AgentConn
O

obra/superpowers

Coding Free

About obra/superpowers

obra/superpowers is the framework-and-methodology layer of the agent-skills movement. While [mattpocock/skills](/agents/mattpocock-skills) is one engineer's personal working set and ComposioHQ/awesome-codex-skills is a community list, superpowers ships as a complete framework: a skills directory plus the methodology for using it. The repo's tagline — 'agentic skills framework & software development methodology that works' — is load-bearing. It is the highest-base agent-skills repo in GitHub trending with 174K total stars (well above the rest of the agent-skills cluster), and held #4 on trending for two consecutive days in late April 2026 with steady +1,600/day growth. The methodology layer addresses what most personal-skills repos leave implicit: how to choose which problems should become skills vs stay in code, how to chain skills together, how to evaluate a skill's effectiveness over time, and how to govern a growing skills library across a team. For teams adopting agent-skills as a production primitive, superpowers is the closest thing to a canonical reference framework available in 2026.

Key Features

  • Agentic skills framework — directory + methodology, not just a skills list
  • 174K total GitHub stars — highest-base agent-skills repo in trending
  • Methodology for choosing which problems become skills vs stay in code
  • Skill chaining and composition patterns documented
  • Team governance guidance for growing skills libraries
  • Compatible with Claude Code, Cursor, OpenCode — harness-agnostic
  • Ships with reference skills you can fork and customize

Overview

obra/superpowers ships the framework layer that personal-skills repos leave implicit: how to choose which engineering problems should become skills, how to compose skills into workflows, and how to govern a skills library as it grows.

The repo combines a working skills directory with the methodology behind it. For teams adopting agent skills as a production primitive — not just as a personal productivity hack — superpowers is the closest thing to a canonical reference framework available in 2026.

Why It Matters

The skills-as-runtime thesis only works at scale if there’s a discipline for managing skills like any other production code. obra/superpowers makes that discipline explicit:

  • Composition patterns. How to chain skills so the model can combine them by scenario, rather than re-implementing the workflow each time.
  • Choice criteria. Which features port well to skills (workflow-shaped, model-driven branching) vs which stay in code (latency-critical, deterministic security boundaries).
  • Skill evaluation. How to measure whether a skill is actually helping vs adding context noise — eval patterns that work without a heavy ML pipeline.
  • Governance. How a team grows a skills library without ending up with 200 overlapping primitives.

Pairs naturally with mattpocock/skills (reference implementation), warp (runtime), and cc-switch (cross-runtime portability).

Similar Agents