. - Execution. You choose subagent driven development, where a fresh subagent handles each task and a review follows each one, or executing plans, which works through every task in the current session with one review of the whole branch at the end. The README calls the first the most thorough and the second the cheapest.
- Test driven development. Red, green, refactor. The skill is strict: if code was written before its test, the instruction is to delete it and start over.
- Code review between tasks. Issues are reported by severity, and critical ones block progress. That review happens inside your session; for a second pass on the pull request itself, see our Claude Code Review setup guide.
- Finishing the branch. It verifies tests, then offers to merge, open a pull request, keep the branch or discard it, and cleans up the worktree.
Two side skills matter in daily use. systematic-debugging forces a root cause investigation before any fix, and after three failed fixes it tells the agent to question the architecture instead of trying a fourth. diagnosing-superpowers reads a session transcript when something went wrong and can package a scrubbed bundle for a bug report.
If you already run several agents at once, this pairs naturally with Claude Code background agents, which also give each session its own worktree.
What Superpowers costs in tokens
The plugin is free, but it is not free to run. claude plugin details superpowers on our install reported:
- Always on: about 706 tokens added to every session, which is the skill descriptions the agent needs to decide when to load each one.
- On invoke: from about 700 tokens (
requesting-code-review) to about 4,900 tokens (subagent-driven-development) each time a skill fires. Brainstorming was about 3,700 and executing plans about 4,500.
The same report lists the hook as having no model context cost, because it counts the hook command itself. The hook's output is a different matter. When we ran it by hand it printed 3,405 characters (about 520 words), the full using-superpowers skill wrapped in a short header, and that text is added to the session at startup, /clear and compaction. Budget for it on top of the always on figure. The bigger cost is behavioral: subagent driven development runs a fresh agent per task plus reviews, so a ten task plan means many more model calls than one agent editing files directly.
Privacy, safety and what we tested
Superpowers is mostly Markdown, which makes it easy to audit. Read the hooks/ folder and the scripts under each skill before you install, as you would for any plugin. The one network call the README discloses is in brainstorming's optional visual companion: its Prime Radiant logo loads from the company's site and includes the Superpowers version, but not your project or prompts. Set SUPERPOWERS_DISABLE_TELEMETRY to turn it off; it also honors Claude Code's DISABLE_TELEMETRY and CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC.
What we checked on Linux with no model involved:
claude plugin validatepassed, with one warning that the plugin has no icon.- The repo's own tests for the session start hook, skill layout and skill size all passed. Every
SKILL.mdis under 20,000 bytes. - The hook emits
additional_contextwhen run under Cursor and the nestedhookSpecificOutputfield under Claude Code, matching what each harness expects. - The
sdd-workspacehelper refused to run outside a git repository (exit code 2, with a message asking forgit initand a first commit). Inside a repo it created.superpowers/sdd/<plan name>/and a.gitignorethat keeps those working files out of commits.
We did not measure whether the skills make a model write better code; that depends on the model and the task. The maintainers publish their own numbers in the release notes. For v7.0.2 they report that in blind graded test conversations, filler questions from brainstorming fell from 41 to 23 compared with 7.0.1. Treat that as the project's internal testing, not an independent benchmark.
Is Superpowers worth it?
Use it when the work has a real design question and a test suite the agent can run: a new feature, a refactor across several files, or a long unattended session where drifting from the plan is the main risk. It is a good fit if you already agree with test first development, since the skills enforce it hard.
Skip it, or turn it off for the session, for quick edits, exploratory prototypes where tests would be thrown away, and repos with no tests at all, where the verify steps have little to check. Our guide to what agentic coding is explains why weak checks hurt any agent, with or without Superpowers. If you mainly want rules the agent always follows, a short instruction file is lighter; see AGENTS.md explained.
FAQ
Is Superpowers free?
Yes. The code is MIT licensed and installs at no cost from Anthropic's marketplace, the author's marketplace or the GitHub repo. You still pay for the model tokens your agent uses, and Superpowers adds some. Prime Radiant offers paid commercial support for companies, which is optional.
Does Superpowers work with Codex and Cursor?
Yes. It is in the official Codex plugin marketplace for the Codex app and CLI, and Cursor installs it with /add-plugin superpowers. Each agent needs its own install, and the same skills load in each.
How do I update or remove Superpowers in Claude Code?
Use /plugin inside Claude Code, or claude plugin update superpowers and claude plugin uninstall superpowers from the terminal. Claude Code notes that a restart is needed after an update. The README says updates are often automatic, depending on the agent.
Superpowers vs Spec Kit: what is the difference?
Both push an agent to specify before it codes. GitHub's Spec Kit is a CLI you install with uv tool install specify-cli, and you drive it with explicit commands such as /speckit-specify, /speckit-plan, /speckit-tasks and /speckit-implement, keeping the spec files as project artifacts. Superpowers triggers its skills automatically and goes further into execution, with worktrees, test first rules, subagent reviews and branch cleanup. Pick Spec Kit if you want a spec process you call by hand, and Superpowers if you want the agent to follow one without being asked.
Can I use only some of the skills?
Mostly. Skills load on demand, so the ones you never trigger cost only their short description. Some agents can filter what loads; for Pi, v7.0.2 fixed the package skills filter so it is respected. The bootstrap is what makes the process mandatory, so removing it turns Superpowers back into an optional skill library.
Can I add my own skills to Superpowers?
You can write your own skills next to it; the writing-skills skill teaches the format and how to test a skill. The project itself says it generally does not accept new skills as contributions, so keep custom skills in your own plugin or repo.


