Superpowers is an agentic skills framework and software development methodology designed for coding AI. It enforces rigorous engineering workflows—such as Test-Driven Development (TDD), design-first planning, and subagent collaboration—transforming AI from a simple code generator into a systematic virtual engineer capable of autonomous planning, execution, and review.
One Minute Overview#
Superpowers is a plugin system that teaches coding agents—like Claude, Codex, or OpenCode—proper software engineering discipline.
It acts as a methodological guardrail. By forcing agents to undergo "brainstorming" and "planning" before coding, and enforcing a strict "Red-Green-Refactor" TDD loop during development, it solves common AI pitfalls like hallucinated code, lack of context, and unmaintainable spaghetti code.
Core Value: Transforms AI coding from "ad-hoc generation" into "engineered delivery" by constraining and guiding agent behavior.
Quick Start#
Installation Difficulty: Low - Requires specific AI coding environments.
Superpowers integrates directly into your AI client. The easiest method is via the Claude Code Plugin Marketplace:
# 1. Register the marketplace
/plugin marketplace add obra/superpowers-marketplace
# 2. Install the plugin
/plugin install superpowers@superpowers-marketplace
# 3. Verify installation
/help
# You should see commands like /superpowers:brainstorm, /superpowers:write-plan
Is this suitable for me?
- ✅ Complex AI Development: When using AI for projects involving multiple files or complex logic.
- ✅ Quality Focused: Ideal if you are tired of AI-generated code lacking tests or failing CI/CD.
- ✅ Workflow Standardization: Necessary when you need AI to follow specific Git workflows or code review standards.
- ❌ Simple Scripts: Likely overkill if you just need a quick 10-line Python script.
Core Capabilities#
1. Brainstorming & Design - Design First, Code Later#
Before writing a single line of code, Superpowers activates a brainstorming protocol. It uses Socratic questioning to clarify requirements, explore alternatives, and presents a structured design document for approval. Actual Value: Prevents AI from making false assumptions based on vague prompts, ensuring the project starts on the right foot.
2. Test-Driven Development - Enforced TDD#
This is the backbone of the system. The agent is forced to follow the cycle: Write failing test -> Watch it fail -> Write minimal code -> Watch it pass. If the AI attempts to write implementation code before the test, it self-corrects. Actual Value: Guarantees code robustness and creates a safety net for future iterations.
3. Subagent-Driven Development - Autonomous Execution#
Development plans are broken down into bite-sized tasks (2-5 mins) and dispatched to "subagents." Each task undergoes a two-stage review: compliance with specifications and code quality checks. Actual Value: Enables high autonomy. The AI can work like a senior engineer for hours without losing context or deviating from the plan.
4. Systematic Debugging#
Replaces ad-hoc "trial and error" with a 4-phase root-cause process, including root-cause tracing, defense-in-depth analysis, and condition-based waiting. Actual Value: When bugs appear, the AI investigates logically rather than guessing, leading to faster fixes.
Tech Stack & Integration#
Languages: Shell, JavaScript, Python, TypeScript Key Dependencies: Requires host environments like Claude Code, Codex, or OpenCode. Integration Methods:
- Claude Code: Native plugin marketplace integration.
- Codex/OpenCode: Instruction injection via remote Markdown fetch.
Maintenance Status#
- Development Activity: Actively maintained with frequent commits and releases.
- Community Response: High engagement with active issues and PRs; comprehensive documentation available.
- Extensibility: Supports custom user-written skills; the project includes a "writing-skills" meta-skill to guide users.