A methodology, delivered as skills that fire on their own
Superpowers is not a grab-bag of prompts. It is a complete software development methodology for coding agents, built on composable skills plus the initial instructions that make sure the agent actually uses them. The distinction matters: most skills repos add capabilities, while Superpowers changes the agent’s process from the first message. That is the reason it sits at the top of this category rather than beside it.
The loop it imposes is the whole product. When you start building, the agent does not jump to code. It steps back and asks what you are really trying to do, teases a spec out of the conversation, and shows it to you in chunks short enough to read. After you sign off, it writes an implementation plan aimed at an enthusiastic junior with no context and an aversion to testing, leaning on red/green TDD, YAGNI, and DRY. Then, on your go, it runs a subagent-driven development process, working through tasks while inspecting and reviewing its own output, sometimes autonomously for hours without drifting from the plan. Because the skills trigger automatically, you do not invoke anything; the agent just behaves this way.
Install
Installation is per harness, and you install separately for each agent you use. For Claude Code, it is on the official plugin marketplace:
/plugin install superpowers@claude-plugins-officialOr via the project’s own marketplace, which carries Superpowers plus related plugins:
/plugin marketplace add obra/superpowers-marketplace
/plugin install superpowers@superpowers-marketplaceThe README documents setup for Codex CLI, Codex App, Factory Droid, Gemini CLI, OpenCode, Cursor, and GitHub Copilot CLI as well, so the methodology is not tied to one agent.
Why it can run unattended
The part that makes the autonomy claim plausible is the review loop, not raw model speed. Once you approve the plan, Superpowers does not hand the whole job to one agent in one long context. It runs subagent-driven development, dispatching each engineering task to a worker, then inspecting and reviewing that work before moving on. That inner check is what keeps a multi-hour run from drifting: a mistake gets caught at the task boundary rather than compounding silently across the session. The plan you signed off on stays the contract, and the red/green TDD discipline gives each step a concrete pass/fail rather than a vibe. It is the combination of an agreed plan, isolated tasks, and per-task review that lets the agent work for a stretch without you babysitting every edit.
What you are actually adopting
The thing to be clear-eyed about is that Superpowers is opinionated by design. It will make your agent slow down, interrogate the spec, and insist on tests before code. If you want a quick one-off script, that ceremony is overhead. If you are building something real and have watched agents charge ahead on a wrong assumption, the ceremony is the point: it front-loads the alignment and testing that prevent the expensive rework later. Adopt it when you want discipline imposed, not when you want a fast hack.
It is MIT-licensed and actively developed, with 275 open issues as of 2026-06 against a very large star count, so the core is stable while the long tail of harness-specific behavior keeps moving.
Superpowers versus other agent-skill repos
| superpowers | anthropics/skills | ECC | |
|---|---|---|---|
| Stars | 222,999 | 148,807 | 212,256 |
| Kind | methodology and framework | reference skill collection | operator system |
| What it changes | the agent’s whole process | adds specific capabilities | configs, memory, security |
| Triggers | automatic | per-skill | per-setup |
Counts are from GitHub as of June 2026. anthropics/skills is the official reference collection: individual capabilities you pick from, not a process. ECC is an operator system focused on configs, memory, and security across harnesses. Superpowers is the one that reshapes how the agent works step to step, from brainstorming to TDD. They are complementary layers rather than substitutes: a methodology, a capability library, and an operations layer.
Related
For the official capability collection, see anthropics/skills; for one engineer’s personal, lighter-weight kit, see mattpocock/skills. Manage these across tools with cc-switch. For what else is climbing, see LLM tooling, the daily digest, and the weekly report.
FAQ
Is Superpowers a set of prompts? No. It is a development methodology built on skills that trigger automatically, so the agent runs a brainstorm to spec to plan to TDD loop without you invoking anything.
How do I install it? Per harness. For Claude Code, /plugin install superpowers@claude-plugins-official, or add obra/superpowers-marketplace. Other agents are documented separately.
Will it slow my agent down? Deliberately, for small tasks. It front-loads spec alignment and testing, which is overhead for a quick script and the whole value for real work.
Does it work outside Claude Code? Yes. The README covers Codex, Gemini CLI, Cursor, OpenCode, Factory Droid, and Copilot CLI, installed per harness.