superskills

Superskills

Production-grade skills for Claude Code that make AI build real products, not prototypes.

These skills transform Claude Code from a code assistant into an autonomous product engineer. They were born from the realization that by 2027, most people using AI coding tools won't be developers — they'll be people with ideas who need AI to build production-ready software without hand-holding.

What's In Here

/genesis — Autonomous Project Creation

Say "go" and Claude autonomously:

• Researches what's trending, what's missing, what new AI capabilities enable
• Picks a project idea based on convergence of real-time signals
• Builds it from scratch with parallel agents
• Runs an adversarial review (an AI-hater challenges every decision)
• Scores it on a 10-dimension rubric
• Presents it to you with specific things to test

You become the tester. Claude becomes the builder with creative ownership.

/production-autopilot — Research-Driven Quality Assurance

A 2+ hour sustained effort that goes far beyond lint and tests:

• Research phase (30 min): arXiv papers, HuggingFace models, competitor analysis, dependency audits
• Engineering phase: Build, types, lint, tests, security, deep code review
• Experience phase: First Contact Test (can a non-technical person use it?), User Journey Testing, Product Polish
• Adversary phase: An AI skeptic challenges creativity, business viability, and soul
• Convergence loop: Keeps going until ALL phases report zero issues in the same cycle

Not done until a non-technical person could use it flawlessly.

/rubric — The Litmus Test

10-dimension scoring system. Below 7 in ANY dimension = NOT PRODUCTION READY.

Dimension	What It Measures
WORKS	Does it actually function? Every edge case?
OBVIOUS	Can someone use it in 30 seconds?
FAST	Does it feel instant?
SOLID	Would it survive a hostile security audit?
TESTED	Would you know if something broke?
ALIVE	Does it have personality and soul?
MONEY	Could this make money? Who pays?
ELEGANT	Would a great engineer respect the code?
READY	One-command deploy? Could you deploy and sleep?
ORIGINAL	Does this already exist? Why is THIS version worth existing?

Every score requires evidence. No vibes-based scoring.

cheatcodes.md — Claude Code Performance Guide

Every hidden lever, environment variable, and undocumented feature that makes Claude Code perform at its ceiling:

• Effort levels and thinking triggers (ultrathink, think very hard)
• 50+ environment variables for model control, context management, and features
• XML tag patterns that increase instruction compliance
• CLAUDE.md optimization techniques (+10% on SWE Bench)
• Keyboard shortcuts, permission modes, model selection

The Philosophy

The Relationship Flip

Traditional: Human asks AI for help → AI assists → Human reviews This: AI builds with creative ownership → Human tests and gives feedback → AI iterates

The Adversary

Every project gets reviewed by an internal AI skeptic who believes AI-generated code is mediocre by default. The adversary challenges creativity, business viability, and whether the project has "soul." Its criticisms must be answered with real code improvements, not justifications.

The Future User

The person running these skills in 2027 is not a developer. They have an idea. They described it to an AI. Now they need it to be production-ready. These skills ensure the AI treats that responsibility seriously — with research, testing, and a quality bar that doesn't flinch.

Installation

Claude Code

Copy the .md files to your skills directory:

cp genesis.md production-autopilot.md rubric.md ~/.claude/skills/

Codex CLI

Copy the files (without .md extension) to your Codex skills directory:

cp genesis ~/.codex/skills/
cp production-autopilot ~/.codex/skills/
cp rubric ~/.codex/skills/

Cheat Codes

Add to your ~/.claude/settings.json:

{
  "effortLevel": "max",
  "env": {
    "ENABLE_TOOL_SEARCH": "auto:0",
    "DISABLE_NON_ESSENTIAL_MODEL_CALLS": "1",
    "CLAUDE_CODE_MAX_OUTPUT_TOKENS": "64000",
    "CLAUDE_AUTOCOMPACT_PCT_OVERRIDE": "80"
  }
}

Prompt Enhancement Techniques Used

These skills incorporate research-backed techniques for maximizing LLM output quality: