Superpowers

by obra

Community Developer Tools 269k likes

A Claude Code plugin that replaces ad hoc prompting with a structured workflow — brainstorm, plan, build with TDD, review — built from composable, auto-updating skills.

Install in Claude Code

/plugin install superpowers@claude-plugins-official

Source: https://github.com/obra/superpowers

About

Superpowers isn't a set of callable MCP tools so much as an installed methodology: it bundles 16+ Markdown-defined skills — brainstorming, writing-plans, test-driven-development, systematic-debugging, requesting-code-review, using-git-worktrees, and more — that your agent picks up automatically once the plugin is active. Each skill surfaces as a capability the agent reasons about mid-task, nudging it through design discussion, a written implementation plan, disciplined test-first coding, and a review checkpoint before anything gets merged.

Built by Jesse Vincent (obra), it runs across more than just Claude Code — Cursor, Codex, Gemini CLI, GitHub Copilot CLI, OpenCode, and several others each have their own install path for the same skill set. That breadth is the point: the workflow lives in the skills themselves rather than any one vendor's plugin format, so switching agents doesn't mean re-teaching your process.

Key features

  • 16+ composable skills covering the full dev lifecycle: brainstorming, planning, TDD, debugging, code review
  • Guides agents through a design → plan → implement → review → merge sequence instead of freeform prompting
  • Ships as native plugins for Claude Code, Cursor, Codex, Gemini CLI, GitHub Copilot CLI, and more
  • Skills auto-update from the upstream repo, improving the methodology without a manual reinstall
  • Includes infrastructure skills for git worktrees and dispatching parallel subagents on independent tasks

Use cases

  • Keeping a coding agent from jumping straight to code before a plan and test cases exist
  • Running multiple related tasks in parallel subagents without losing track of review checkpoints
  • Standardizing how a team's agents debug, since systematic-debugging enforces a repeatable process
  • Adopting one workflow across several different AI coding tools instead of separate prompts for each

Available tools

test-driven-development

Pushes the agent to write a failing test before writing the implementation that makes it pass.

brainstorming

Structures the design conversation before any code exists, surfacing tradeoffs and open questions.

writing-plans

Turns a design decision into a concrete, checkable implementation plan the agent executes step by step.

systematic-debugging

Walks the agent through a repeatable root-cause process instead of guessing at fixes.

requesting-code-review

Triggers a review checkpoint before a branch is considered done or merged.

Frequently asked questions

Does Superpowers expose MCP tools I can call directly?

Not in the classic sense — it's a skills plugin. Claude Code and compatible agents pick up its Markdown-defined skills automatically once installed, rather than you invoking named MCP functions.

Do I need Claude Code specifically to use it?

No. Superpowers ships install paths for Cursor, Codex, Gemini CLI, GitHub Copilot CLI, OpenCode, and a few other agent platforms, all built from the same skill set.