Superpowers plugin for Claude Code

• Asks what you want, then writes a design for you to approve

  Before any creative work, Claude asks questions about what you want and writes a design doc. You approve it before any code is written.

  Try: “Let's make a react todo list”

  Write · brainstorming

• Turns the approved design into small step-by-step tasks

  Breaks the design into short tasks with exact file paths, complete code and checks, so work can start without guessing.

  Try: “Write the implementation plan for this design”

  Write · writing-plans

• Creates a separate branch and workspace before feature work

  Gives the work its own git worktree so your main checkout stays untouched while Claude builds.

  Runs code · using-git-worktrees

• Builds the plan with a fresh subagent and review per task

  Hands each task to a new subagent and reviews it before the next. The most thorough option, and the slowest and most token-hungry.

  Try: “Execute this plan with subagent-driven-development”

  Runs code · subagent-driven-development

• Builds the plan in this session with one review at the end

  The cheaper way to run a plan: every task runs in the current session and you get one final review.

  Try: “Execute this plan with executing-plans”

  Runs code · executing-plans

• Writes a failing test first, then the code

  Enforces red, green, refactor. Claude must see a test fail before it writes the code, and code written before tests is deleted.

  Runs code · test-driven-development

• Finds the root cause before proposing any fix

  Runs a four-phase investigation of a bug or failing test before Claude suggests a change.

  Try: “Use systematic-debugging on this failing test”

  Runs code · systematic-debugging

• Makes Claude run checks before saying it is done

  Claude must run the verification commands and read their output before it claims work is finished.

  Runs code · verification-before-completion

• Reviews finished work against the requirements

  Checks a finished task or feature against the plan and requirements before merging.

  Read · requesting-code-review

• Checks review feedback instead of agreeing blindly

  When review comments arrive, Claude tests whether they are technically right before changing code.

  Read · receiving-code-review

• Splits independent tasks across parallel subagents

  When two or more tasks do not depend on each other, runs them side by side.

  Runs code · dispatching-parallel-agents

• Offers merge, PR, keep or discard when tests pass

  Once tests pass, gives you the choice to merge, open a pull request, keep or discard the branch, then cleans up the worktree.

  Runs code · finishing-a-development-branch

• Guide for creating and testing your own skills

  Helps you write, edit and test new skills of your own.

  Try: “Help me write a new skill”

  Write · writing-skills

• Explains from a transcript why Superpowers misbehaved

  Reads past session transcripts, reports what went wrong with evidence and can build a scrubbed bug report. It files a GitHub issue only after you approve the text.

  Try: “figure out what went wrong with superpowers in this session”

  Network · diagnosing-superpowers

• Used by the assistant behind the scenes

  Tells Claude to check for a relevant skill before replying

  Loaded at the start of every session. It makes Claude look for a matching skill before it answers.

  Read · using-superpowers

• Loads the using-superpowers skill into each new session

  Runs a bash script on startup, /clear and compaction. It reads the using-superpowers skill from disk and adds it to Claude's context. No network calls.

  Runs code · SessionStart hook

• Optional local web page for visual brainstorming

  A Node server on 127.0.0.1 at a random port that shows mockups while you brainstorm. It stops after 4 hours idle and loads its logo from primeradiant.com.

  Runs code · Visual companion server

Watch out for

• Copies named superpowers are not this plugin

  Many unrelated repos copy the name. The real one is obra/superpowers by Jesse Vincent, the source Anthropic's official marketplace points to. Install from the official marketplace and check the owner is obra. github.com

• That page sends the Superpowers version to primeradiant.com

  Its logo loads from primeradiant.com with the version number. The README says no project or prompt data is sent. Turn it off with SUPERPOWERS_DISABLE_TELEMETRY. github.com

• A startup hook runs a script in every new session

  One SessionStart hook (on startup, /clear and compaction) runs the plugin's session-start bash script. It reads the using-superpowers skill from disk and adds it to Claude's context. It makes no network calls. github.com

• Claude is told to write files, commit and merge

  The skills ask Claude to create worktrees and branches, edit code and commit. Keep Claude Code's permission prompts on and read what it proposes before merges. github.com

• Optional local web page for visual brainstorming

  If you accept it, a Node server starts on 127.0.0.1 at a random port with a key in the URL and stops after 4 hours idle. Do not bind it to a non-local address unless you need a remote setup. github.com

• The diagnosing skill reads your past sessions

  diagnosing-superpowers reads session transcripts to explain a misfire. It files a GitHub issue only after you approve the exact text, and bug bundles are scrubbed first. github.com

• Helper scripts run on your machine

  Some skills ship shell and Node scripts (for example find-polluter.sh) that Claude runs locally. They are plain files in the plugin folder you can read first. github.com

Security advisories

In their words