GPTCode Skills

Skills are the secret sauce behind gt’s ability to generate production-quality code. When you run any gt command, it automatically detects your project’s language and injects the relevant skill into the AI’s system prompt.

💡 The result: Instead of generic code that “works”, you get idiomatic code that follows community best practices, proper error handling, and language-specific patterns.

Why Skills Matter

Without skills, AI models produce generic code:

• No language idioms
• Inconsistent error handling
• Poor naming conventions
• Missing documentation patterns

With skills, gt produces production-ready code:

• Idiomatic patterns (e.g., Go’s explicit error handling, Elixir’s pattern matching)
• Consistent style following community guidelines
• Proper documentation and testing patterns
• Framework-specific best practices (Rails, Phoenix, React)

How Skills Work

┌──────────────────────────────────────────────────────────┐
│  You run: gt do "add user authentication"                │
│                        ↓                                 │
│  gt detects: Ruby on Rails project (Gemfile, config/)   │
│                        ↓                                 │
│  gt injects: Rails skill + Ruby skill into prompt       │
│                        ↓                                 │
│  AI generates: Service objects, proper migrations,      │
│                RSpec tests, Devise patterns             │
└──────────────────────────────────────────────────────────┘

Available Skills

Language-Specific

General

Design & Product

Ops

Installing Skills

# List available skills
gt skills list

# Install a specific skill
gt skills install ruby

# Install all built-in skills
gt skills install-all

# View skill content
gt skills show ruby

Creating Custom Skills

You can create custom skills for your team or Stack:

1. Create a markdown file in ~/.gptcode/skills/
2. Add frontmatter with name, language, and description
3. The skill will be automatically loaded when working with that language

Example Custom Skill

---
name: my-company-style
language: typescript
description: Our company's TypeScript conventions
---

# Company TypeScript Style

## Always use strict mode
...

Contributing Skills

Want to add a skill for your favorite language? Open a PR with your skill markdown file.

Using Skills in Other Tools

The skills that power gt can also be used in other AI coding tools. Here’s how to apply them in your favorite environment:

Cursor

Cursor has the strongest support for project rules via .cursorrules:

# Export skills directly to Cursor format
gt skills show go > .cursorrules

# Or combine multiple skills
gt skills show go >> .cursorrules
gt skills show tdd-bug-fix >> .cursorrules

Cursor reads .cursorrules automatically and applies them to every interaction—no need to ask.

VS Code (GitHub Copilot)

GitHub Copilot supports custom instructions via files in your project:

Project-wide instructions:

# Export a gt skill to Copilot format
gt skills show ruby > .github/copilot-instructions.md

The file .github/copilot-instructions.md is automatically read by Copilot before every interaction.

Reusable prompts:

# Create a prompt for TDD workflow
gt skills show tdd-bug-fix > .github/prompts/tdd.prompt.md

Use in Copilot Chat by referencing the prompt.

Antigravity (Gemini in IDE)

If you’re using Google’s Antigravity (Gemini IDE integration):

1. Create .gemini/settings.json in your project:

   {
     "customInstructions": "See .gemini/skills/ for coding guidelines"
   }
   

2. Export skills to .gemini/skills/:

   mkdir -p .gemini/skills
   gt skills show python > .gemini/skills/python.md
   gt skills show tdd-bug-fix > .gemini/skills/tdd.md
   

Google Gemini (AI Studio / API)

For Gemini, use skills as system instructions:

# Show skill content to copy
gt skills show typescript

Paste the content into:

• AI Studio: System Instructions field
• API: system_instruction parameter in your request

# Example with Gemini API
import google.generativeai as genai

model = genai.GenerativeModel(
    'gemini-pro',
    system_instruction=open('typescript-skill.md').read()
)

Claude (Anthropic Console / API)

Claude supports system prompts where you can inject skills:

# Export for Claude
gt skills show elixir > claude-system.md

Use in:

• claude.ai: Start conversation with “Follow these guidelines:” + paste skill
• API: Set as system parameter

# Example with Claude API
import anthropic

client = anthropic.Anthropic()
message = client.messages.create(
    model="claude-3-opus-20240229",
    system=open('elixir-skill.md').read(),
    messages=[{"role": "user", "content": "Create a GenServer"}]
)

Replit (Agent / Ghostwriter)

Replit doesn’t auto-load rule files yet, but you can use a convention:

# Export to a RULES file
gt skills show python > RULES.md

Tips for Replit:

1. Keep RULES.md open in an editor tab—Ghostwriter prioritizes open files
2. When using Replit Agent, start with: “Read RULES.md and follow strictly”

Quick Reference

💡 Pro tip: Commit your skills files to git so your whole team benefits from consistent AI-generated code!