Superpowers
Jesse Vincent's open-source Agentic skills framework and software development methodology plugin
标签:AI Agent Plugins And SkillsPlugins and SkillsWhat are Superpowers?
Superpowers is an open-source agentic skills framework and software development methodology developed by Jesse Vincent and the Prime Radiant team, specifically designed for AI coding agents such as Claude, Cursor, and GitHub Copilot. Through 14 composable skills and initial instructions, it prevents AI coding assistants from directly jumping into code writing. Instead, it first refines requirements through Socratic questioning, generates verifiable design documents, and formulates detailed implementation plans before entering a sub-agent-driven development process. Its core emphasis is on RED-GREEN-REFACTOR test-driven development, the YAGNI principle, and the DRY principle. It supports advanced workflows such as Git Worktrees isolated development, parallel sub-agent scheduling, and two-phase code review, allowing agents like Claude to run autonomously for hours without deviating from their planned trajectory.
Superpowers’ main functions
-
Socratic brainstorming : This involves refining requirements through questioning, presenting design solutions in chunks for user confirmation before proceeding to coding.
-
Design document generation : Transforms rough ideas into structured, verifiable specifications and saves them automatically.
-
Detailed implementation plans : Break down the work into 2-5 minute micro-tasks, including precise file paths, complete code, and verification steps.
-
Subagent-driven development : Dispatch an independent subagent to each task and perform a two-phase review (specification compliance + code quality).
-
Test-driven development (TDD) enforces a red-green-refactor loop, requiring tests to be written before code, and removing any code written before testing.
-
Systematic debugging : A four-stage root cause analysis process that includes techniques such as root cause tracing, defense in depth, and conditional waiting.
-
Git Worktrees workflow (using git-worktrees) : Develop in separate branches and working directories, and validate clean test baselines.
-
Parallel agent scheduling : Executes multiple sub-agent workflows concurrently, improving efficiency in large projects.
-
Code review mechanism (requesting-code-review / receiving-code-review) : Automatic review between tasks, reporting issues according to severity, and blocking progress for critical issues.
-
Finishing a development branch : Validating tests, providing merge/PR/keep/discard options, and cleaning up the workspace.
-
Self-expanding skills (writing-skills) : Creating and testing new skills following best practices, supporting teams in developing internal methodologies.
-
Multi-platform plugin support : Covers Claude Code Official Marketplace, Cursor, OpenAI Codex CLI/App, GitHub Copilot CLI, Gemini CLI, and OpenCode.
How to use Superpowers
-
Claude Code Official Marketplace : Running
/plugin install superpowers@claude-plugins-official. -
Claude Code Superpowers Market : Execute first
/plugin marketplace add obra/superpowers-marketplace, then execute/plugin install superpowers@superpowers-marketplace. -
OpenAI Codex CLI : Open the plugin search interface, search for “Superpowers” and select to install.
-
OpenAI Codex App : Click Plugins in the sidebar, find Superpowers in the Coding category, and click
+Install. -
Cursor : Search for “superpowers” in the add-on marketplace and install it via Agent chat.
-
OpenCode : Input
Fetch and follow instructions from https://raw.githubusercontent.com/obra/superpowers/refs/heads/main/.opencode/INSTALL.md. -
GitHub Copilot CLI : Run
copilot plugin marketplace add obra/superpowers-marketplaceand in sequencecopilot plugin install superpowers@superpowers-marketplace. -
Gemini CLI :
gemini extensions install https://github.com/obra/superpowersUsed for running and updatinggemini extensions update superpowers. -
Basic workflow : Brainstorming → Design confirmation → Git Worktree creation → Plan writing → Sub-agent execution/Plan execution → TDD implementation → Code review → Branch cleanup.
Key information and usage requirements for Superpowers
-
Open source license : MIT License, which allows free use, modification, and commercial application.
-
Maintainers : Jesse Vincent and the Prime Radiant team.
-
Core Philosophy : TDD Priority, Systematic Processes, Complexity Reduction, Evidence Verification.
-
Number of skills : 14 composable skills covering testing, debugging, collaboration, planned execution, parallel agents, and Git workflows.
-
Installation prerequisites : A Coding Agent that supports plugin systems (Claude Code, Cursor, Codex, etc.) must be used.
-
Workflow Mandatory : The agent must check relevant skills before performing a task; this is not a suggestion but a mandatory process.
-
Community support : Discord community, GitHub Issues, email subscription release notifications.
-
Contribution Guidelines : Adding new skills arbitrarily is not allowed. Modifications must be verified across all supported platforms and follow
writing-skillsthe guidelines. -
Update mechanism : Some platforms update automatically, depending on the Coding Agent’s plugin system.
Superpowers’ core advantages
-
Instead of writing code directly, understand the requirements first : When the agent sees a development task, it takes a step back and clarifies the real goal by asking questions, avoiding illusions and rework.
-
Systematized rather than arbitrary : Process-based methodologies replace guesswork programming, with complexity reduction as the primary goal.
-
Evidence speaks louder than words : Success is declared only after verification, avoiding false completions that “appear to work”.
-
Automatic skill trigger : Skills are automatically activated when a relevant task is detected, without the need for manual activation or memorizing complex commands.
-
Long-term autonomous operation : Agents such as Claude can work autonomously for hours without deviating from their target according to a predetermined plan.
-
Prioritize mandatory testing : TDD, as a mandatory workflow rather than an optional suggestion, fundamentally ensures code quality.
-
Multi-Agent Ecosystem Compatibility : The same skill set works seamlessly across platforms such as Claude, Cursor, Copilot, Codex, and Gemini.
-
Micro-task breakdown : Task descriptions with a granularity of 2-5 minutes allow even “enthusiastic junior engineers” to execute them accurately.
-
Isolated development : Git Worktrees avoids polluting the main branch, supports parallel experimentation, and allows for safe rollback.
Application scenarios of Superpowers
-
Building new features from scratch : Ensure requirements are understood correctly through brainstorming and design documents to avoid large-scale rework later.
-
Large-scale refactoring project : using Git Worktrees to isolate experiments, with sub-agents handling the modifications of different modules in parallel.
-
Bug Fixing and Root Cause Analysis : Systematic debugging skills guide agents to perform four-stage root cause tracing, rather than blindly trying and failing.
-
Automated code review : Before submission, the Agent performs a two-stage review to proactively identify specification deviations and quality issues.
-
Complex changes involving multiple files : Break down large requirements into micro-tasks, assign sub-agents for gradual implementation and continuous verification.
-
Cross-platform development collaboration : The same set of Superpowers skills provides a consistent experience across Claude, Cursor, and Copilot.
Frequently Asked Questions about Superpowers
A: It combines both. It’s a complete software development methodology based on composable skills, and it’s also distributed as an open-source plugin to major AI coding platforms.
A: No. Skills are automatically triggered based on the current task. Users only need to interact with the Coding Agent normally, and Superpowers will take over the workflow in the background.
A: Claude Code (official marketplace), Cursor, OpenAI Codex CLI/App, GitHub Copilot CLI, Gemini CLI, OpenCode.
A: Yes, but you must follow writing-skillsthe skill guidelines and ensure compatibility across all supported platforms. The official team generally does not directly accept new skill contributions.
A: Generating code directly can easily lead to misperceptions and rework. Superpowers ensures that the Agent truly understands the requirements before taking action by designing documents and plans, reducing the cost of later modifications.
A: No. RED-GREEN-REFACTOR is a mandatory workflow; the agent removes any code written before testing to ensure test coverage.