⚡ Superpowers：195k Stars 的 Agent 开发方法论，装上你的 Coding Agent 就直接开挂

# ⚡ Superpowers：195k Stars 的 Agent 开发方法论，装上你的 Coding Agent 就直接开挂 > 项目地址：[obra/superpowers](https://github.com/obra/superpowers) | ⭐ 195,070 Stars | 🛠 Shell | 作者：Jesse Vincent (Prime Radiant) 老实说，我见过太多 Coding Agent 项目了——功能花里胡哨，但实际用起来该踩的坑一个没少。Agent 整天瞎猜我的意图，写出来的代码又臭又长，测试？不存在的。 Superpowers 不一样。它不是另一个 Agent 框架，而是一套**完整的软件开发方法论**，以 Skill 形式注入你的 Coding Agent 大脑。装上之后，Agent 不再一上来就写代码，而是先做设计、写计划、TDD 循环、代码审查——全程自动化。 ## 装一个，所有 Agent 都能用 Superpowers 覆盖了主流 Coding Agent 平台，安装方式各不相同。 **Claude Code（官方市场）：** **Codex CLI：** **Gemini CLI：** **Cursor：** **GitHub Copilot CLI：** ## 它的工作流有多猛？ 装上 Superpowers 后，Agent 的干活流程完全变了： 1. **brainstorming** — 先跟你讨论需求，不做代码空想，输出完整设计文档 2. **using-git-worktrees** — 设计确定后自动创建独立分支，保持主分支干净 3. **writing-plans** — 把任务拆成 2-5 分钟能搞定的子任务，每个都有文件路径、代码、验证步骤 4. **subagent-driven-development** — 每个子任务派一个子 Agent 独立执行，两级审查（先查设计合规，再查代码质量） 5. **test-driven-development** — 强制 RED-GREEN-REFACTOR 循环，先写测试再写代码 6. **requesting-code-review** — 任务之间自动审查，关键问题阻断进度 7. **finishing-a-development-branch** — 全部完成后验证测试，提供 merge/PR/保留/丢弃选项 最骚的是，这些步骤全是**自动触发**的。Agent 在执行任何任务前都会检查相关的 Skill，不是"建议"而是"强制流程"。 ## 一句话总结 - **195k Stars**，Jesse Vincent 出品，Prime Radiant 维护 - 不是框架，是**方法论 + Skill 合集**，注入 Agent 后改变工作方式 - 支持 Claude Code / Codex CLI / Gemini CLI / Cursor / Copilot CLI 等主流平台 - 核心：设计先行 → 计划拆解 → TDD 循环 → 代码审查 → 干净收尾 装上它，你的 Agent 就不再是瞎写的脚本小子，而是一个有纪律的 Senior Engineer。

# ⚡ Superpowers: 195k Stars Agentic Skill Framework That Turns Your Coding Agent Into a Senior Engineer > Repo: [obra/superpowers](https://github.com/obra/superpowers) | ⭐ 195,070 Stars | 🛠 Shell | Author: Jesse Vincent (Prime Radiant) I've seen plenty of coding agent projects — fancy features, but when you actually use them, the same old problems: agents guessing your intent, writing bloated code, skipping tests entirely. Superpowers is different. It's not another agent framework. It's a **complete software development methodology** delivered as composable skills that inject directly into your coding agent's brain. Once installed, your agent stops blindly writing code and starts following a disciplined process: design → plan → TDD → review → clean finish. ## Install Across All Platforms **Claude Code (Official Marketplace):** **Codex CLI:** **Gemini CLI:** **Cursor:** **GitHub Copilot CLI:** ## The Workflow Once Superpowers is active, your agent follows this process automatically: 1. **brainstorming** — Discusses requirements, explores alternatives, saves a design document 2. **using-git-worktrees** — Creates isolated workspace on new branch post-design 3. **writing-plans** — Breaks work into 2-5 minute tasks with exact file paths and verification 4. **subagent-driven-development** — Dispatches subagents per task with two-stage review (spec compliance then code quality) 5. **test-driven-development** — Enforces RED-GREEN-REFACTOR: write failing test → minimal code → pass → commit 6. **requesting-code-review** — Auto-review between tasks, critical issues block progress 7. **finishing-a-development-branch** — Verifies tests, offers merge/PR/keep/discard options The kicker: **all steps are mandatory, not suggestions.** The agent checks for relevant skills before every task. ## Bottom Line - **195k Stars**, by Jesse Vincent (Prime Radiant) - Not a framework — a **methodology + skill library** that changes how your agent works - Supports Claude Code / Codex CLI / Gemini CLI / Cursor / Copilot CLI - Core: design first → plan → TDD loop → code review → clean finish Install it and your agent stops being a script kiddie and becomes a disciplined senior engineer.