[
    {
        "role": "user",
        "content": "tell%20me%20about%20superpowers%20framework%2C%20possibly%20for%20claude%20or%20codex"
    },
    {
        "role": "assistant",
        "content": "Here\u2019s a detailed breakdown of the *Superpowers* framework \u2014 what it is, how it works (especially with Claude Code and Codex), its benefits & trade-offs, and who it\u2019s best suited for.\n\n---\n\n## What is Superpowers?\n\n- *Superpowers* is an open-source, agentic *skills framework* created by **Jesse Vincent** (of Prime Radiant) that layers strong software engineering discipline onto AI coding assistants. It was first released in **October 2025**. ([verdent.ai](https://www.verdent.ai/guides/what-is-superpowers-ai-coding-framework?utm_source=openai))  \n- Rather than a standalone tool, it\u2019s a plugin or instruction set (\u201cskills\u201d) that gets injected into platforms like **Claude Code, Cursor, Codex, Gemini CLI, GitHub Copilot CLI**, etc. ([kdjingpai.com](https://www.kdjingpai.com/en/superpowers/?utm_source=openai))  \n- The core idea is to enforce professional, battle-tested software development workflows: clarity, planning, test-first discipline, code review, debugging, and modular execution. ([verdent.ai](https://www.verdent.ai/guides/what-is-superpowers-ai-coding-framework?utm_source=openai))\n\n---\n\n## Key Components & Workflow\n\nSuperpowers introduces a structured, multi-phase workflow enforced via \u201cskills\u201d\u2014these are self-contained files (`SKILL.md`) defining when and how the AI should behave. ([kdjingpai.com](https://www.kdjingpai.com/en/superpowers/?utm_source=openai)) Here\u2019s how it typically flows:\n\n| Phase | What Happens |\n|---|---|\n| **Phase 0 / Master Hook**: \u201cusing-superpowers\u201d | At session start, Superpowers intercepts your prompt and decides which skills apply, routing automatically. ([popularaitools.ai](https://popularaitools.ai/blog/superpowers-plugin-10x-claude-code-2026?utm_source=openai)) |\n| **Phase 1: Brainstorm / Clarify** | Claude asks clarifying questions, refines requirements, and confirms design before starting any implementation. ([mymcpshelf.com](https://www.mymcpshelf.com/blog/superpowers-skills-framework-claude-code?utm_source=openai)) |\n| **Phase 2: Git Worktrees / Isolated Workspace** | Sets an isolated branch or worktree to prevent messing up existing code until plan is validated. ([mymcpshelf.com](https://www.mymcpshelf.com/blog/superpowers-skills-framework-claude-code?utm_source=openai)) |\n| **Phase 3: Planning** | Break task into small (2-5 minute) microtasks. Plan architecture, component responsibilities, file structure. Includes verification checkpoints. ([mymcpshelf.com](https://www.mymcpshelf.com/blog/superpowers-skills-framework-claude-code?utm_source=openai)) |\n| **Phase 4: Subagent-Driven Development / Execution with Review** | Each microtask is executed by a \u201csubagent.\u201d After implementation, there are reviews: one for spec compliance, another for code quality. Critical issues block progress. ([mymcpshelf.com](https://www.mymcpshelf.com/blog/superpowers-skills-framework-claude-code?utm_source=openai)) |\n| **Phase 5: Test-Driven Development (TDD) with RED-GREEN-REFACTOR** | Tests are written first (failing), then code to pass, then refactoring. Any code written before tests (except in certain utility cases) is deleted. ([mymcpshelf.com](https://www.mymcpshelf.com/blog/superpowers-skills-framework-claude-code?utm_source=openai)) |\n| **Other Skills & Meta-skills** | Skills for writing new skills (`writing-skills`), debugging (`systematic-debugging`), visual companions for design, etc. ([popularaitools.ai](https://popularaitools.ai/blog/superpowers-plugin-10x-claude-code-2026?utm_source=openai)) |\n\n---\n\n## How It Works with Claude Code & Codex\n\n- **Claude Code** is the primary target: everything (session hooks, subagents, skill activation) works most fully there. ([verdent.ai](https://www.verdent.ai/guides/what-is-superpowers-ai-coding-framework?utm_source=openai))  \n- With **Codex** (OpenAI\u2019s