← Back to Discover

superpowers

An open-source skills framework and development method for coding agents, covering brainstorming, planning, test-driven implementation and code review, with install routes for many agent tools.

Agent frameworksCoding agents

What it does

Superpowers is a set of composable skills plus starter instructions that make a coding agent follow a fuller development process. The skills trigger automatically, so you do not have to invoke them. It is MIT licensed and published by Prime Radiant. The workflow runs in stages. Before any code is written, the agent asks what you are trying to achieve and shows a design in pieces short enough to read before you sign off. It then sets up an isolated git worktree, writes a plan of small tasks with exact file paths and verification steps, and either hands each task to a fresh subagent with a review after each, or works through them inline with one review at the end. Implementation follows red/green test-driven development, with review between tasks and a final step to finish the branch. The library also has skills for systematic debugging, verifying a fix before declaring it done, dispatching parallel agents, handling review feedback and writing new skills. Installation is separate for each agent tool, and the page lists routes for Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot CLI, OpenCode and others. Enterprise users can contact the team about commercial support. The page also discloses an optional logo load in the brainstorming visual companion that reports only the Superpowers version, which can be switched off with an environment variable.

Capabilities

  • Skills that trigger automatically inside the coding agent you already use
  • Design-first workflow: brainstorming, written plans, then implementation
  • Test-driven development and code review built into the process
  • Subagent-driven development with a review after each task
  • Install routes for Claude Code, Codex, Cursor, Gemini CLI and more

When to use it

  • You want a coding agent to plan and verify its work instead of jumping straight to code
  • You hand an agent long multi-step tasks and want review checkpoints along the way
  • You use several agent tools and want the same skills across them