Superpowers for Claude Code: review, how it works and whether it is worth it (2026)

What you get in one minute#

Superpowers is a free plugin that makes Claude Code work like a disciplined engineer (ask first, plan, write tests first, review), and it pays off for multi-file features but is overkill for small fixes.

Logo: the official Superpowers project (github.com/obra/superpowers)

Over 300,000 GitHub stars and about 1 million installs (October 2026), but opinions are split.

What it does#

You type your request as usual, and Superpowers changes what Claude does before it answers. "Add a settings page" now starts with questions and a design you approve, not code.

Added at version 7.0.1:

• 15 skills: brainstorming, writing-plans, using-git-worktrees, subagent-driven-development, executing-plans, test-driven-development, requesting-code-review, receiving-code-review, finishing-a-development-branch, systematic-debugging, verification-before-completion, dispatching-parallel-agents, diagnosing-superpowers, writing-skills and using-superpowers.
• 1 hook: SessionStart, on start, clear and compact.
• 0 slash commands, 0 named agents, 0 MCP servers (the marketplace page still lists old ones). Call a skill by hand with /superpowers:<skill>.

How it works in the background#

One hook puts the rulebook into every session, and that rulebook sends Claude from skill to skill in a fixed order. The picture follows one feature request.

• Steps 3 to 7: you approve the design and read the plan.
• Step 8: subagent-driven-development is the most thorough; executing-plans the cheapest (README).

It is all written instructions; nothing forces Claude to follow a skill.

Under the hood: dissection#

Superpowers is mostly Markdown instructions (81% of its size) with a few Bash and JavaScript helpers, and its only code that runs by itself is one start-up hook.

The file tree#

superpowers/
├── .claude-plugin/
│   ├── plugin.json            name, version 7.0.1, MIT, author
│   └── marketplace.json       lets you add the repo itself as a marketplace
├── hooks/
│   ├── hooks.json             one SessionStart hook
│   ├── run-hook.cmd           one file that runs on Windows and Unix
│   └── session-start          Bash: prints the context to inject
└── skills/                    15 skills, 84 files in total
    ├── using-superpowers/SKILL.md          the rulebook the hook injects
    ├── brainstorming/
    │   ├── SKILL.md
    │   ├── builder-check-prompt.md
    │   ├── visual-companion.md
    │   └── scripts/ (server.cjs, helper.js, start-server.sh, stop-server.sh, frame-template.html)
    ├── writing-plans/SKILL.md
    ├── using-git-worktrees/SKILL.md
    ├── subagent-driven-development/
    │   ├── SKILL.md, implementer-prompt.md, task-reviewer-prompt.md, re-review-prompt.md
    │   ├── references/ (fix-loop, final-review, preflight, ...)
    │   └── scripts/ (sdd-workspace, task-brief, review-package)
    ├── executing-plans/ (SKILL.md, scripts/task-start, scripts/task-done)
    ├── test-driven-development/ (SKILL.md, writing-good-tests.md)
    ├── requesting-code-review/ (SKILL.md, code-reviewer.md)
    ├── receiving-code-review/SKILL.md
    ├── finishing-a-development-branch/SKILL.md
    ├── systematic-debugging/ (SKILL.md, techniques .md, find-polluter.sh, one .ts example)
    ├── verification-before-completion/SKILL.md
    ├── dispatching-parallel-agents/SKILL.md
    ├── diagnosing-superpowers/ (SKILL.md, prompts/, templates/, references/)
    └── writing-skills/ (SKILL.md, references/, render-graphs.js)

What each file is#

Two key excerpts#

hooks/hooks.json:

"SessionStart": [{
  "matcher": "startup|clear|compact",
  "hooks": [{ "type": "command",
    "command": "\"${CLAUDE_PLUGIN_ROOT}/hooks/run-hook.cmd\" session-start",
    "shell": "bash", "async": false }]
}]

skills/using-superpowers/SKILL.md:

If you think there is even a 1% chance a skill might apply to what you are doing,
you ABSOLUTELY MUST invoke the skill.

Disk and network#

• Writes through Claude: docs/superpowers/ and .superpowers/sdd/, plus git branches and worktrees.
• Network: the hook makes none; the optional visual companion loads a logo with the version number (SUPERPOWERS_DISABLE_TELEMETRY turns it off).
• Needs: Bash (Git for Windows on Windows) and git.