LLM), many of the core workflow and planning components work, though some of the subagent or parallelization aspects are more limited. Skills are portable via the same `SKILL.md` files. ([kdjingpai.com](https://www.kdjingpai.com/en/superpowers/?utm_source=openai))  \n- **Cursor**, **Gemini CLI**, **GitHub Copilot CLI**, etc., are also supported, though with feature-tradeoffs depending on how well they support multiple agents or isolated contexts. ([verdent.ai](https://www.verdent.ai/guides/what-is-superpowers-ai-coding-framework?utm_source=openai))  \n\n---\n\n## Why It Matters & What It Solves\n\nWithout Superpowers, common pain points include:\n\n- AI immediately jumping into writing code\u2014skipping requirements, testing, architecture decisions. ([mymcpshelf.com](https://www.mymcpshelf.com/blog/superpowers-skills-framework-claude-code?utm_source=openai))  \n- Inconsistent code quality, missing test coverage, duplicated code, mischosen libraries, messy file structures. ([kdjingpai.com](https://www.kdjingpai.com/en/superpowers/?utm_source=openai))  \n- When working collaboratively or over time, drift in style, architecture, or reliability. \n\nSuperpowers aims to:\n\n- Cultivate reliable, repeatable software engineering behavior from your coding agent.  \n- Reduce wasted effort by ensuring features are well-defined before work starts.  \n- Improve maintainability, test coverage, and correctness through enforced planning and reviews. ([popularaitools.ai](https://popularaitools.ai/blog/superpowers-plugin-10x-claude-code-2026?utm_source=openai))  \n\n---\n\n## Pros & Trade-Offs\n\n| Pros | Trade-Offs / Limitations |\n|---|---|\n| Much higher code quality and reliability. | Adds overhead\u2014startup time for planning and asking clarifying questions. Not ideal when speed matters or tasks are trivial. ([verdent.ai](https://www.verdent.ai/guides/what-is-superpowers-ai-coding-framework?utm_source=openai)) |\n| Enforced testing and review reduces bugs down the line. | Learning curve for users not used to disciplined workflows. May feel rigid at first. |\n| Better consistency across teams and large codebases. | Some platforms support fewer Superpowers features\u2014subagents, parallel tasks, etc.\u2014limiting what\u2019s possible. |\n| Measurable checkpoints & quality gates; stops bad assumptions early. | Might slow you down in exploratory work, prototyping, or very simple tasks. |\n\n---\n\n## Who It\u2019s For & When to Use It\n\n**Ideal for:**\n\n- Medium to large projects or features where architecture/design matters  \n- Teams wanting consistency, test coverage, and maintainability  \n- Client-facing or production systems where bugs are costly  \n\n**Less ideal for:**\n\n- One-off scripts, throwaways, hacking something up quickly  \n- Rapid experimentation or prototyping where iteration matters more than formality  \n- Minor bug fixes or tiny tasks where the workflow overhead may dominate time spent coding  \n\n---\n\n## Getting Started\n\n1. Install the plugin / skill set on your platform.  \n   - For Claude Code:  \n     ```\n     /plugin marketplace add obra/superpowers-marketplace  \n     /plugin install superpowers@superpowers-marketplace  \n     ```\n     ([mymcpshelf.com](https://www.mymcpshelf.com/blog/superpowers-skills-framework-claude-code?utm_source=openai))  \n2. Start a new coding session: Superpowers injects a session hook that activates the framework. ([mymcpshelf.com](https://www.mymcpshelf.com/blog/superpowers-skills-framework-claude-code?utm_source=openai))  \n3. Use slash-commands or prompts to invoke specific skills if needed (e.g. `/superpowers:brainstorming`). ([claudepluginhub.com](https://www.claudepluginhub.com/plugins/fradser-superpowers-superpowers?utm_source=openai))  \n4. Follow the phases: clarify \u2192 design \u2192 plan \u2192 execute with review \u2192 enforce TDD \u2192 debug systematically.  \n\n---\n\nIf you like, I can walk through a **live example**: what it looks like to build a small feature using Superpowers with Claude Code. Would that be helpful?"
    }
]