Open AI Engineering Standard (OAIES)
The definitive open-source standard for building production-grade AI systems. Not a tutorial. Not a template. A standard — opinionated, battle-tested, and built by engineers who ship.
Why This Exists
Every serious engineering discipline has a standard.
- Web: React, Next.js conventions
- APIs: REST, OpenAPI, GraphQL
- DevOps: The Twelve-Factor App
- AI Engineering: Nothing. Until now.
Most AI repositories give you "here are five ways to do X." This repository gives you one way — the right way — informed by production deployments, enterprise constraints, and the collective wisdom of engineers who have shipped AI systems at scale.
If you are starting any AI product, you fork this first.
Maturity Model
This repository is structured as a 10-level AI Engineering Maturity Model. Enter at your current level. Ascend deliberately.
| Level | Name | When You're Ready |
|---|---|---|
| L0 | AI Foundations | You're new to building with LLMs |
| L1 | Prompt Engineering | You understand tokens and context |
| L2 | Context Engineering | You're building multi-turn systems |
| L3 | Skill Engineering | You're automating recurring workflows |
| L4 | Agent Engineering | You're building autonomous agents |
| L5 | Multi-Agent Systems | You're orchestrating agent networks |
| L6 | Memory & Knowledge | You need persistent agent memory |
| L7 | LLMOps | You're running AI in production |
| L8 | AI SDLC | You're building AI-first processes |
| L9 | Enterprise AI | You need governance and compliance |
| L10 | AI Org Playbook | You're building an AI-first organization |
The Standard Workflow
Every feature. Every time. No exceptions.
Idea → Research → Requirements → Story Kickoff → Context Collection →
Knowledge Gathering → Planning → Architecture → Human Approval →
Prompt Generation → Coding → Self Review → Agent Review → Testing →
Evaluation → Security → Performance → Accessibility → Documentation →
Deployment → Monitoring → Continuous Improvement → Postmortem →
Lessons Learned → Knowledge Base Update
See the AI SDLC for the complete workflow specification with prompts for every stage.
What's Inside
🏗️ Core Sections (Maturity Levels)
- 00-foundations/ — How LLMs think, the harness principle, failure modes
- 01-prompt-engineering/ — XML-structured prompts, patterns, anti-patterns
- 02-context-engineering/ — Context fidelity, progressive disclosure, state management
- 03-skill-engineering/ — 26 reusable agent skills
- 04-agent-engineering/ — 20 agent definitions, subagents, hooks
- 05-multi-agent-systems/ — Orchestration patterns, A2A communication
- 06-memory-knowledge/ — Memory types, knowledge graphs, hybrid RAG
- 07-llmops/ — Evaluation, PromptOps, cost optimization, observability
- 08-ai-sdlc/ — Complete SDLC with 15 production prompts
- 09-enterprise-ai/ — Governance, security, compliance, templates
- 10-ai-org-playbook/ — Roles, operating models, culture
Unique differentiators
- patterns/ — 12 AI Engineering Patterns (13-component operational specs; start with planner)
- cookbook/ — Technology cookbooks with prompts, skills, and checklists (see React)
- mcps/ — MCP integration specs for production tooling
Operational resources (canonical paths)
- SDLC prompts — Production prompts including implementation-plan
- Skills — Reusable agent skills
- Agents — Agent definitions
- Hooks — Lifecycle hooks
- Checklists — Quality gates for every stage
- Evaluation — Framework selection and eval practice
- Enterprise security & governance — Security, compliance, templates
- Observability — Tracing, PromptOps, cost, dashboards
- Templates — ADR and related templates
- Resources — Curated books, talks, repositories, standards
- Normative standard — Machine-checkable controls, evidence, conformance (when claiming OAIES conformance)
Quick Start
Use with Claude Code
# Clone into your project
git clone https://github.com/gauravprwl14/open-ai-engineering-standard .oaies
# Copy the behavioral contract (repo root)
cp .oaies/CLAUDE.md ./CLAUDE.md
# Install skills from their canonical location
mkdir -p .claude/skills
cp .oaies/content/03-skill-engineering/skills/*.skill.md .claude/skills/
# Install agents from their canonical location
mkdir -p .claude/agents
cp .oaies/content/04-agent-engineering/agents/*.agent.md .claude/agents/
# Start a feature using the standard workflow
# Reference: content/08-ai-sdlc/prompts/story-kickoff.prompt.md
# Planning: content/08-ai-sdlc/prompts/implementation-plan.prompt.md
# Pattern: content/patterns/planner-pattern/README.md
Use with Cursor
# Point Cursor rules / docs at the content tree, e.g.:
# content/02-context-engineering/templates/CLAUDE.md.template
# content/08-ai-sdlc/prompts/
# content/patterns/planner-pattern/
Use standalone
Browse the maturity model from Level 0 and follow the standard workflow in Level 8. For complex work, run the Planner Pattern before coding.
Design Principles
| Principle | What It Means |
|---|---|
| Opinionated | One standard, not a menu of options |
| Production-first | Recommendations include workflows, prompts, and failure recovery you can run |
| Enterprise-ready | Governance, compliance, and audit trails are specified where they matter |
| Living standard | Updated as tooling and practice change |
| Context-complete | Every decision includes Why, When, Tradeoffs, Anti-patterns |
The 50/50 Rule
50% AI reasoning. 50% deterministic code.
The biggest mistake in AI engineering is treating the LLM as the entire system. The model is one component. The deterministic harness around it — validation, authorization, execution, logging — is equally important. This repository enforces both halves equally.
Contributing
Read CONTRIBUTING.md before opening a PR.
The standard for contributions is high by design. Every addition must include:
- Clear "Why" rationale
- Documented tradeoffs
- At least one anti-pattern
- Enterprise consideration
- Working example
Versioning
| Version | Focus |
|---|---|
| v1.0 | AI Foundations, Prompt Engineering, Context Engineering |
| v2.0 | Agent Engineering, Multi-Agent Systems |
| v3.0 | LLMOps, AI SDLC |
| v4.0 | Enterprise AI, AI Org Playbook |
| v5.0 | Patterns Library complete |
| v10.0 | Full Cookbook across all technology stacks |
License
MIT — Use it, adapt it, build on it. If it helps you, give it a star and consider contributing back.
OAIES v1.0 — "The standard doesn't tell you what you can do. It tells you what you must do."