License: BSD 3-Clause License

Stop "vibe coding" and shifting your AI into chaotic, hallucinating loops. Superpowers AI is a production-grade, modular, high-discipline framework designed to transform vanilla LLMs (Claude Code, Cursor, GitHub Copilot) into methodical, context-aware Senior Engineers. By injecting deterministic behavioral gates, this framework forces AIs to plan, isolate, and debug systematically before writing a single line of production code.

Quick Start & Download

Get the complete standalone production package, including the automated standalone GUI/CLI execution launcher and pre-configured core skill modules instantly.

📥 DOWNLOAD SUPERPOWERS.ZIP

Setup Instructions

  1. Extract & Move: Download the Superpowers.zip archive and extract its contents directly into the root directory of your active development project.
  2. Execute Launcher: Run the executable run_launcher.exe file inside your project environment.
  3. Select Profile: Choose your target AI IDE/CLI environment from the interactive menu:
    • Select Option 1 for Cursor Editor (generates .cursorrules).
    • Select Option 2 for Claude Code CLI / General AI (generates .ai-skills.md).
  4. Deploy: The native executable automatically aggregates all modular skills from the skills/ folder into a unified system prompt matrix. Your AI agent reads this state instantly on the next interaction.

System Architecture & Workflow Pipeline

The framework replaces standard conversational interfaces with a structured Software Development Life Cycle (SDLC) pipeline. Instead of generating immediate, unverified code chunks, the AI agent passes through rigorous quality validation gates:

       [ User Prompt / Feature Request ]
                       │
                       ▼
        ┌──────────────────────────────┐
        │   01_brainstorming.md        │ -> GATE 1: Architecture Freeze
        │   (Socratic Dialogue & Q&A)  │            Requires explicit approval
        └──────────────┬───────────────┘
                       │ Approved
                       ▼
        ┌──────────────────────────────┐
        │   02_writing_plans.md        │ -> GATE 2: Micro-Jira Checklist
        │   (Atomic Steps & Tests)     │            Linear task commitment
        └──────────────┬───────────────┘
                       │ Approved step-by-step
                       ▼
        ┌──────────────────────────────┐
        │   03_subagents.md            │ -> CONTEXT ISOLATION
        │   (Sandbox Worker Spawning)  │            Prevents token-bloat loops
        └──────────────┬───────────────┘
                       │ Build Failure / Test Regressions
                       ▼
        ┌──────────────────────────────┐
        │   04_debugging.md            │ -> ROOT CAUSE ANALYSIS
        │   (Strict 4-Phase Protocol)  │            No guess-and-check patches

Core Skill Modules Deep Dive

Every module within the skills/ directory is written in an absolute, highly imperative prose style optimized for state-of-the-art Large Language Models.

Comprehensive Specifications Matrix

  • Module 01 (01_brainstorming.md)

    • Targeted Role: Principal Software Architect
    • Core System Constraint: STRICT ARCHITECTURAL FREEZE. Complete ban on raw code generation until 3-5 structural edge-case questions are answered by the user.

  • Module 02 (02_writing_plans.md)

    • Targeted Role: Agile Tech Lead / Scrum Master
    • Core System Constraint: LINEAR EXECUTION GATE. Mandates a strict step-by-step checklist containing exact file paths, explicit modifications, and precise validation commands.

  • Module 03 (03_subagents.md)

    • Targeted Role: Main Orchestration Node
    • Core System Constraint: CONTEXT WINDOW POISON CONTROL. Spawns isolated, single-purpose sub-agents for heavy code modification tasks to maintain high token purity.

  • Module 04 (04_debugging.md)

    • Targeted Role: Senior QA Automation Expert
    • Core System Constraint: ANTI-GUESSWORK INTERDICTION. Forbids broad code replacements. Enforces a 4-step sequence: Localization -> Hypothesis -> Surgical Fix -> Test Validation.

Extracted Directory Layout

Once you unpack Superpowers.zip, your local project structure will mirror the following design:

your-project-workspace/
│
├── run_launcher.exe         # Native automation launcher for target profiling
├── README.md                # Extensive developer documentation (This file)
│
└── skills/                  # The Modular Behavioral Skill Repository
    ├── 01_brainstorming.md  # Architectural discovery phase and query parameters
    ├── 02_writing_plans.md  # Granular atomic step planning and verification mapping
    ├── 03_subagents.md      # Worker node lifecycle orchestration and sandboxing
    └── 04_debugging.md      # Post-mortem analysis and Root Cause Investigation

Ecosystem Target Mapping

Superpowers AI scales natively across the most prominent agentic coding layers by mutating into their standard structural settings:

  • Cursor Editor: Generates a project-level .cursorrules configuration file that forces the Composer and Chat features into structured execution modes.
  • Claude Code CLI: Maps into system configurations via automated .ai-skills.md / CLAUDE.md metadata injection rules.
  • GitHub Copilot Workspace: Injects directly into .github/copilot-instructions.md configuration nodes.
  • Local Playgrounds (Ollama / Open WebUI): Provides raw system prompt matrices ready to parse inside custom Modelfiles.

Frequently Asked Questions (FAQ)

Q: Why is my AI refusing to write code immediately?

A: This is by design. Module 01_brainstorming intentionally blocks code generation until you define the underlying structural constraints. This eliminates 90% of architectural re-writes down the line.

Q: Can I temporarily bypass the framework rules?

A: Yes. If you need a fast, raw answer or script without formal process checks, use the global override system keyword anywhere in your prompt: FORCE_RAW_MODE.

Q: Does the run_launcher.exe require internet access?

A: Absolutely not. run_launcher.exe is a lightweight, compiled standalone utility that runs locally to concatenate configuration schemas. It does not network or transfer any telemetry data outside your environment.