superpowers is a plugin that turns a development process for coding agents into a set of skills. It has the agent clarify requirements before building, break the plan into pieces, and go through tests and review. Skills are pulled in automatically depending on the kind of work.
From actually using it
superpowers was installed when I reworked this whole site, and a session-start hook put the skill-usage rules into the conversation, so the brainstorming skill ran on the very first request. When the skill wanted to go through questions first, I said "Skip the questions and just go ahead," and it skipped the question stage and started the work. That confirmed first-hand that the skills give priority to the user's explicit instructions.
What the plugin does
The README describes it as a software development methodology for coding agents. It is made of composable skills, plus instructions that make the agent use those skills.
Before starting work, the agent checks whether a relevant skill exists. The README describes this as a required workflow, not a suggestion.
Installing it
In Claude Code, you install it by its name in the official marketplace. The README also describes adding the author's marketplace (obra/superpowers-marketplace) and installing from there.
/plugin install superpowers@claude-plugins-official
The basic flow in 7 steps
- brainstorming: before building, it asks what you want, sorts it out, and shows the design in small chunks.
- using-git-worktrees: after the design is approved, it creates an isolated branch workspace and checks that tests start out clean.
- writing-plans: it breaks the work into tasks of 2 to 5 minutes each and writes down file paths, code, and verification steps.
- subagent-driven-development or executing-plans: either a fresh subagent plus review for each task, or continuing in the current session with a review at the end.
- test-driven-development: it enforces the order of a failing test, a passing test, then cleanup.
- requesting-code-review: it reviews between tasks, and serious problems block progress.
- finishing-a-development-branch: it checks the tests, has you choose between merge, PR, keep, or discard, and then cleans up.
What it suits and what it does not
- Suits: work like a new feature where the requirements are still fuzzy, that touches several files, and that can be verified with tests.
- Does not suit: fixing a typo or changing one line of config. The cost of going through the process is bigger than the task. In that case, just tell it to go ahead.
- There are separate skills for debugging (systematic-debugging) and for before you declare something done (verification-before-completion).
- The skill library also includes dispatching-parallel-agents, receiving-code-review, writing-skills, and others.
Compared with other tools
The code-review plugin is a command that reviews a single PR, while superpowers is the whole procedure from planning to wrap-up. The feature-dev plugin in the official marketplace is also a development flow using exploration, design, and review agents, so its purpose is similar.
Real examples
You are starting a new feature with vague requirements
I want to add email verification after sign-up.
The brainstorming skill does not write code right away. It asks questions first, such as how long the verification link should be valid and whether resending is allowed. It shows the design in parts and asks for your approval, then moves on to writing the plan. Which questions actually come up differs from project to project.
You do not need the question stage
Skip the questions and just go ahead.
Skills give priority to the user's explicit instructions, so it skips the question stage and starts the work. You only need to say this once in the conversation.
Common mistakes and when not to use it
- Even a small fix gets design, planning, and review attached, which adds time. For a small task, say up front, "This is a simple fix, just do it."
- Because subagents run for each task, token usage can go up. Keep an eye on it with /usage.
- The README tells you to install the plugin separately for each tool that uses it. What you install in Claude Code does not automatically apply to other tools.
- The visual aid in brainstorming loads a logo and sends the superpowers version. To turn this off, set SUPERPOWERS_DISABLE_TELEMETRY to a true value (per the README).
- If a skill suddenly stops being invoked, the README suggests starting a new session. That note is under the Hermes agent entry, so I cannot say the same holds for Claude Code.
FAQ
Can I call a skill directly?
By default, a skill can be called by you with /name, and Claude can also load it automatically from its description. Plugin skills appear in the form /superpowers:skill-name.
Can I use it in other AI coding tools?
The README has install commands for several tools, including Codex, Cursor, Gemini CLI, and GitHub Copilot CLI. The install method differs by tool, and you install it separately for each.
Checked against the official docs on 2026-10-30. Sources: obra/superpowers README, Claude Code docs: Skills, Claude Code docs: Install and manage plugins. How we write