What it is
Superpowers is a set of agent skills plus a short bootstrap prompt that tells the agent to use them. When you ask for a feature, the agent first asks what you are trying to achieve, writes a spec you approve in chunks, turns it into a plan of 2 to 5 minute tasks, and then works through the plan with subagents, strict test-driven development and code review between tasks.
Jesse Vincent wrote it, and his company Prime Radiant maintains it and sells commercial support. It is listed in Anthropic's official Claude Code plugin marketplace and OpenAI's Codex plugin marketplace, and the README gives install steps for 16 harnesses, including Cursor, Gemini CLI, Copilot CLI, Devin CLI, Grok Build and OpenCode.
Our review
Superpowers is the skills plugin most worth installing for agent-driven feature work, because it changes what the agent does before it writes code rather than handing it more tools. It has about 304,000 GitHub stars, sits in both Anthropic's and OpenAI's official plugin marketplaces, and shipped v7.0.1 on 21 October 2026, eleven days after v7.0.0.
The value is the enforced sequence. Its 15 skills cover brainstorming, git worktrees, written plans, subagent-driven execution, red-green-refactor TDD (the README says code written before its test gets deleted), code review that blocks on critical issues, and a branch-finishing step. A SessionStart hook injects the bootstrap so the skills fire without being named, and the README claims agents commonly work unattended for a couple of hours without leaving the plan.
The cost is time and tokens, and the maintainers are still tuning both. Subagent-driven development dispatches a fresh subagent and a review for every task; executing-plans runs inline with one review at the end and is the cheaper path. The v6.4.2 notes admit that some frontier models tried to build the whole project while writing the plan, and that the fix cut planning time to a quarter and tokens to about a third. v7.0.0 then rebuilt brainstorming, drew an open issue titled "v7's brainstorm questions are rather annoying", and was retuned within a fortnight.
It is an opinionated project rather than a community library. The maintainers generally refuse new skills and reject agent-written pull requests nobody asked for. The optional visual companion loads a logo from Prime Radiant's site that reveals your version; SUPERPOWERS_DISABLE_TELEMETRY or Claude Code's own DISABLE_TELEMETRY turns it off.
Stacking it with another full workflow framework such as Everything Claude Code or BMAD gives the agent two competing planning regimes in one context. Matt Pocock's skills sit at the opposite end: a handful of pieces you choose, with no mandatory sequence at all.
Where it earns its place
A backend engineer handing Claude Code a multi-file feature in an existing service
Get a feature built from a loose ticket without supervising every step
Brainstorming reads the repo before asking anything, writing-plans splits the work into 2 to 5 minute tasks, and subagents implement each with a review after it. Expect token use to grow with the number of tasks; switch to executing-plans when one end-of-branch review is enough.
An engineering manager whose team uses Claude Code, Cursor and Codex side by side
Make every developer's agent follow the same test-first, review-before-merge process
Each harness needs its own install, but the skills are identical across them. The plugin is MIT and free, so the only spend is extra model usage; Prime Radiant offers paid support and managed spending for larger rollouts.
A developer chasing a flaky integration test
Find the root cause instead of patching the symptom
systematic-debugging runs a four-phase root-cause process and verification-before-completion makes the agent prove the fix. When a session goes wrong, diagnosing-superpowers reads the transcript and reports what happened with line-level evidence.
Before you commit
- Setup
- One command in Claude Code (/plugin install superpowers@claude-plugins-official); every other harness needs its own install. The README's smoke test is a fresh session asking for a React todo list: a working install starts brainstorming before any code.
- Integrations
- Documented for 16 harnesses, including Claude Code, Codex app and CLI, Cursor, Gemini CLI, Copilot CLI, OpenCode, Devin CLI, Kimi Code and Grok Build. Pi gets a small extension that re-injects the bootstrap after compaction; Hermes has no such hook and loses it in very long sessions.
- Data and privacy
- No server component: the skills are text files and local shell hooks. The only outbound call is the optional visual companion's logo, which carries your Superpowers version and can be disabled. GitHub lists no security advisories for the repo.
- Cost at scale
- Free under MIT. The real bill is model usage, which scales with the number of tasks in a plan under subagent-driven development rather than with the size of the change; commercial support is priced on request.
- Limits
- 352 open issues, many asking for more harnesses or reporting slower responses. Contributions must work across every supported agent, and maintainers rarely accept new skills, so gaps are filled on their schedule.
What's good
- Enforces spec, plan, TDD and review in a fixed order
- In Anthropic's and OpenAI's official plugin marketplaces
- Install steps for 16 agent harnesses
- Cheaper inline mode (executing-plans) when subagents are overkill
What's not
- Slow and token-heavy for small changes
- Default behaviour still shifts between releases, as v7.0.0's brainstorming rebuild showed
- New skill contributions are generally not accepted
Best for
Developers who want an agent to run multi-hour features with a plan, tests and review built in.
Alternatives
Superpowers: common questions
- Is Superpowers worth it?
- A discipline layer that makes a coding agent behave like a careful junior engineer, and worth its extra tokens on anything bigger than a one-line fix. We rate it 4.7 out of 5.
- How much does Superpowers cost?
- Free and open source (MIT). You pay for the model usage of the agent you run it in.
- What are the downsides of Superpowers?
- Slow and token-heavy for small changes; default behaviour still shifts between releases, as v7.0.0's brainstorming rebuild showed; new skill contributions are generally not accepted.
- Who is Superpowers best for?
- Developers who want an agent to run multi-hour features with a plan, tests and review built in.
- How much does Superpowers cost for a team or in production?
- Free under MIT. The real bill is model usage, which scales with the number of tasks in a plan under subagent-driven development rather than with the size of the change; commercial support is priced on request.
- Is Superpowers safe for company data?
- No server component: the skills are text files and local shell hooks. The only outbound call is the optional visual companion's logo, which carries your Superpowers version and can be disabled. GitHub lists no security advisories for the repo.
- What are the best Superpowers alternatives?
- The closest alternatives we have reviewed are Matt Pocock's Skills, Agent Skills (Addy Osmani), Everything Claude Code (ECC) and Spec Kit.
Answers checked Oct 23, 2026.
Make Superpowers? Show this rating on your site
