Superpowers — agentic skills framework & software development methodology
Complete software development methodology for coding agents, built on composable skills and initial instructions that make the agent stop, ask, spec, plan, and then execute.
Snapshot
- Stars: 201,206
- Forks: 17,923
- Open issues: 298
- Language: Shell
- License: MIT
- Latest release: v5.1.0 (2026-05-04)
- Supports: Claude Code, Codex CLI/App, Factory Droid, Gemini CLI, OpenCode, Cursor, GitHub Copilot CLI
What it does
- Pushes agents to turn vague ideas into a readable spec before coding.
- Reinforces red/green TDD, YAGNI, and DRY.
- Encourages subagent-driven development once the plan is approved.
- Uses the same skill pack across multiple harnesses, but installs separately per harness.
Why it matters
Superpowers is one of the clearest examples of the shift from “prompting” to agent operating procedures. The repo is less about a single tool and more about a repeatable methodology for coding agents.
Similar OSS
- 2026-04-19-everything-claude-code-ecc-guide — harness-native operator system with memory, security, and many agents
- 2026-03-21-gstack — Claude Code planning / review / QA / shipping system
- 2026-05-21-oh-my-agent — vendor-agnostic multi-agent harness
- 2026-05-21-ouroboros — spec-first Agent OS with replayable events
- 2026-05-21-agent-harness-landscape — comparative landscape note