obra superpowers
COMMUNITYLABSCO SUMMARY
The 14 chain into one pipeline: brainstorming turns a rough idea into a written design, writing-plans breaks the design into small tasks, using-git-worktrees isolates the work on its own branch, and then subagent-driven-development or executing-plans runs the tasks under red-green test-driven-development with a two-stage review (requesting-code-review, receiving-code-review) before finishing-a-development-branch decides whether to merge. Two skills sit outside that chain: writing-skills, for authoring new skills in this same format, and dispatching-parallel-agents, for running independent tasks concurrently. Nine of the 14 need nothing beyond the agent itself; the other five assume a local tool already on the machine, git chief among them for the worktree skill.
This suits someone who wants their agent to default to a disciplined, plan-then-test-then-review process and is willing to let it run unattended for a couple of hours against a plan they signed off on first. It does not suit someone who wants to borrow one skill, like test-driven-development, on its own: using-superpowers is written to load automatically and to require a skill check before the agent answers anything, including a clarifying question.
READ THE FULL ANALYSIS
Ten install paths, one skill set. The README gives separate install steps for Claude Code (two competing marketplaces), Antigravity, Codex App, Codex CLI, Cursor, Factory Droid, GitHub Copilot CLI, Kimi Code, OpenCode, and Pi, and says plainly that using more than one of these means installing Superpowers again in each — there is no single install that covers every agent.
One skill phones home by default. The brainstorming skill's optional visual companion loads a logo from the maintainer's own site and reports which Superpowers version is running. The maintainers say it carries no project, prompt, or click data; it is turned off by setting the SUPERPOWERS_DISABLE_TELEMETRY environment variable, or by either of Claude Code's own telemetry opt-outs.
Who wrote it, and who can change it. Superpowers is maintained by Jesse Vincent and Prime Radiant, not Anthropic, and the README says the project does not generally accept new skills as contributions — only fixes to the existing 14, tested across every agent it supports.
ALSO IN THIS PACKAGE
superpowers
Core skills library for Claude Code: TDD, debugging, collaboration patterns, and proven techniques
analyst common
cost and time
plan adherence
quality evidence
repeated work
request conflicts
scrub audit
scrub
similar session
skill timeline
stumbles
WHAT'S INSIDE
14 showing · 14 totalNothing else to set up — install it and go.
brainstorming
A rule against jumping straight into code — the assistant has to talk the idea through with you and get a clear yes on what it intends to build, however small the job looks.
writing-plans
Turns an agreed design into a written, step-by-step plan detailed enough that someone who has never seen the project could follow it without asking questions.
finishing-a-development-branch
The last step of a piece of work: run the whole test suite, and only once it is green, offer you the choice of folding the work into the main code, opening it up for review, or leaving it where it is.
receiving-code-review
Someone criticises the code and the easy reply is "you're absolutely right". This forbids that: the suggestion gets checked against the real codebase first, and argued with when it is wrong.
requesting-code-review
Hands the finished code to a second, separate assistant to go over with fresh eyes, and brings back what it found sorted into must-fix, should-fix and minor.
verification-before-completion
Nothing gets called done until it has been shown to work: the command is run, the output is read, and only then does the claim get made.
using-superpowers
Before answering anything at all — even a clarifying question — the assistant is made to check whether one of these skills covers the situation, and to follow it when one does.
writing-skills
How to write one of these skill documents so that it actually changes what an assistant does: watch it get the job wrong first, write the instruction, then put it through the same job again.
dispatching-parallel-agents
When several unrelated things are broken at once, this hands each one to its own helper to work on at the same time, then checks that the fixes don't collide.
executing-plans
Takes a plan that was already written down and works through it in order, proving each step really passes before starting the next, and stopping to ask instead of guessing when something does not fit.
subagent-driven-development
Works through a written plan one task at a time, handing each task to a separate assistant that starts with a clean slate, and reviewing the result before moving on.
systematic-debugging
Bans the guessing that usually follows a bug — the assistant has to work out why it is happening, and show its evidence, before it is allowed to change anything.
test-driven-development
Write the test before the code, and watch it fail first — because a test you never saw fail might not be testing anything.
using-git-worktrees
A separate copy of the project to work in, so nothing that happens next can disturb what you have open — plus a check for whether one already exists before another gets made.
HOW TO GET IT
npx skills add obra/superpowersnpx skills add obra/superpowers --skill <name> --full-depthPick the skill name from the Skills tab — each entry there installs independently.
/plugin marketplace add obra/superpowers/plugin install superpowers@superpowers-devTyped inside the agent's own prompt, not in a terminal. The marketplace is called superpowers-dev, which is the part after the @.