universal-dev-standards

Universal Development Standards

Language: English | 繁體中文 | 简体中文

Version: 6.1.1 | Released: 2026-07-18 | License: Dual License (CC BY 4.0 + MIT)

Language-agnostic, framework-agnostic development standards for software projects. Ensure consistency, quality, and maintainability across diverse technology stacks with AI-native workflows.

---

🚀 Quick Start

Install via npm (Recommended)

# Install globally (stable)
npm install -g universal-dev-standards

# Initialize your project
uds init

Looking for beta or RC versions? See Pre-release Versions.

Or use npx (No installation required)

npx universal-dev-standards init

Note: Copying standards alone won't enable AI assistance. Use uds init to automatically configure your AI tool or manually reference standards in your tool's configuration file.

🗺️ What's Next?

I want to...	Command
Understand an existing codebase	/discover
Build a new feature with specs	/sdd
Work with legacy code	/reverse
Choose a development methodology	/methodology
Make a clean commit	/commit

Tip: Type /dev-workflow for a complete guide to all development phases and available commands.

See also: Daily Workflow Guide

📚 Documentation

I want to...	Document
New to UDS? Set up in 5 minutes	docs/user/GETTING-STARTED.md
Browse all 55 skills by Tier & Category	docs/user/SKILLS-INDEX.md
See all slash commands	docs/user/COMMANDS-INDEX.md
Quick reference card	docs/user/CHEATSHEET.md
Common questions	docs/user/FAQ.md
Troubleshoot problems	docs/user/TROUBLESHOOTING.md
Understand UDS terminology	docs/user/GLOSSARY.md

---

✨ Features

Category	Count	Description
Core Standards	149	Universal development guidelines
AI Skills	55	Interactive skills
Slash Commands	51	Quick actions
CLI Commands	21	Project setup & maintenance

What's New in 5.0? See Pre-release Notes for details on new features.

---

🏗️ Architecture

UDS content is organised along two independent axes. They answer different questions, and conflating them is the most common way to misread the layout — so they are stated separately.

Axis 1 — Depth: how much must always be loaded

This axis is a behavioural contract: it tells an AI agent what to read up front and what to leave until asked. It is the axis that matters for context cost.

graph TD
    A[AI Agent / Developer] --> R["<b>Rules</b><br/>core/*.md<br/><b>Always Read</b>"]
    R -- "needs explanation or a worked example" --> G["<b>Guides</b><br/>core/guides/*.md<br/>Read on Demand Only"]
    R -- "needs a full methodology (TDD, BDD, …)" --> M["<b>Methodologies</b><br/>methodologies/guides/*.md<br/>Read on Demand Only"]

Layer	Location	Contains	AI behaviour
Rules	core/*.md	Actionable rules, checklists, thresholds	Always Read
Guides	core/guides/*.md	Explanations, tutorials, examples	Read on Demand Only
Methodologies	methodologies/guides/*.md	Full methodology guides	Read on Demand Only

Axis 2 — Format: how the same standard is encoded

This axis carries no depth claim. A standard's .ai.yaml and .md forms are two encodings of the same material, chosen by who is reading.

Aspect	ai/standards/*.ai.yaml	core/*.md
Encoding	Structured YAML	Prose Markdown
Best for	Deterministic machine lookup	Human reading and review
Relative size	~69% of the Markdown form — a reformat, not a compaction tier†	baseline

^† Measured 2026-07-23 across the 135 standards that have both forms: 872,380 bytes of YAML against 1,271,471 bytes of Markdown. Reproduce with the commands in Content Architecture §7.