Superpowers is an open source agentic skills framework and software development methodology created by Jesse Vincent and the Prime Radiant team. Available on GitHub with more than 130,000 stars, it has established itself as one of the most widely adopted tools for structuring the work of AI coding agents.
The founding principle is simple: instead of letting the agent rush into code, Superpowers teaches it a disciplined workflow. As soon as a project is detected, the agent first engages in a brainstorming phase to understand the actual need. It then presents the specification in chunks short enough to be genuinely read and validated by the human. Once the design is approved, the agent builds an implementation plan before writing a single line of code.
The framework relies on SKILL.md files in markdown that serve as structured instructions. These skills cover five domains: testing (strict red/green TDD, asynchronous tests, anti-patterns), debugging (systematic approach, root cause tracing, verification), collaboration (brainstorming, planning, code review, parallel agents), development (git worktrees, branch management, subagent workflows), and meta-skills (creating, testing, and sharing new skills).
The architecture is remarkably token-efficient: the core injects fewer than 2,000 tokens at startup. Additional skills are loaded on demand via a shell script that searches for and reads them. This frugality is essential to avoid wasting the agent's context.
Superpowers also integrates a conversational memory system. It duplicates Claude conversation transcripts outside the .claude directory (to avoid their automatic deletion after one month), indexes them in a SQLite database with vector search, and generates summaries via Claude Haiku.
The project is compatible with several agent platforms: Claude Code via its native plugin system, Cursor, Codex, and OpenCode (with manual installation for the latter two). An ecosystem of satellite repositories complements the main project: a curated marketplace, a lab for experimental skills, a collection of community skills, and tools for developing Claude Code plugins.
The underlying methodological principles — strict TDD, YAGNI, DRY — reflect a vision in which software quality is not sacrificed to automation but instead reinforced by a structure that guides the agent toward best practices.