Superpowers

Pricing

Open source

Platforms

Claude Code, Codex, Cursor, Gemini CLI, OpenCode, Repo workflows

Website

https://github.com/obra/superpowers

Free access note

Superpowers is open source on GitHub and can be used as a skills layer for supported AI coding tools.

Caveat

It does not provide model quota by itself; you still need access to Claude Code, Codex, Cursor, Gemini CLI, OpenCode, or another supported agent.

My take

Superpowers is one of the more important Claude Code-adjacent projects because it does not try to be a better model. It tries to make the agent behave like a disciplined engineer. That is the right problem.

My take: most AI coding failures in 2026 are not caused by the model being unable to write code. They are caused by weak task framing, skipped design, no test-first loop, guess-driven debugging, missing verification, and large diffs that nobody wants to review. Superpowers addresses that layer directly.

What It Is

Superpowers is an open-source skills framework from obra/superpowers. The project positions itself as a way to give agents structured capabilities across tools such as Claude Code, Codex, Cursor, Gemini CLI, OpenCode, and other coding agents.

In Coding Agent Tools terms, Superpowers belongs between the specification layer and the agent execution layer. It is not a replacement for Claude Code or Codex. It is the set of reusable workflows that tells those agents how to approach work.

Why It Matters For Claude Code

Claude Code is already strong at reading a repository, editing files, running commands, and leaving a diff. The weak point is not raw action. The weak point is whether the action is staged, inspectable, and easy to review.

Superpowers improves the Claude Code workflow by making common engineering gates explicit:

• brainstorming before implementation, so vague feature requests become scoped decisions.
• writing-plans before edits, so a task has steps, boundaries, risks, and verification commands.
• test-driven-development when behavior can be captured before implementation.
• systematic-debugging when a failure needs root cause rather than another guess.
• verification-before-completion before the agent claims the task is finished.
• requesting-code-review and receiving-code-review so review becomes a workflow, not a vibe.
• using-git-worktrees when concurrent agent work should be isolated.

My practical view: if Claude Code is your execution engine, Superpowers is a process harness. It gives Claude Code a stronger default operating mode.

How It Relates To Codex

Codex now has its own Skills and Automations story, but the Superpowers idea still applies. Codex is strongest when it receives a task brief with acceptance criteria, proof commands, and review expectations. Superpowers-style workflows help produce that brief.

Use Superpowers with Codex when you want repeatable task contracts: problem statement, plan, tests, validation evidence, review summary, and branch handoff. That matters even more for cloud or connected-host Codex work because the human reviewer may not have watched every step.

Best Skills To Start With

Do not install every workflow and hope for discipline. Start with the failure mode you already see.

• If agents start coding too early, start with brainstorming and writing-plans.
• If agents patch symptoms, start with systematic-debugging.
• If agents say “done” without proof, start with verification-before-completion.
• If diffs are hard to merge, add using-git-worktrees and review-focused skills.
• If product intent keeps drifting, pair Superpowers with OpenSpec and the Skills Matrix.

When I Would Use It

I would use Superpowers for real product work, refactors, bug fixes, migrations, test recovery, frontend implementation, documentation systems, and any change where review cost matters.

I would not use the full workflow for a tiny copy edit. The point is not ceremony. The point is to force the agent to create evidence when the work is risky enough to need evidence.

Claude Code Setup Pattern

The safest setup pattern is simple:

1. Install or enable Superpowers from the GitHub project.
2. Keep project rules in CLAUDE.md or the equivalent agent memory.
3. Start new feature work with brainstorming or planning.
4. Require verification before completion.
5. Review the final diff as normal engineering work.

That creates a clean chain: project context, workflow skill, agent execution, proof command, human review.

Coding Agent Tools Verdict

Superpowers is a strong addition to Claude Code because it changes the agent’s behavior at the moments where AI coding usually fails. It is also useful beside Codex, Cursor, Gemini CLI, OpenCode, and opencode when you want skills to be portable across tools.

My verdict: use Superpowers when the output needs to be maintainable, reviewed, and explainable. Skip it only when the task is truly trivial.

Source Notes

• The official Superpowers repository is obra/superpowers.
• The project describes Superpowers as a skills framework for agentic coding tools, including Claude Code, Codex, Cursor, Gemini CLI, OpenCode, and other agents.
• The Claude Code page on Coding Agent Tools links here first so readers can understand the workflow layer before leaving for GitHub.