skip to content
文章 · 2026年6月 Posts · June 2026

superpowers

作者 By Yc Lin·2026年6月27日27 June 2026·9 min read

ai· github

obra/superpowers

Last Updated: 2026-06-27 ⭐ Stars: 240,000 🍴 Forks: 21,300 🔗 GitHub: https://github.com/obra/superpowers

Overview

In the field of software engineering, large language models have been widely used to assist in programming. However, traditional AI assistants tend to jump straight into code generation without holistically analyzing the architecture, often leading to a lack of test coverage and chaotic engineering logic. To resolve the lack of systematic methodology, blind refactoring, and low delivery quality associated with autonomous code generation by AI, obra/superpowers was created as an agentic skills framework and software development methodology for coding agents.

Based on the core philosophy of “process over guessing,” the project builds a structured ecosystem around a set of composable skills. It constrains AI behavior by enforcing mandatory workflows: initiating a Socratic brainstorming session to refine requirements before writing code, automatically creating an isolated Git worktree after design approval, and generating highly granular, step-by-step implementation plans. During the implementation phase, the framework mandates a strict Test-Driven Development (TDD) cycle following the RED-GREEN-REFACTOR workflow, even going so far as to delete unverified code written before its corresponding tests.

Ultimately, this project injects standardized software engineering capabilities into various AI programming clients (such as Claude Code, Gemini CLI, GitHub Copilot CLI, etc.). It breaks down complex development tasks into subagent-driven workflows, utilizing multi-stage specifications and code quality reviews to guarantee delivery compliance. It is highly applicable to automated software collaboration scenarios that demand higher autonomy from AI coding agents, robust test coverage in large codebases, and a reduction in “agentic slop.”

Tech Stack

  • Shell Script (51.6%): Used for writing the foundational lifecycle hooks, automating environment initialization, managing Git worktrees, and maintaining cross-platform toolchain adaptation scripts.

  • JavaScript / TypeScript (43.9%): Used to implement the main plugin logic, build extensions that interface with various AI coding harnesses (such as Claude Code, Codex, Kimi, etc.), and develop test runners.

  • HTML (2.2%): Used for rendering the optional visual companion feature and local visual telemetry interfaces.

  • Python (1.9%): Used for auxiliary scripts, automation workflow invocations, and integration evaluation tests.

Applications / Impact

  • Application Areas: Ideal for autonomous agent-driven development, deploying automated TDD workflows within agile engineering teams, multi-agent collaborative programming, and auditing the quality of enterprise AI development processes.

  • Notable Organizations/Projects: The framework officially supports and deeply integrates with mainstream AI coding tools, including Anthropic’s Claude Code, Google’s Gemini CLI, GitHub Copilot CLI, Kimi Code, and OpenCode. Developed by Prime Radiant, the project has garnered substantial traction within the open-source community, accumulating 240k stars.

评论Comments