Agentic Codex: GH-600 Reference Series
The Agentic Codex reference series — six GH-600 domain deep-dives supporting the Agentic AI certification track, mapped to quests and skills measured.
Table of Contents
This is the Agentic Codex reference series — a set of six domain deep-dives that support the GH-600 Agentic AI certification track and its companion quest arc. Each article maps directly to one domain of the GH-600 Developing in Agentic AI Systems exam, explaining the concepts the certification tests and pointing to the hands-on quests where you implement them.
Use this series as the conceptual companion to the quests: read the domain article to understand why, then complete the linked quests to practice how.
What Is the GH-600?
The GH-600 certification — Developing in Agentic AI Systems — tests your ability to design, build, evaluate, and govern AI agents that work on your behalf inside the GitHub platform.
The exam tests six knowledge domains:
| Domain | Weight | Topic |
|---|---|---|
| D1 | 18% | Agentic AI in the SDLC |
| D2 | 18% | Tools, environments, and agent capabilities |
| D3 | 19% | Memory, context, and state management |
| D4 | 19% | Evaluating and improving agent performance |
| D5 | 17% | Multi-agent system design |
| D6 | 9% | Responsible autonomy and guardrails |
The arc covers all of these with hands-on exercises that use only GitHub-native tools: Copilot coding agent, GitHub Actions, MCP servers, GitHub Environments, and the GitHub Models API.
The Reference Series
Each article below is a deep-dive on one GH-600 domain:
- Embedding Agents in the SDLC (GH-600 Domain 1) — bounded agency, plan-then-execute, and observability across the lifecycle.
- MCP Servers and Agent Tooling in Practice (Domain 2) — the Model Context Protocol, tool scoping, and dev-environment integration.
- Taming Agent Memory and Context Drift (Domain 3) — the three memory tiers, state persistence, and drift detection.
- Evaluating and Tuning Agents with GitHub Signals (Domain 4) — machine-verifiable success criteria, root cause analysis, and behaviour tuning.
- Orchestrating Multi-Agent Workflows on GitHub (Domain 5) — fan-out, correlation IDs, failure recovery, and lifecycle management.
- Agent Guardrails and Responsible Autonomy (Domain 6) — autonomy levels, guardrails, human-in-the-loop, and audit trails.
The Arc Structure
Domain 1 (Q1–Q3) → Domain 2 (Q4–Q7) → Domain 3 (Q8–Q10)
→ Domain 4 (Q11–Q13) → Domain 5 (Q14–Q17) → Domain 6 (Q18–Q19)
→ Grand Capstone (all domains)
Each quest builds on the previous one. Q1 starts with the basics of where agents fit in an SDLC. The Capstone requires you to deploy a working 3-agent orchestration system with full observability, evaluation, and governance in place.
How to Begin
- Read the GH-600 certification hub
- Review the skills measured
- Start at Q1: Agentic SDLC Integration
The arc was designed to take approximately 40 hours of hands-on work from start to capstone. The exam itself is 70% pass threshold across all domains.
Good luck, adventurer. The Codex is open.