Chart: file sizes from the GitHub tree at tag v7.0.1, October 2026. Only loaded skill bodies count against your context.

What users say#

47 user reports read from 20 pages: 5 praise, 5 mixed, 37 complaints, but that split is skewed. GitHub issues attract complaints, 14 of them from one version 7 thread. Reddit could not be read, so it is not counted.

Chart: our count of the 47 reports, October 2026. Not a satisfaction rate.

Chart: themes of the 37 complaints, labelled by DIY Lab Hub, October 2026.

Pros: why it is useful#

It gives Claude a repeatable engineering routine, and that is worth the most on work where mistakes are expensive.

• Fewer wrong builds: the design approval catches "built the wrong thing" before code exists.
• Tests by construction and a reviewable paper trail in your repository.
• Long unattended runs: the README says a couple of hours without leaving the plan.
• Easy to inspect: MIT licensed, text and short scripts. 46 versions in 13 months.

Chart: version headings with dates in the project's RELEASE-NOTES file at v7.0.1, read in October 2026.

Cons: why it is not useful#

It is a heavy process, and the plugin cannot tell when your task is small.

• Cost: a fresh subagent per task plus reviews multiplies model calls.
• Slow for small work; "skip planning" helps, but friction stays.
• Overlap with plan mode, subagents and worktrees, and behaviour changes between versions.

Should you consider it? When it is useful#

Consider it for multi-file features and unclear goals, skip it for small tasks, and test it on one real task first. Pin 6.4.2 if the version 7 questions annoy you.

Install and check it#

You install it from Anthropic's official marketplace with one command, inside Claude Code or from your shell.

Linux#

Inside Claude Code:

/plugin install superpowers@claude-plugins-official

Or from your shell, at user scope:

claude plugin install superpowers@claude-plugins-official

Windows (PowerShell)#

Install Git for Windows first (it provides the Bash the hook needs), then:

claude plugin install superpowers@claude-plugins-official

Without Bash the plugin still loads, but the start-up injection is silently skipped.

Mac (Terminal)#

Same commands as Linux:

claude plugin install superpowers@claude-plugins-official

To stay on version 6.4.2 (from the docs, not tested by us)#

The official entry is pinned; the repository is its own marketplace:

/plugin marketplace add obra/superpowers#v6.4.2
/plugin install superpowers@superpowers-dev

Do not keep both copies installed.

Check it worked#

claude plugin list
claude plugin details superpowers

The second shows 15 skills, 1 hook and the token cost. Then type Let's make a react todo list; brainstorming should start before any code.

Things to keep in mind#

• Security. A plugin runs code with your user rights. The hook is short and we read it fully; the risk is in the skills. Read hooks/hooks.json and hooks/session-start first.
• Updates can change behaviour, as version 7 did. Conflicts: overlaps with feature-dev and code-review; install one at a time.
• Your repository gets docs/superpowers/ and .superpowers/; work in git so you can undo.
• Organisations can limit marketplaces; if install is refused, ask yours.

Troubleshooting#

FAQ#

Is Superpowers made by Anthropic? No. It is a community project; Anthropic lists it in its marketplace, which is a catalog, not an endorsement.

Can I turn it off for one project? Yes. Install at project scope, or run claude plugin disable superpowers@claude-plugins-official.

How much does it cost? The plugin is free but uses more of your Claude usage than plain Claude Code on most tasks.

Sources#

• github.com/obra/superpowers: README, hooks/, skills/, .claude-plugin/ and RELEASE-NOTES at tag v7.0.1 (October 2026).
• Superpowers on the Claude marketplace: install count and listed components (October 2026).
• anthropics/claude-plugins-official: marketplace entry with its pinned commit (October 2026).
• Install and manage plugins, Measure plugin cost, Plugin security and trust and the plugin manifest reference: official Claude Code docs (October 2026).
• Maintainer announcement (October 2025).
• User reports: issue 2530, 750, 1152, 2612, 2551, 743, 2017, 895, 512, 275, 37, 2552, 2562, 2618 (2025 to 2026); Hacker News thread (April 2026); blog review (April 2026); self-run test (April 2026); DataCamp (July 2026); guide (October 2026); deep dive (February 2026).

Not run by DIY Lab Hub: we read the files and docs but did not install the plugin on a test machine.

Checked against the official docs on 2026-10-27