[
    {
        "role": "user",
        "content": "tell%20me%20about%20superpowers%20framework%2C%20possibly%20for%20claude%20code%20or%20codex%20skill%3F"
    },
    {
        "role": "assistant",
        "content": "Here\u2019s a detailed breakdown of **Superpowers** \u2014 what it is, how it works, its core principles, and how it relates to Claude Code (and Codex):\n\n---\n\n## What Is Superpowers?\n\n- **Superpowers** is an open-source, agentic skills framework for AI coding assistants (Claude Code, Codex, Cursor, etc.) developed by Jesse Vincent (github handle *obra*) and the team at Prime Radiant. ([verdent.ai](https://www.verdent.ai/guides/what-is-superpowers-ai-coding-framework?utm_source=openai))  \n- It encodes software engineering best practices\u2014such as test-driven development (TDD), structured planning, code review, and systematic debugging\u2014into mandatory skills that trigger based on context. The goal is to ensure AI behaves like a senior engineer, not just a code generator. ([verdent.ai](https://www.verdent.ai/guides/what-is-superpowers-ai-coding-framework?utm_source=openai))\n\n---\n\n## How It Works\n\n### Skill-Based Architecture\n\n- Each \u201cskill\u201d is a `SKILL.md` file describing when it should activate, what instructions it enforces, and so on. Skills are composable and context-sensitive. ([verdent.ai](https://www.verdent.ai/guides/what-is-superpowers-ai-coding-framework?utm_source=openai))  \n- There\u2019s a **master skill** (`using-superpowers`) that acts like a dispatcher: It intercepts tasks early, determines which other skills should apply, and forces the agent to follow them rather than skipping directly into code. ([claudemarketplaces.com](https://claudemarketplaces.com/skills/obra/superpowers/using-superpowers?utm_source=openai))\n\n### Workflow Phases\n\nSuperpowers enforces a structured workflow made up of multiple phases. Depending on the write-up you read, there are slightly different breakdowns (5-phase, 7-phase) but they share the same fundamentals. Here\u2019s a typical sequence: ([ayautomate.com](https://www.ayautomate.com/blog/best-claude-code-skills?utm_source=openai))\n\n1. **Clarification / Brainstorming**  \n   The agent asks clarifying questions, surface requirements, dependencies, constraints before writing any code. Generates a specification or design document. ([verdent.ai](https://www.verdent.ai/guides/what-is-superpowers-ai-coding-framework?utm_source=openai))\n\n2. **Design**  \n   Architectural decisions, component layout, choices on structure, etc. Alternatives are weighed. ([popularaitools.ai](https://popularaitools.ai/blog/superpowers-plugin-10x-claude-code-2026?utm_source=openai))\n\n3. **Planning**  \n   Break down tasks into small, manageable units (2-5 minutes each), with explicit file paths, steps, verification criteria. Each task should be testable. ([mymcpshelf.com](https://www.mymcpshelf.com/blog/superpowers-skills-framework-claude-code?utm_source=openai))\n\n4. **Test-Driven Development (TDD)**  \n   Enforce \u201cred-green-refactor\u201d: write tests first (which must fail), then implement code to pass, then refactor. Any code written before tests is deleted. ([claude.com](https://claude.com/plugins/superpowers?utm_source=openai))\n\n5. **Subagent-Driven Development / Execution**  \n   For larger plans, Superpowers spawns \u201csubagents\u201d for each task. Each subagent implements a task and undergoes checks: spec compliance, test passing, code quality etc. Parallel execution is supported where possible. ([verdent.ai](https://www.verdent.ai/guides/what-is-superpowers-ai-coding-framework?utm_source=openai))\n\n6. **Review & Finalization**  \n   After implementation edges and tests pass, a code review happens\u2014automatically, often between tasks. Critical issues block further progress until resolved. Then the feature is merged/finalized. ([verdent.ai](https://www.verdent.ai/guides/what-is-superpowers-ai-coding-framework?utm_source=openai))\n\n7. **Debugging / Systematic Debugging**  \n   When errors occur, a four-phase methodology kicks in: root cause investigation, hypothesis testing, pattern analysis, and verification of fixes. ([claude.com](https://claude.com/plugins/superpowers?utm_source=openai))\n\n---\n\n## Why It\u2019s Significant\n\n- **Discipline over speed**: Superpowers aims to prevent early rush-to-code by requiring design, tests, review etc. ([verdent.ai](https://www.verdent.ai/guides/what-is-superpowers-ai-coding-framework?utm_source=openai))  \n- **Model-agnostic**: Although Claude Code is its primary platform, it also supports Codex and others. Skills are enforced via configuration rather than modifications to the underlying model. ([verdent.ai](https://www.verdent.ai/guides/what-is-superpowers-ai-coding-framework?utm_source=openai))  \n- **Open source & growing**: Released October 2025; popular\u2014tens of thousands of stars on GitHub; official marketplace listing for Claude Code since around January 2026. ([verdent.ai](https://www.verdent.ai/guides/what-is-superpowers-ai-coding-framework?utm_source=openai))  \n\n---\n\n## Limitations & Trade-offs\n\n- **Added overhead**: The brainstorming, planning, code review phases add time. For simple tasks or small bug fixes, it might feel slow or overkill. ([verdent.ai](https://www.verdent.ai/guides/what-is-superpowers-ai-coding-framework?utm_source=openai))  \n- **Relevance**: Not every task naturally fits the workflow. Exploratory prototyping or rapid experiments may want more flexibility. ([verdent.ai](https://www.verdent.ai/guides/what-is-superpowers-ai-coding-framework?utm_source=openai))  \n- **Platform-specific gaps**: Certain skills rely on features available in Claude Code (e.g. subagent spawning, worktrees). On platforms without those, some skills degrade gracefully or just don\u2019t get invoked. ([verdent.ai](https://www.verdent.ai/guides/what-is-superpowers-ai-coding-framework?utm_source=openai))  \n\n---\n\n## How to Use It in Claude Code (and Codex)\n\n- **Install Superpowers** via Claude Code plugin marketplace:  \n  ```\n  /plugin marketplace add obra/superpowers-marketplace  \n  /plugin install superpowers@superpowers-marketplace  \n  ```  \n  After installation, you\u2019ll see new slash commands like `/superpowers:brainstorming`, `/superpowers:writing-plans`, `/superpowers:execute-plan` etc. ([claudedirectory.org](https://www.claudedirectory.org/skills/superpowers?utm_source=openai))\n\n- **Use the master skill**: `using-superpowers` is always active and prevents skipping essential phases when tasks require them. ([claudemarketplaces.com](https://claudemarketplaces.com/skills/obra/superpowers/using-superpowers?utm_source=openai))\n\n- For **Codex** users, some adaptations are needed: you may have to manually integrate SKILL.md files or map certain tools like subagent dispatch depending on your environment. ([github.com](https://github.com/obra/superpowers/blob/main/skills/using-superpowers/references/codex-tools.md?utm_source=openai))\n\n---\n\nIf you'd like, I can walk you through a live example of a feature build using Superpowers, or help you compare how a typical Claude Code vs Superpowers workflow looks. Do you want to see that?"
    }
]