Superpowers for Claude Code: installing 7.0, how the workflow runs, and where the old slash commands went

Superpowers is an open-source skills library for coding agents, and a development workflow at the same time: before building anything it works out what you actually want, writes that up as a design and a plan, then builds it test-first with reviews along the way. The project is obra/superpowers on GitHub, by Jesse Vincent and the Prime Radiant team, under the MIT licence, with about 317k stars when checked on 2026-11-13. The latest release was 7.0.2, published on 2026-11-03. Besides Claude Code it supports Codex, Cursor, OpenCode, Copilot CLI and about a dozen other tools.

The most common problem with older tutorials: the /brainstorm, /write-plan and /execute-plan commands are gone. They were removed in 5.1.0 on 2026-04-30. You no longer need commands, because the skills trigger on their own based on what you're doing. To use one by hand, ask the agent to use that skill, for example superpowers:brainstorming, superpowers:writing-plans or superpowers:executing-plans. Version 7.0, released in October, also rebuilt the requirements conversation, so it no longer asks questions the way older videos show.

Install

Claude Code (official plugin marketplace recommended)

/plugin install superpowers@claude-plugins-official

You can also use the author's own marketplace, which carries a few related plugins:

/plugin marketplace add obra/superpowers-marketplace
/plugin install superpowers@superpowers-marketplace

Open a new session after installing. To check it works, send Let's make a react todo list. With a working install it starts brainstorming and asks about what you want instead of writing code straight away.

Other tools: install it separately in each one.

The workflow

Once it's installed, Claude Code checks for relevant skills before every task. The project describes these as mandatory workflows, not suggestions:

1. brainstorming: asks why you want the thing and what it should be, then plays its understanding back for you to correct. Small changes get agreed in chat; projects get a written design document. Since 7.0.2 it asks fewer questions: if an answer wouldn't change what gets built, it states its call for you to correct instead, and when there's a real choice it lists the options with a recommendation.
2. using-git-worktrees: after you approve the design, it creates an isolated git worktree on a new branch, runs project setup and checks the tests pass before any changes.
3. writing-plans: breaks the work into tasks of 2 to 5 minutes each, with exact file paths, code and verification steps.
4. Carrying out the plan, one of two ways (see the token section below):
   • subagent-driven-development: a fresh subagent for each task, with a review after each. The most thorough option.
   • executing-plans: every task done in the current session, with one review of the whole branch at the end. The cheapest option.
5. test-driven-development: write a failing test, watch it fail, write the least code that passes, commit. Code written before its test gets deleted, so expect that.
6. requesting-code-review: reviews against the plan between tasks and reports issues by severity. Critical issues block progress.
7. finishing-a-development-branch: runs the tests, offers to merge, open a PR, keep or discard the branch, then cleans up the worktree.

The author says it's not unusual for the agent to work on its own for a couple of hours without straying from the plan.

Not everything goes through the whole sequence. Since 7.0 it sizes the work first: it does a quick, clear task right away, agrees a small change in chat, and writes a design document only for a project. You can overrule its call.

Token use: the two execution modes differ a lot

Superpowers itself calls no API; it changes how Claude Code works. That way of working involves far more turns than writing code directly: discussion, a plan, tests, reviews, and possibly many subagents while carrying out the plan. On per-token billing, the biggest variable is which mode you pick in step 4:

• subagent-driven-development: a new subagent per task, plus a review per task. Each subagent re-reads the relevant files, so with many tasks token use is clearly higher. In return, every step gets checked.
• executing-plans: all tasks in one session with a single review at the end, which the project calls the cheapest option.

For a small feature that touches a few files, you can ask it to use executing-plans. Save the subagent mode for larger work where mistakes are expensive.

If a session burns more tokens than expected, repeats work or ignores its plan, ask it to "figure out what went wrong with superpowers in this session". It runs the diagnosing-superpowers skill, reads the session transcript and points to where things went wrong.

The reliable answer is your bill. With CCSub, the Usage page in the dashboard lists tokens and charge for every request; add up a time range to get the cost of one task.

Things to watch

• It creates branches and worktrees. using-git-worktrees opens a worktree on a new branch and asks what to do with it at the end. In a folder that isn't a git repo, run git init and make a first commit.
• Test-driven development deletes code. An implementation written before its test is deleted and rewritten. If you only want a small bug fixed, say so; it will treat it as a small task.
• Visual companion telemetry. The optional visual page in brainstorming loads a logo from the author's site, including the Superpowers version but nothing about your project. To turn it off, set SUPERPOWERS_DISABLE_TELEMETRY=1; it also honours Claude Code's DISABLE_TELEMETRY and CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC.
• Updates. Most tools update the plugin automatically. If its behaviour doesn't match what's described here, update to the latest version first.

Does it depend on your API provider?

No. Superpowers doesn't touch your API settings, and its subagents use whatever endpoint and key Claude Code already uses. It installs and behaves the same with an official account or a third-party service, with nothing extra to configure. The only thing to watch is the token use described above.

If you haven't set up Claude Code yet, see the Claude Code setup guide; for Codex, see the Codex CLI guide. claude-mem, another Claude Code plugin, does need an extra step with a third-party API; see the claude-mem guide.