AI Engineering Standard

AI Development, Training & Agent Engineering Standards

v2.1.0 — Contract-First Multi-Agent Engineering

Language: 한국어

What is AI Engineering Standard?

AI Engineering Standard is a reusable engineering standard for AI-assisted development, model training, experimentation, LLM/Vision workflows, general ML/DL workflows, and AI coding agents.

Version 2 establishes a machine-readable architecture and policy contract, explicit agent/Skill routing, environment-aware runtime behavior, cross-platform installation lifecycle controls, Korean localization quality gates, executable conformance checks, and dependency-compatibility alignment rules. Version 2.1 extends this foundation with framework-neutral multi-agent contracts for Agent, Agent Role, Agent Contract, Work Unit, Handoff, Evidence, Evaluation, and Acceptance.

Independent repository model

This repository is the single source of truth for development, validation, and release.

flowchart TD
    A[Feature / Fix / Docs] --> B[Pull Request]
    B --> C[Local Validation Gate]
    B --> D[GitHub Actions CI]
    C --> E[main]
    D --> E
    E --> F[Version Tag]
    F --> G[GitHub Release]

Local validation and GitHub Actions CI verify the same repository state before changes reach main. GitHub Actions is an automated verification and bounded execution mechanism; it is not a second source of truth or a replacement repository.

There is no separate private repository, development repository, staging repository, or promotion/export step.

2.1 architecture

Version 2.1 extends the 2.0 foundation without replacing it. The canonical unit is the Work Unit, with explicit Agent/Role contracts, Handoffs, Evidence, Evaluation, and Acceptance. Contract definitions live under core/contracts/2.1/; architecture guidance lives under docs/architecture/2.1/.

Version 2.0.0, 2.0.1, and 2.0.2 remain preserved as historical release provenance. Starting with 2.0.1, development and release are performed directly in this repository under the independent release contract.

Quick start

AIEngineeringStandard 2.2 is installed as a versioned package. The consumer project does not need to clone the AIEngineeringStandard repository.

Recommended: pipx

pipx install ai-engineering-standard

Then install the standard into the current project:

ai-engineering-standard install --language ko --domain all

Preview without writing:

ai-engineering-standard install --language ko --domain all --dry-run

Alternative: pip

python -m pip install ai-engineering-standard
ai-engineering-standard install --language ko --domain all

Pinned installation:

pipx install ai-engineering-standard==2.2.0

The package provides one cross-platform CLI for installation, state inspection, update/reconciliation, validation, and safe uninstall. Shell and PowerShell installer scripts remain development/source-tree compatibility entrypoints rather than the primary consumer interface.

Available domains are common, ml, llm, vision, colab, and all. The canonical source language is English, with Korean as the only supported localized locale. Supported locales are defined by i18n/languages.json; the installer derives its accepted locale set from that catalog.

Installation lifecycle

Successful installs record ownership and hashes in .codingstandard/installation.json. The installer supports state inspection, update/reconciliation, safe uninstall, and explicit force mode for recovery.

See INSTALL.md for the complete installation contract.